diff --git a/.gitignore b/.gitignore
index e1e018ea..8282d4b0 100644
--- a/.gitignore
+++ b/.gitignore
@@ -12,4 +12,5 @@ cmd.md
 .env
 /one-api
 temp
-.DS_Store
\ No newline at end of file
+.DS_Store
+/__debug_bin*
diff --git a/Dockerfile b/Dockerfile
index 346d9c5b..66edb360 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -44,4 +44,4 @@ COPY --from=builder2 /build/one-api /
 
 EXPOSE 3000
 WORKDIR /data
-ENTRYPOINT ["/one-api"]
\ No newline at end of file
+ENTRYPOINT ["/one-api"]
diff --git a/README.md b/README.md
index 5decf662..33ffb385 100644
--- a/README.md
+++ b/README.md
@@ -385,7 +385,7 @@ graph LR
    + 例子：`NODE_TYPE=slave`
 9. `CHANNEL_UPDATE_FREQUENCY`：设置之后将定期更新渠道余额，单位为分钟，未设置则不进行更新。
    + 例子：`CHANNEL_UPDATE_FREQUENCY=1440`
-10. `CHANNEL_TEST_FREQUENCY`：设置之后将定期检查渠道，单位为分钟，未设置则不进行检查。 
+10. `CHANNEL_TEST_FREQUENCY`：设置之后将定期检查渠道，单位为分钟，未设置则不进行检查。
    +例子：`CHANNEL_TEST_FREQUENCY=1440`
 11. `POLLING_INTERVAL`：批量更新渠道余额以及测试可用性时的请求间隔，单位为秒，默认无间隔。
     + 例子：`POLLING_INTERVAL=5`
diff --git a/common/config/config.go b/common/config/config.go
index a235a8df..c881e097 100644
--- a/common/config/config.go
+++ b/common/config/config.go
@@ -109,6 +109,9 @@ var RequestInterval = time.Duration(requestInterval) * time.Second
 
 var SyncFrequency = env.Int("SYNC_FREQUENCY", 10*60) // unit is second
 
+// ForceEmailTLSVerify is used to determine whether to force TLS verification for email
+var ForceEmailTLSVerify = env.Bool("FORCE_EMAIL_TLS_VERIFY", false)
+
 var BatchUpdateEnabled = false
 var BatchUpdateInterval = env.Int("BATCH_UPDATE_INTERVAL", 5)
 
@@ -164,3 +167,6 @@ var UserContentRequestTimeout = env.Int("USER_CONTENT_REQUEST_TIMEOUT", 30)
 
 var EnforceIncludeUsage = env.Bool("ENFORCE_INCLUDE_USAGE", false)
 var TestPrompt = env.String("TEST_PROMPT", "Output only your specific model name with no additional text.")
+
+// OpenrouterProviderSort is used to determine the order of the providers in the openrouter
+var OpenrouterProviderSort = env.String("OPENROUTER_PROVIDER_SORT", "")
diff --git a/common/conv/any.go b/common/conv/any.go
index 467e8bb7..33d34aa7 100644
--- a/common/conv/any.go
+++ b/common/conv/any.go
@@ -1,6 +1,9 @@
 package conv
 
 func AsString(v any) string {
-	str, _ := v.(string)
-	return str
+	if str, ok := v.(string); ok {
+		return str
+	}
+
+	return ""
 }
diff --git a/common/env/helper.go b/common/env/helper.go
index fdb9f827..44af5efd 100644
--- a/common/env/helper.go
+++ b/common/env/helper.go
@@ -3,13 +3,14 @@ package env
 import (
 	"os"
 	"strconv"
+	"strings"
 )
 
 func Bool(env string, defaultValue bool) bool {
 	if env == "" || os.Getenv(env) == "" {
 		return defaultValue
 	}
-	return os.Getenv(env) == "true"
+	return strings.ToLower(os.Getenv(env)) == "true"
 }
 
 func Int(env string, defaultValue int) int {
diff --git a/common/helper/audio.go b/common/helper/audio.go
new file mode 100644
index 00000000..9e99a947
--- /dev/null
+++ b/common/helper/audio.go
@@ -0,0 +1,62 @@
+package helper
+
+import (
+	"bytes"
+	"context"
+	"io"
+	"os"
+	"os/exec"
+	"strconv"
+
+	"github.com/pkg/errors"
+)
+
+// SaveTmpFile saves data to a temporary file. The filename would be apppended with a random string.
+func SaveTmpFile(filename string, data io.Reader) (string, error) {
+	if data == nil {
+		return "", errors.New("data is nil")
+	}
+
+	f, err := os.CreateTemp("", "*-"+filename)
+	if err != nil {
+		return "", errors.Wrapf(err, "failed to create temporary file %s", filename)
+	}
+	defer f.Close()
+
+	_, err = io.Copy(f, data)
+	if err != nil {
+		return "", errors.Wrapf(err, "failed to copy data to temporary file %s", filename)
+	}
+
+	return f.Name(), nil
+}
+
+// GetAudioTokens returns the number of tokens in an audio file.
+func GetAudioTokens(ctx context.Context, audio io.Reader, tokensPerSecond float64) (float64, error) {
+	filename, err := SaveTmpFile("audio", audio)
+	if err != nil {
+		return 0, errors.Wrap(err, "failed to save audio to temporary file")
+	}
+	defer os.Remove(filename)
+
+	duration, err := GetAudioDuration(ctx, filename)
+	if err != nil {
+		return 0, errors.Wrap(err, "failed to get audio tokens")
+	}
+
+	return duration * tokensPerSecond, nil
+}
+
+// GetAudioDuration returns the duration of an audio file in seconds.
+func GetAudioDuration(ctx context.Context, filename string) (float64, error) {
+	// ffprobe -v error -show_entries format=duration -of default=noprint_wrappers=1:nokey=1 {{input}}
+	c := exec.CommandContext(ctx, "/usr/bin/ffprobe", "-v", "error", "-show_entries", "format=duration", "-of", "default=noprint_wrappers=1:nokey=1", filename)
+	output, err := c.Output()
+	if err != nil {
+		return 0, errors.Wrap(err, "failed to get audio duration")
+	}
+
+	// Actually gpt-4-audio calculates tokens with 0.1s precision,
+	// while whisper calculates tokens with 1s precision
+	return strconv.ParseFloat(string(bytes.TrimSpace(output)), 64)
+}
diff --git a/common/helper/audio_test.go b/common/helper/audio_test.go
new file mode 100644
index 00000000..2fdede4d
--- /dev/null
+++ b/common/helper/audio_test.go
@@ -0,0 +1,68 @@
+package helper
+
+import (
+	"context"
+	"io"
+	"net/http"
+	"os"
+	"os/exec"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+func TestGetAudioDuration(t *testing.T) {
+	// skip if there is no ffmpeg installed
+	_, err := exec.LookPath("ffmpeg")
+	if err != nil {
+		t.Skip("ffmpeg not installed, skipping test")
+	}
+
+	t.Run("should return correct duration for a valid audio file", func(t *testing.T) {
+		tmpFile, err := os.CreateTemp("", "test_audio*.mp3")
+		require.NoError(t, err)
+		defer os.Remove(tmpFile.Name())
+
+		// download test audio file
+		resp, err := http.Get("https://s3.laisky.com/uploads/2025/01/audio-sample.m4a")
+		require.NoError(t, err)
+		defer resp.Body.Close()
+
+		_, err = io.Copy(tmpFile, resp.Body)
+		require.NoError(t, err)
+		require.NoError(t, tmpFile.Close())
+
+		duration, err := GetAudioDuration(context.Background(), tmpFile.Name())
+		require.NoError(t, err)
+		require.Equal(t, duration, 3.904)
+	})
+
+	t.Run("should return an error for a non-existent file", func(t *testing.T) {
+		_, err := GetAudioDuration(context.Background(), "non_existent_file.mp3")
+		require.Error(t, err)
+	})
+}
+
+func TestGetAudioTokens(t *testing.T) {
+	// skip if there is no ffmpeg installed
+	_, err := exec.LookPath("ffmpeg")
+	if err != nil {
+		t.Skip("ffmpeg not installed, skipping test")
+	}
+
+	t.Run("should return correct tokens for a valid audio file", func(t *testing.T) {
+		// download test audio file
+		resp, err := http.Get("https://s3.laisky.com/uploads/2025/01/audio-sample.m4a")
+		require.NoError(t, err)
+		defer resp.Body.Close()
+
+		tokens, err := GetAudioTokens(context.Background(), resp.Body, 50)
+		require.NoError(t, err)
+		require.Equal(t, tokens, 195.2)
+	})
+
+	t.Run("should return an error for a non-existent file", func(t *testing.T) {
+		_, err := GetAudioTokens(context.Background(), nil, 1)
+		require.Error(t, err)
+	})
+}
diff --git a/common/helper/helper.go b/common/helper/helper.go
index 65f4fd29..c80ee0a3 100644
--- a/common/helper/helper.go
+++ b/common/helper/helper.go
@@ -6,13 +6,13 @@ import (
 	"html/template"
 	"log"
 	"net"
+	"net/http"
 	"os/exec"
 	"runtime"
 	"strconv"
 	"strings"
 
 	"github.com/gin-gonic/gin"
-
 	"github.com/songquanpeng/one-api/common/random"
 )
 
@@ -32,6 +32,14 @@ func OpenBrowser(url string) {
 	}
 }
 
+// RespondError sends a JSON response with a success status and an error message.
+func RespondError(c *gin.Context, err error) {
+	c.JSON(http.StatusOK, gin.H{
+		"success": false,
+		"message": err.Error(),
+	})
+}
+
 func GetIp() (ip string) {
 	ips, err := net.InterfaceAddrs()
 	if err != nil {
diff --git a/common/helper/time.go b/common/helper/time.go
index f0bc6021..c1cda8d1 100644
--- a/common/helper/time.go
+++ b/common/helper/time.go
@@ -5,6 +5,7 @@ import (
 	"time"
 )
 
+// GetTimestamp get current timestamp in seconds
 func GetTimestamp() int64 {
 	return time.Now().Unix()
 }
diff --git a/common/image/image.go b/common/image/image.go
index beebd0c6..14488627 100644
--- a/common/image/image.go
+++ b/common/image/image.go
@@ -3,7 +3,6 @@ package image
 import (
 	"bytes"
 	"encoding/base64"
-	"github.com/songquanpeng/one-api/common/client"
 	"image"
 	_ "image/gif"
 	_ "image/jpeg"
@@ -13,6 +12,8 @@ import (
 	"strings"
 	"sync"
 
+	"github.com/pkg/errors"
+	"github.com/songquanpeng/one-api/common/client"
 	_ "golang.org/x/image/webp"
 )
 
@@ -22,27 +23,54 @@ var dataURLPattern = regexp.MustCompile(`data:image/([^;]+);base64,(.*)`)
 func IsImageUrl(url string) (bool, error) {
 	resp, err := client.UserContentRequestHTTPClient.Head(url)
 	if err != nil {
-		return false, err
+		return false, errors.Wrapf(err, "failed to fetch image URL: %s", url)
 	}
-	if !strings.HasPrefix(resp.Header.Get("Content-Type"), "image/") {
-		return false, nil
+	defer resp.Body.Close()
+
+	if resp.StatusCode == http.StatusNotFound {
+		// this file may not support HEAD method
+		resp, err = client.UserContentRequestHTTPClient.Get(url)
+		if err != nil {
+			return false, errors.Wrapf(err, "failed to fetch image URL: %s", url)
+		}
+		defer resp.Body.Close()
 	}
+
+	if resp.StatusCode != http.StatusOK {
+		return false, errors.Errorf("failed to fetch image URL: %s, status code: %d", url, resp.StatusCode)
+	}
+
+	if resp.ContentLength > 10*1024*1024 {
+		return false, errors.Errorf("image size should not exceed 10MB: %s, size: %d", url, resp.ContentLength)
+	}
+
+	contentType := strings.ToLower(resp.Header.Get("Content-Type"))
+	if !strings.HasPrefix(contentType, "image/") &&
+		!strings.Contains(contentType, "application/octet-stream") {
+		return false,
+			errors.Errorf("invalid content type: %s, expected image type", contentType)
+	}
+
 	return true, nil
 }
 
 func GetImageSizeFromUrl(url string) (width int, height int, err error) {
 	isImage, err := IsImageUrl(url)
+	if err != nil {
+		return 0, 0, errors.Wrap(err, "failed to fetch image URL")
+	}
 	if !isImage {
-		return
+		return 0, 0, errors.New("not an image URL")
 	}
 	resp, err := client.UserContentRequestHTTPClient.Get(url)
 	if err != nil {
-		return
+		return 0, 0, errors.Wrap(err, "failed to get image from URL")
 	}
 	defer resp.Body.Close()
+
 	img, _, err := image.DecodeConfig(resp.Body)
 	if err != nil {
-		return
+		return 0, 0, errors.Wrap(err, "failed to decode image")
 	}
 	return img.Width, img.Height, nil
 }
@@ -58,22 +86,35 @@ func GetImageFromUrl(url string) (mimeType string, data string, err error) {
 	}
 
 	isImage, err := IsImageUrl(url)
-	if !isImage {
-		return
+	if err != nil {
+		return mimeType, data, errors.Wrap(err, "failed to fetch image URL")
 	}
+	if !isImage {
+		return mimeType, data, errors.New("not an image URL")
+	}
+
 	resp, err := http.Get(url)
 	if err != nil {
-		return
+		return mimeType, data, errors.Wrap(err, "failed to get image from URL")
 	}
 	defer resp.Body.Close()
+
+	if resp.StatusCode != http.StatusOK {
+		return mimeType, data, errors.Errorf("failed to fetch image URL: %s, status code: %d", url, resp.StatusCode)
+	}
+	if resp.ContentLength > 10*1024*1024 {
+		return mimeType, data, errors.Errorf("image size should not exceed 10MB: %s, size: %d", url, resp.ContentLength)
+	}
+
 	buffer := bytes.NewBuffer(nil)
 	_, err = buffer.ReadFrom(resp.Body)
 	if err != nil {
-		return
+		return mimeType, data, errors.Wrap(err, "failed to read image data from response")
 	}
+
 	mimeType = resp.Header.Get("Content-Type")
 	data = base64.StdEncoding.EncodeToString(buffer.Bytes())
-	return
+	return mimeType, data, nil
 }
 
 var (
diff --git a/common/image/image_test.go b/common/image/image_test.go
index 5b669b51..5774ef1d 100644
--- a/common/image/image_test.go
+++ b/common/image/image_test.go
@@ -1,8 +1,8 @@
 package image_test
 
 import (
+	"bytes"
 	"encoding/base64"
-	"github.com/songquanpeng/one-api/common/client"
 	"image"
 	_ "image/gif"
 	_ "image/jpeg"
@@ -13,8 +13,8 @@ import (
 	"strings"
 	"testing"
 
+	"github.com/songquanpeng/one-api/common/client"
 	img "github.com/songquanpeng/one-api/common/image"
-
 	"github.com/stretchr/testify/assert"
 	_ "golang.org/x/image/webp"
 )
@@ -51,6 +51,8 @@ func TestMain(m *testing.M) {
 }
 
 func TestDecode(t *testing.T) {
+	t.Parallel()
+
 	// Bytes read: varies sometimes
 	// jpeg: 1063892
 	// png: 294462
@@ -96,6 +98,8 @@ func TestDecode(t *testing.T) {
 }
 
 func TestBase64(t *testing.T) {
+	t.Parallel()
+
 	// Bytes read:
 	// jpeg: 1063892
 	// png: 294462
@@ -149,6 +153,8 @@ func TestBase64(t *testing.T) {
 }
 
 func TestGetImageSize(t *testing.T) {
+	t.Parallel()
+
 	for i, c := range cases {
 		t.Run("Decode:"+strconv.Itoa(i), func(t *testing.T) {
 			width, height, err := img.GetImageSize(c.url)
@@ -160,6 +166,8 @@ func TestGetImageSize(t *testing.T) {
 }
 
 func TestGetImageSizeFromBase64(t *testing.T) {
+	t.Parallel()
+
 	for i, c := range cases {
 		t.Run("Decode:"+strconv.Itoa(i), func(t *testing.T) {
 			resp, err := http.Get(c.url)
@@ -175,3 +183,83 @@ func TestGetImageSizeFromBase64(t *testing.T) {
 		})
 	}
 }
+
+func TestGetImageFromUrl(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name       string
+		input      string
+		wantMime   string
+		wantErr    bool
+		errMessage string
+	}{
+		{
+			name:     "Valid JPEG URL",
+			input:    cases[0].url, // Using the existing JPEG test case
+			wantMime: "image/jpeg",
+			wantErr:  false,
+		},
+		{
+			name:     "Valid PNG URL",
+			input:    cases[1].url, // Using the existing PNG test case
+			wantMime: "image/png",
+			wantErr:  false,
+		},
+		{
+			name:     "Valid Data URL",
+			input:    "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mP8z8BQDwAEhQGAhKmMIQAAAABJRU5ErkJggg==",
+			wantMime: "image/png",
+			wantErr:  false,
+		},
+		{
+			name:       "Invalid URL",
+			input:      "https://invalid.example.com/nonexistent.jpg",
+			wantErr:    true,
+			errMessage: "failed to fetch image URL",
+		},
+		{
+			name:       "Non-image URL",
+			input:      "https://ario.laisky.com/alias/doc",
+			wantErr:    true,
+			errMessage: "invalid content type",
+		},
+	}
+
+	for _, tt := range tests {
+		tt := tt // capture range variable
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			mimeType, data, err := img.GetImageFromUrl(tt.input)
+
+			if tt.wantErr {
+				assert.Error(t, err)
+				if tt.errMessage != "" {
+					assert.Contains(t, err.Error(), tt.errMessage)
+				}
+				return
+			}
+
+			assert.NoError(t, err)
+			assert.NotEmpty(t, data)
+
+			// For data URLs, we should verify the mime type matches the input
+			if strings.HasPrefix(tt.input, "data:image/") {
+				assert.Equal(t, tt.wantMime, mimeType)
+				return
+			}
+
+			// For regular URLs, verify the base64 data is valid and can be decoded
+			decoded, err := base64.StdEncoding.DecodeString(data)
+			assert.NoError(t, err)
+			assert.NotEmpty(t, decoded)
+
+			// Verify the decoded data is a valid image
+			reader := bytes.NewReader(decoded)
+			_, format, err := image.DecodeConfig(reader)
+			assert.NoError(t, err)
+			assert.Equal(t, strings.TrimPrefix(tt.wantMime, "image/"), format)
+		})
+	}
+}
diff --git a/common/message/email.go b/common/message/email.go
index 85a83d6d..02074e8b 100644
--- a/common/message/email.go
+++ b/common/message/email.go
@@ -10,6 +10,7 @@ import (
 	"strings"
 	"time"
 
+	"github.com/pkg/errors"
 	"github.com/songquanpeng/one-api/common/config"
 	"github.com/songquanpeng/one-api/common/logger"
 )
@@ -27,17 +28,17 @@ func SendEmail(subject string, receiver string, content string) error {
 	}
 	encodedSubject := fmt.Sprintf("=?UTF-8?B?%s?=", base64.StdEncoding.EncodeToString([]byte(subject)))
 
-	// Extract domain from SMTPFrom
+	// Extract domain from SMTPFrom with fallback
+	domain := "localhost"
 	parts := strings.Split(config.SMTPFrom, "@")
-	var domain string
-	if len(parts) > 1 {
+	if len(parts) > 1 && parts[1] != "" {
 		domain = parts[1]
 	}
+
 	// Generate a unique Message-ID
 	buf := make([]byte, 16)
-	_, err := rand.Read(buf)
-	if err != nil {
-		return err
+	if _, err := rand.Read(buf); err != nil {
+		return errors.Wrap(err, "failed to generate random bytes for Message-ID")
 	}
 	messageId := fmt.Sprintf("<%x@%s>", buf, domain)
 
@@ -50,59 +51,85 @@ func SendEmail(subject string, receiver string, content string) error {
 		receiver, config.SystemName, config.SMTPFrom, encodedSubject, messageId, time.Now().Format(time.RFC1123Z), content))
 
 	auth := smtp.PlainAuth("", config.SMTPAccount, config.SMTPToken, config.SMTPServer)
-	addr := fmt.Sprintf("%s:%d", config.SMTPServer, config.SMTPPort)
-	to := strings.Split(receiver, ";")
+	addr := net.JoinHostPort(config.SMTPServer, fmt.Sprintf("%d", config.SMTPPort))
+
+	// Clean up recipient addresses
+	receiverEmails := []string{}
+	for _, email := range strings.Split(receiver, ";") {
+		email = strings.TrimSpace(email)
+		if email != "" {
+			receiverEmails = append(receiverEmails, email)
+		}
+	}
+
+	if len(receiverEmails) == 0 {
+		return errors.New("no valid recipient email addresses")
+	}
 
 	if config.SMTPPort == 465 || !shouldAuth() {
 		// need advanced client
 		var conn net.Conn
 		var err error
+
+		// Add connection timeout
+		dialer := &net.Dialer{
+			Timeout: 30 * time.Second,
+		}
+
 		if config.SMTPPort == 465 {
 			tlsConfig := &tls.Config{
-				InsecureSkipVerify: true,
+				InsecureSkipVerify: !config.ForceEmailTLSVerify,
 				ServerName:         config.SMTPServer,
 			}
-			conn, err = tls.Dial("tcp", fmt.Sprintf("%s:%d", config.SMTPServer, config.SMTPPort), tlsConfig)
+			conn, err = tls.DialWithDialer(dialer, "tcp", addr, tlsConfig)
 		} else {
-			conn, err = net.Dial("tcp", fmt.Sprintf("%s:%d", config.SMTPServer, config.SMTPPort))
+			conn, err = dialer.Dial("tcp", addr)
 		}
+
 		if err != nil {
-			return err
+			return errors.Wrap(err, "failed to connect to SMTP server")
 		}
+
 		client, err := smtp.NewClient(conn, config.SMTPServer)
 		if err != nil {
-			return err
+			return errors.Wrap(err, "failed to create SMTP client")
 		}
 		defer client.Close()
+
 		if shouldAuth() {
 			if err = client.Auth(auth); err != nil {
-				return err
+				return errors.Wrap(err, "SMTP authentication failed")
 			}
 		}
+
 		if err = client.Mail(config.SMTPFrom); err != nil {
-			return err
+			return errors.Wrap(err, "failed to set MAIL FROM")
 		}
-		receiverEmails := strings.Split(receiver, ";")
+
 		for _, receiver := range receiverEmails {
 			if err = client.Rcpt(receiver); err != nil {
-				return err
+				return errors.Wrapf(err, "failed to add recipient: %s", receiver)
 			}
 		}
+
 		w, err := client.Data()
 		if err != nil {
-			return err
+			return errors.Wrap(err, "failed to create message data writer")
 		}
-		_, err = w.Write(mail)
-		if err != nil {
-			return err
+
+		if _, err = w.Write(mail); err != nil {
+			return errors.Wrap(err, "failed to write email content")
 		}
-		err = w.Close()
-		if err != nil {
-			return err
+
+		if err = w.Close(); err != nil {
+			return errors.Wrap(err, "failed to close message data writer")
 		}
+
 		return nil
 	}
-	err = smtp.SendMail(addr, auth, config.SMTPAccount, to, mail)
+
+	// Use the same sender address in the SMTP protocol as in the From header
+	err := smtp.SendMail(addr, auth, config.SMTPFrom, receiverEmails, mail)
 	if err != nil && strings.Contains(err.Error(), "short response") { // 部分提供商返回该错误，但实际上邮件已经发送成功
 		logger.SysWarnf("short response from SMTP server, return nil instead of error: %s", err.Error())
 		return nil
diff --git a/controller/channel-test.go b/controller/channel-test.go
index 3894691c..849aaef5 100644
--- a/controller/channel-test.go
+++ b/controller/channel-test.go
@@ -106,6 +106,8 @@ func testChannel(ctx context.Context, channel *model.Channel, request *relaymode
 	if err != nil {
 		return "", err, nil
 	}
+	c.Set(ctxkey.ConvertedRequest, convertedRequest)
+
 	jsonData, err := json.Marshal(convertedRequest)
 	if err != nil {
 		return "", err, nil
diff --git a/model/ability.go b/model/ability.go
index 5cfb9949..6180fbe0 100644
--- a/model/ability.go
+++ b/model/ability.go
@@ -7,6 +7,7 @@ import (
 
 	"gorm.io/gorm"
 
+	"github.com/pkg/errors"
 	"github.com/songquanpeng/one-api/common"
 	"github.com/songquanpeng/one-api/common/utils"
 )
@@ -42,7 +43,7 @@ func GetRandomSatisfiedChannel(group string, model string, ignoreFirstPriority b
 		err = channelQuery.Order("RAND()").First(&ability).Error
 	}
 	if err != nil {
-		return nil, err
+		return nil, errors.Wrap(err, "get random satisfied channel")
 	}
 	channel := Channel{}
 	channel.Id = ability.ChannelId
diff --git a/relay/adaptor/aiproxy/adaptor.go b/relay/adaptor/aiproxy/adaptor.go
index 42d49c0a..d85f42d1 100644
--- a/relay/adaptor/aiproxy/adaptor.go
+++ b/relay/adaptor/aiproxy/adaptor.go
@@ -38,7 +38,7 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	return aiProxyLibraryRequest, nil
 }
 
-func (a *Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+func (a *Adaptor) ConvertImageRequest(_ *gin.Context, request *model.ImageRequest) (any, error) {
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
diff --git a/relay/adaptor/ali/adaptor.go b/relay/adaptor/ali/adaptor.go
index 4aa8a11a..79b51ac3 100644
--- a/relay/adaptor/ali/adaptor.go
+++ b/relay/adaptor/ali/adaptor.go
@@ -67,7 +67,7 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	}
 }
 
-func (a *Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+func (a *Adaptor) ConvertImageRequest(_ *gin.Context, request *model.ImageRequest) (any, error) {
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
diff --git a/relay/adaptor/anthropic/adaptor.go b/relay/adaptor/anthropic/adaptor.go
index bd0949be..cd005b7c 100644
--- a/relay/adaptor/anthropic/adaptor.go
+++ b/relay/adaptor/anthropic/adaptor.go
@@ -36,8 +36,8 @@ func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *me
 
 	// https://x.com/alexalbert__/status/1812921642143900036
 	// claude-3-5-sonnet can support 8k context
-	if strings.HasPrefix(meta.ActualModelName, "claude-3-5-sonnet") {
-		req.Header.Set("anthropic-beta", "max-tokens-3-5-sonnet-2024-07-15")
+	if strings.HasPrefix(meta.ActualModelName, "claude-3-7-sonnet") {
+		req.Header.Set("anthropic-beta", "output-128k-2025-02-19")
 	}
 
 	return nil
@@ -47,10 +47,10 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
-	return ConvertRequest(*request), nil
+	return ConvertRequest(c, *request)
 }
 
-func (a *Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+func (a *Adaptor) ConvertImageRequest(_ *gin.Context, request *model.ImageRequest) (any, error) {
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
diff --git a/relay/adaptor/anthropic/constants.go b/relay/adaptor/anthropic/constants.go
index 9b515c1c..b3a05ee8 100644
--- a/relay/adaptor/anthropic/constants.go
+++ b/relay/adaptor/anthropic/constants.go
@@ -3,11 +3,13 @@ package anthropic
 var ModelList = []string{
 	"claude-instant-1.2", "claude-2.0", "claude-2.1",
 	"claude-3-haiku-20240307",
-	"claude-3-5-haiku-20241022",
 	"claude-3-5-haiku-latest",
+	"claude-3-5-haiku-20241022",
 	"claude-3-sonnet-20240229",
 	"claude-3-opus-20240229",
+	"claude-3-5-sonnet-latest",
 	"claude-3-5-sonnet-20240620",
 	"claude-3-5-sonnet-20241022",
-	"claude-3-5-sonnet-latest",
+	"claude-3-7-sonnet-latest",
+	"claude-3-7-sonnet-20250219",
 }
diff --git a/relay/adaptor/anthropic/main.go b/relay/adaptor/anthropic/main.go
index d3e306c8..78765ed6 100644
--- a/relay/adaptor/anthropic/main.go
+++ b/relay/adaptor/anthropic/main.go
@@ -2,18 +2,21 @@ package anthropic
 
 import (
 	"bufio"
+	"context"
 	"encoding/json"
 	"fmt"
-	"github.com/songquanpeng/one-api/common/render"
 	"io"
+	"math"
 	"net/http"
 	"strings"
 
 	"github.com/gin-gonic/gin"
+	"github.com/pkg/errors"
 	"github.com/songquanpeng/one-api/common"
 	"github.com/songquanpeng/one-api/common/helper"
 	"github.com/songquanpeng/one-api/common/image"
 	"github.com/songquanpeng/one-api/common/logger"
+	"github.com/songquanpeng/one-api/common/render"
 	"github.com/songquanpeng/one-api/relay/adaptor/openai"
 	"github.com/songquanpeng/one-api/relay/model"
 )
@@ -36,7 +39,16 @@ func stopReasonClaude2OpenAI(reason *string) string {
 	}
 }
 
-func ConvertRequest(textRequest model.GeneralOpenAIRequest) *Request {
+// isModelSupportThinking is used to check if the model supports extended thinking
+func isModelSupportThinking(model string) bool {
+	if strings.Contains(model, "claude-3-7-sonnet") {
+		return true
+	}
+
+	return false
+}
+
+func ConvertRequest(c *gin.Context, textRequest model.GeneralOpenAIRequest) (*Request, error) {
 	claudeTools := make([]Tool, 0, len(textRequest.Tools))
 
 	for _, tool := range textRequest.Tools {
@@ -61,7 +73,27 @@ func ConvertRequest(textRequest model.GeneralOpenAIRequest) *Request {
 		TopK:        textRequest.TopK,
 		Stream:      textRequest.Stream,
 		Tools:       claudeTools,
+		Thinking:    textRequest.Thinking,
 	}
+
+	if isModelSupportThinking(textRequest.Model) &&
+		c.Request.URL.Query().Has("thinking") && claudeRequest.Thinking == nil {
+		claudeRequest.Thinking = &model.Thinking{
+			Type:         "enabled",
+			BudgetTokens: int(math.Min(1024, float64(claudeRequest.MaxTokens/2))),
+		}
+	}
+
+	if isModelSupportThinking(textRequest.Model) &&
+		claudeRequest.Thinking != nil {
+		if claudeRequest.MaxTokens <= 1024 {
+			return nil, errors.New("max_tokens must be greater than 1024 when using extended thinking")
+		}
+
+		// top_p must be nil when using extended thinking
+		claudeRequest.TopP = nil
+	}
+
 	if len(claudeTools) > 0 {
 		claudeToolChoice := struct {
 			Type string `json:"type"`
@@ -127,7 +159,9 @@ func ConvertRequest(textRequest model.GeneralOpenAIRequest) *Request {
 			var content Content
 			if part.Type == model.ContentTypeText {
 				content.Type = "text"
-				content.Text = part.Text
+				if part.Text != nil {
+					content.Text = *part.Text
+				}
 			} else if part.Type == model.ContentTypeImageURL {
 				content.Type = "image"
 				content.Source = &ImageSource{
@@ -142,13 +176,14 @@ func ConvertRequest(textRequest model.GeneralOpenAIRequest) *Request {
 		claudeMessage.Content = contents
 		claudeRequest.Messages = append(claudeRequest.Messages, claudeMessage)
 	}
-	return &claudeRequest
+	return &claudeRequest, nil
 }
 
 // https://docs.anthropic.com/claude/reference/messages-streaming
 func StreamResponseClaude2OpenAI(claudeResponse *StreamResponse) (*openai.ChatCompletionsStreamResponse, *Response) {
 	var response *Response
 	var responseText string
+	var reasoningText string
 	var stopReason string
 	tools := make([]model.Tool, 0)
 
@@ -158,6 +193,10 @@ func StreamResponseClaude2OpenAI(claudeResponse *StreamResponse) (*openai.ChatCo
 	case "content_block_start":
 		if claudeResponse.ContentBlock != nil {
 			responseText = claudeResponse.ContentBlock.Text
+			if claudeResponse.ContentBlock.Thinking != nil {
+				reasoningText = *claudeResponse.ContentBlock.Thinking
+			}
+
 			if claudeResponse.ContentBlock.Type == "tool_use" {
 				tools = append(tools, model.Tool{
 					Id:   claudeResponse.ContentBlock.Id,
@@ -172,6 +211,10 @@ func StreamResponseClaude2OpenAI(claudeResponse *StreamResponse) (*openai.ChatCo
 	case "content_block_delta":
 		if claudeResponse.Delta != nil {
 			responseText = claudeResponse.Delta.Text
+			if claudeResponse.Delta.Thinking != nil {
+				reasoningText = *claudeResponse.Delta.Thinking
+			}
+
 			if claudeResponse.Delta.Type == "input_json_delta" {
 				tools = append(tools, model.Tool{
 					Function: model.Function{
@@ -189,9 +232,20 @@ func StreamResponseClaude2OpenAI(claudeResponse *StreamResponse) (*openai.ChatCo
 		if claudeResponse.Delta != nil && claudeResponse.Delta.StopReason != nil {
 			stopReason = *claudeResponse.Delta.StopReason
 		}
+	case "thinking_delta":
+		if claudeResponse.Delta != nil && claudeResponse.Delta.Thinking != nil {
+			reasoningText = *claudeResponse.Delta.Thinking
+		}
+	case "ping",
+		"message_stop",
+		"content_block_stop":
+	default:
+		logger.SysErrorf("unknown stream response type %q", claudeResponse.Type)
 	}
+
 	var choice openai.ChatCompletionsStreamResponseChoice
 	choice.Delta.Content = responseText
+	choice.Delta.Reasoning = &reasoningText
 	if len(tools) > 0 {
 		choice.Delta.Content = nil // compatible with other OpenAI derivative applications, like LobeOpenAICompatibleFactory ...
 		choice.Delta.ToolCalls = tools
@@ -209,11 +263,23 @@ func StreamResponseClaude2OpenAI(claudeResponse *StreamResponse) (*openai.ChatCo
 
 func ResponseClaude2OpenAI(claudeResponse *Response) *openai.TextResponse {
 	var responseText string
-	if len(claudeResponse.Content) > 0 {
-		responseText = claudeResponse.Content[0].Text
-	}
+	var reasoningText string
+
 	tools := make([]model.Tool, 0)
 	for _, v := range claudeResponse.Content {
+		switch v.Type {
+		case "thinking":
+			if v.Thinking != nil {
+				reasoningText += *v.Thinking
+			} else {
+				logger.Errorf(context.Background(), "thinking is nil in response")
+			}
+		case "text":
+			responseText += v.Text
+		default:
+			logger.Warnf(context.Background(), "unknown response type %q", v.Type)
+		}
+
 		if v.Type == "tool_use" {
 			args, _ := json.Marshal(v.Input)
 			tools = append(tools, model.Tool{
@@ -226,11 +292,13 @@ func ResponseClaude2OpenAI(claudeResponse *Response) *openai.TextResponse {
 			})
 		}
 	}
+
 	choice := openai.TextResponseChoice{
 		Index: 0,
 		Message: model.Message{
 			Role:      "assistant",
 			Content:   responseText,
+			Reasoning: &reasoningText,
 			Name:      nil,
 			ToolCalls: tools,
 		},
@@ -277,6 +345,8 @@ func StreamHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusC
 		data = strings.TrimPrefix(data, "data:")
 		data = strings.TrimSpace(data)
 
+		logger.Debugf(c.Request.Context(), "stream <- %q\n", data)
+
 		var claudeResponse StreamResponse
 		err := json.Unmarshal([]byte(data), &claudeResponse)
 		if err != nil {
@@ -344,6 +414,9 @@ func Handler(c *gin.Context, resp *http.Response, promptTokens int, modelName st
 	if err != nil {
 		return openai.ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), nil
 	}
+
+	logger.Debugf(c.Request.Context(), "response <- %s\n", string(responseBody))
+
 	var claudeResponse Response
 	err = json.Unmarshal(responseBody, &claudeResponse)
 	if err != nil {
diff --git a/relay/adaptor/anthropic/model.go b/relay/adaptor/anthropic/model.go
index 47f193fa..6dd299c4 100644
--- a/relay/adaptor/anthropic/model.go
+++ b/relay/adaptor/anthropic/model.go
@@ -1,5 +1,7 @@
 package anthropic
 
+import "github.com/songquanpeng/one-api/relay/model"
+
 // https://docs.anthropic.com/claude/reference/messages_post
 
 type Metadata struct {
@@ -22,6 +24,9 @@ type Content struct {
 	Input     any    `json:"input,omitempty"`
 	Content   string `json:"content,omitempty"`
 	ToolUseId string `json:"tool_use_id,omitempty"`
+	// https://docs.anthropic.com/en/docs/build-with-claude/extended-thinking#implementing-extended-thinking
+	Thinking  *string `json:"thinking,omitempty"`
+	Signature *string `json:"signature,omitempty"`
 }
 
 type Message struct {
@@ -54,6 +59,7 @@ type Request struct {
 	Tools         []Tool    `json:"tools,omitempty"`
 	ToolChoice    any       `json:"tool_choice,omitempty"`
 	//Metadata    `json:"metadata,omitempty"`
+	Thinking *model.Thinking `json:"thinking,omitempty"`
 }
 
 type Usage struct {
@@ -84,6 +90,8 @@ type Delta struct {
 	PartialJson  string  `json:"partial_json,omitempty"`
 	StopReason   *string `json:"stop_reason"`
 	StopSequence *string `json:"stop_sequence"`
+	Thinking     *string `json:"thinking,omitempty"`
+	Signature    *string `json:"signature,omitempty"`
 }
 
 type StreamResponse struct {
diff --git a/relay/adaptor/aws/adaptor.go b/relay/adaptor/aws/adaptor.go
index 62221346..45bfbdf6 100644
--- a/relay/adaptor/aws/adaptor.go
+++ b/relay/adaptor/aws/adaptor.go
@@ -72,7 +72,7 @@ func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *me
 	return nil
 }
 
-func (a *Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+func (a *Adaptor) ConvertImageRequest(_ *gin.Context, request *model.ImageRequest) (any, error) {
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
diff --git a/relay/adaptor/aws/claude/adapter.go b/relay/adaptor/aws/claude/adapter.go
index eb3c9fb8..2f6a4cc5 100644
--- a/relay/adaptor/aws/claude/adapter.go
+++ b/relay/adaptor/aws/claude/adapter.go
@@ -21,7 +21,11 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 		return nil, errors.New("request is nil")
 	}
 
-	claudeReq := anthropic.ConvertRequest(*request)
+	claudeReq, err := anthropic.ConvertRequest(c, *request)
+	if err != nil {
+		return nil, errors.Wrap(err, "convert request")
+	}
+
 	c.Set(ctxkey.RequestModel, request.Model)
 	c.Set(ctxkey.ConvertedRequest, claudeReq)
 	return claudeReq, nil
diff --git a/relay/adaptor/aws/claude/main.go b/relay/adaptor/aws/claude/main.go
index 3fe3dfd8..c20827b0 100644
--- a/relay/adaptor/aws/claude/main.go
+++ b/relay/adaptor/aws/claude/main.go
@@ -36,6 +36,8 @@ var AwsModelIDMap = map[string]string{
 	"claude-3-5-sonnet-20241022": "anthropic.claude-3-5-sonnet-20241022-v2:0",
 	"claude-3-5-sonnet-latest":   "anthropic.claude-3-5-sonnet-20241022-v2:0",
 	"claude-3-5-haiku-20241022":  "anthropic.claude-3-5-haiku-20241022-v1:0",
+	"claude-3-7-sonnet-latest":   "anthropic.claude-3-7-sonnet-20250219-v1:0",
+	"claude-3-7-sonnet-20250219": "anthropic.claude-3-7-sonnet-20250219-v1:0",
 }
 
 func awsModelID(requestModel string) (string, error) {
@@ -47,13 +49,14 @@ func awsModelID(requestModel string) (string, error) {
 }
 
 func Handler(c *gin.Context, awsCli *bedrockruntime.Client, modelName string) (*relaymodel.ErrorWithStatusCode, *relaymodel.Usage) {
-	awsModelId, err := awsModelID(c.GetString(ctxkey.RequestModel))
+	awsModelID, err := awsModelID(c.GetString(ctxkey.RequestModel))
 	if err != nil {
 		return utils.WrapErr(errors.Wrap(err, "awsModelID")), nil
 	}
 
+	awsModelID = utils.ConvertModelID2CrossRegionProfile(awsModelID, awsCli.Options().Region)
 	awsReq := &bedrockruntime.InvokeModelInput{
-		ModelId:     aws.String(awsModelId),
+		ModelId:     aws.String(awsModelID),
 		Accept:      aws.String("application/json"),
 		ContentType: aws.String("application/json"),
 	}
@@ -101,13 +104,14 @@ func Handler(c *gin.Context, awsCli *bedrockruntime.Client, modelName string) (*
 
 func StreamHandler(c *gin.Context, awsCli *bedrockruntime.Client) (*relaymodel.ErrorWithStatusCode, *relaymodel.Usage) {
 	createdTime := helper.GetTimestamp()
-	awsModelId, err := awsModelID(c.GetString(ctxkey.RequestModel))
+	awsModelID, err := awsModelID(c.GetString(ctxkey.RequestModel))
 	if err != nil {
 		return utils.WrapErr(errors.Wrap(err, "awsModelID")), nil
 	}
 
+	awsModelID = utils.ConvertModelID2CrossRegionProfile(awsModelID, awsCli.Options().Region)
 	awsReq := &bedrockruntime.InvokeModelWithResponseStreamInput{
-		ModelId:     aws.String(awsModelId),
+		ModelId:     aws.String(awsModelID),
 		Accept:      aws.String("application/json"),
 		ContentType: aws.String("application/json"),
 	}
diff --git a/relay/adaptor/aws/claude/model.go b/relay/adaptor/aws/claude/model.go
index 10622887..b0dd6800 100644
--- a/relay/adaptor/aws/claude/model.go
+++ b/relay/adaptor/aws/claude/model.go
@@ -1,6 +1,9 @@
 package aws
 
-import "github.com/songquanpeng/one-api/relay/adaptor/anthropic"
+import (
+	"github.com/songquanpeng/one-api/relay/adaptor/anthropic"
+	"github.com/songquanpeng/one-api/relay/model"
+)
 
 // Request is the request to AWS Claude
 //
@@ -17,4 +20,5 @@ type Request struct {
 	StopSequences    []string            `json:"stop_sequences,omitempty"`
 	Tools            []anthropic.Tool    `json:"tools,omitempty"`
 	ToolChoice       any                 `json:"tool_choice,omitempty"`
+	Thinking         *model.Thinking     `json:"thinking,omitempty"`
 }
diff --git a/relay/adaptor/aws/llama3/main.go b/relay/adaptor/aws/llama3/main.go
index e5fcd89f..76b06f91 100644
--- a/relay/adaptor/aws/llama3/main.go
+++ b/relay/adaptor/aws/llama3/main.go
@@ -70,13 +70,14 @@ func ConvertRequest(textRequest relaymodel.GeneralOpenAIRequest) *Request {
 }
 
 func Handler(c *gin.Context, awsCli *bedrockruntime.Client, modelName string) (*relaymodel.ErrorWithStatusCode, *relaymodel.Usage) {
-	awsModelId, err := awsModelID(c.GetString(ctxkey.RequestModel))
+	awsModelID, err := awsModelID(c.GetString(ctxkey.RequestModel))
 	if err != nil {
 		return utils.WrapErr(errors.Wrap(err, "awsModelID")), nil
 	}
 
+	awsModelID = utils.ConvertModelID2CrossRegionProfile(awsModelID, awsCli.Options().Region)
 	awsReq := &bedrockruntime.InvokeModelInput{
-		ModelId:     aws.String(awsModelId),
+		ModelId:     aws.String(awsModelID),
 		Accept:      aws.String("application/json"),
 		ContentType: aws.String("application/json"),
 	}
@@ -140,13 +141,14 @@ func ResponseLlama2OpenAI(llamaResponse *Response) *openai.TextResponse {
 
 func StreamHandler(c *gin.Context, awsCli *bedrockruntime.Client) (*relaymodel.ErrorWithStatusCode, *relaymodel.Usage) {
 	createdTime := helper.GetTimestamp()
-	awsModelId, err := awsModelID(c.GetString(ctxkey.RequestModel))
+	awsModelID, err := awsModelID(c.GetString(ctxkey.RequestModel))
 	if err != nil {
 		return utils.WrapErr(errors.Wrap(err, "awsModelID")), nil
 	}
 
+	awsModelID = utils.ConvertModelID2CrossRegionProfile(awsModelID, awsCli.Options().Region)
 	awsReq := &bedrockruntime.InvokeModelWithResponseStreamInput{
-		ModelId:     aws.String(awsModelId),
+		ModelId:     aws.String(awsModelID),
 		Accept:      aws.String("application/json"),
 		ContentType: aws.String("application/json"),
 	}
diff --git a/relay/adaptor/aws/utils/adaptor.go b/relay/adaptor/aws/utils/adaptor.go
index 4cb880f2..f5fc0038 100644
--- a/relay/adaptor/aws/utils/adaptor.go
+++ b/relay/adaptor/aws/utils/adaptor.go
@@ -39,7 +39,7 @@ func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *me
 	return nil
 }
 
-func (a *Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+func (a *Adaptor) ConvertImageRequest(_ *gin.Context, request *model.ImageRequest) (any, error) {
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
diff --git a/relay/adaptor/aws/utils/consts.go b/relay/adaptor/aws/utils/consts.go
new file mode 100644
index 00000000..c91f342e
--- /dev/null
+++ b/relay/adaptor/aws/utils/consts.go
@@ -0,0 +1,75 @@
+package utils
+
+import (
+	"context"
+	"slices"
+	"strings"
+
+	"github.com/songquanpeng/one-api/common/logger"
+)
+
+// CrossRegionInferences is a list of model IDs that support cross-region inference.
+//
+// https://docs.aws.amazon.com/bedrock/latest/userguide/inference-profiles-support.html
+//
+// document.querySelectorAll('pre.programlisting code').forEach((e) => {console.log(e.innerHTML)})
+var CrossRegionInferences = []string{
+	"us.amazon.nova-lite-v1:0",
+	"us.amazon.nova-micro-v1:0",
+	"us.amazon.nova-pro-v1:0",
+	"us.anthropic.claude-3-5-haiku-20241022-v1:0",
+	"us.anthropic.claude-3-5-sonnet-20240620-v1:0",
+	"us.anthropic.claude-3-5-sonnet-20241022-v2:0",
+	"us.anthropic.claude-3-7-sonnet-20250219-v1:0",
+	"us.anthropic.claude-3-haiku-20240307-v1:0",
+	"us.anthropic.claude-3-opus-20240229-v1:0",
+	"us.anthropic.claude-3-sonnet-20240229-v1:0",
+	"us.meta.llama3-1-405b-instruct-v1:0",
+	"us.meta.llama3-1-70b-instruct-v1:0",
+	"us.meta.llama3-1-8b-instruct-v1:0",
+	"us.meta.llama3-2-11b-instruct-v1:0",
+	"us.meta.llama3-2-1b-instruct-v1:0",
+	"us.meta.llama3-2-3b-instruct-v1:0",
+	"us.meta.llama3-2-90b-instruct-v1:0",
+	"us.meta.llama3-3-70b-instruct-v1:0",
+	"us-gov.anthropic.claude-3-5-sonnet-20240620-v1:0",
+	"us-gov.anthropic.claude-3-haiku-20240307-v1:0",
+	"eu.amazon.nova-lite-v1:0",
+	"eu.amazon.nova-micro-v1:0",
+	"eu.amazon.nova-pro-v1:0",
+	"eu.anthropic.claude-3-5-sonnet-20240620-v1:0",
+	"eu.anthropic.claude-3-haiku-20240307-v1:0",
+	"eu.anthropic.claude-3-sonnet-20240229-v1:0",
+	"eu.meta.llama3-2-1b-instruct-v1:0",
+	"eu.meta.llama3-2-3b-instruct-v1:0",
+	"apac.amazon.nova-lite-v1:0",
+	"apac.amazon.nova-micro-v1:0",
+	"apac.amazon.nova-pro-v1:0",
+	"apac.anthropic.claude-3-5-sonnet-20240620-v1:0",
+	"apac.anthropic.claude-3-5-sonnet-20241022-v2:0",
+	"apac.anthropic.claude-3-haiku-20240307-v1:0",
+	"apac.anthropic.claude-3-sonnet-20240229-v1:0",
+}
+
+// ConvertModelID2CrossRegionProfile converts the model ID to a cross-region profile ID.
+func ConvertModelID2CrossRegionProfile(model, region string) string {
+	var regionPrefix string
+	switch prefix := strings.Split(region, "-")[0]; prefix {
+	case "us", "eu":
+		regionPrefix = prefix
+	case "ap":
+		regionPrefix = "apac"
+	default:
+		// not supported, return original model
+		return model
+	}
+
+	newModelID := regionPrefix + "." + model
+	if slices.Contains(CrossRegionInferences, newModelID) {
+		logger.Debugf(context.TODO(), "convert model %s to cross-region profile %s", model, newModelID)
+		return newModelID
+	}
+
+	// not found, return original model
+	return model
+}
diff --git a/relay/adaptor/baidu/adaptor.go b/relay/adaptor/baidu/adaptor.go
index 15306b95..664e0e77 100644
--- a/relay/adaptor/baidu/adaptor.go
+++ b/relay/adaptor/baidu/adaptor.go
@@ -109,7 +109,7 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	}
 }
 
-func (a *Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+func (a *Adaptor) ConvertImageRequest(_ *gin.Context, request *model.ImageRequest) (any, error) {
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
diff --git a/relay/adaptor/cloudflare/adaptor.go b/relay/adaptor/cloudflare/adaptor.go
index 97e3dbb2..8958466d 100644
--- a/relay/adaptor/cloudflare/adaptor.go
+++ b/relay/adaptor/cloudflare/adaptor.go
@@ -19,7 +19,7 @@ type Adaptor struct {
 }
 
 // ConvertImageRequest implements adaptor.Adaptor.
-func (*Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+func (*Adaptor) ConvertImageRequest(_ *gin.Context, request *model.ImageRequest) (any, error) {
 	return nil, errors.New("not implemented")
 }
 
diff --git a/relay/adaptor/cloudflare/main.go b/relay/adaptor/cloudflare/main.go
index 980a2891..e164d473 100644
--- a/relay/adaptor/cloudflare/main.go
+++ b/relay/adaptor/cloudflare/main.go
@@ -19,9 +19,8 @@ import (
 )
 
 func ConvertCompletionsRequest(textRequest model.GeneralOpenAIRequest) *Request {
-	p, _ := textRequest.Prompt.(string)
 	return &Request{
-		Prompt:      p,
+		Prompt:      textRequest.Prompt,
 		MaxTokens:   textRequest.MaxTokens,
 		Stream:      textRequest.Stream,
 		Temperature: textRequest.Temperature,
diff --git a/relay/adaptor/cohere/adaptor.go b/relay/adaptor/cohere/adaptor.go
index 6fdb1b04..dd90bd7b 100644
--- a/relay/adaptor/cohere/adaptor.go
+++ b/relay/adaptor/cohere/adaptor.go
@@ -15,7 +15,7 @@ import (
 type Adaptor struct{}
 
 // ConvertImageRequest implements adaptor.Adaptor.
-func (*Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+func (*Adaptor) ConvertImageRequest(_ *gin.Context, request *model.ImageRequest) (any, error) {
 	return nil, errors.New("not implemented")
 }
 
diff --git a/relay/adaptor/coze/adaptor.go b/relay/adaptor/coze/adaptor.go
index 44f560e8..21d91e76 100644
--- a/relay/adaptor/coze/adaptor.go
+++ b/relay/adaptor/coze/adaptor.go
@@ -38,7 +38,7 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	return ConvertRequest(*request), nil
 }
 
-func (a *Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+func (a *Adaptor) ConvertImageRequest(_ *gin.Context, request *model.ImageRequest) (any, error) {
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
diff --git a/relay/adaptor/deepl/adaptor.go b/relay/adaptor/deepl/adaptor.go
index d018a096..5a03c261 100644
--- a/relay/adaptor/deepl/adaptor.go
+++ b/relay/adaptor/deepl/adaptor.go
@@ -39,7 +39,7 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	return convertedRequest, nil
 }
 
-func (a *Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+func (a *Adaptor) ConvertImageRequest(_ *gin.Context, request *model.ImageRequest) (any, error) {
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
diff --git a/relay/adaptor/gemini/adaptor.go b/relay/adaptor/gemini/adaptor.go
index 84083f60..a1b3f2fa 100644
--- a/relay/adaptor/gemini/adaptor.go
+++ b/relay/adaptor/gemini/adaptor.go
@@ -25,7 +25,7 @@ func (a *Adaptor) Init(meta *meta.Meta) {
 
 func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
 	defaultVersion := config.GeminiVersion
-	if strings.Contains(meta.ActualModelName, "gemini-2.0") ||
+	if strings.Contains(meta.ActualModelName, "gemini-2") ||
 		strings.Contains(meta.ActualModelName, "gemini-1.5") {
 		defaultVersion = "v1beta"
 	}
@@ -66,7 +66,7 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	}
 }
 
-func (a *Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+func (a *Adaptor) ConvertImageRequest(_ *gin.Context, request *model.ImageRequest) (any, error) {
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
diff --git a/relay/adaptor/gemini/constants.go b/relay/adaptor/gemini/constants.go
index d220b25f..424cf637 100644
--- a/relay/adaptor/gemini/constants.go
+++ b/relay/adaptor/gemini/constants.go
@@ -19,6 +19,9 @@ var ModelsSupportSystemInstruction = []string{
 	// "gemini-1.5-pro-experimental",
 	"gemini-2.0-flash", "gemini-2.0-flash-exp",
 	"gemini-2.0-flash-thinking-exp-01-21",
+	"gemini-2.0-flash-lite",
+	// "gemini-2.0-flash-exp-image-generation",
+	"gemini-2.0-pro-exp-02-05",
 }
 
 // IsModelSupportSystemInstruction check if the model support system instruction.
diff --git a/relay/adaptor/gemini/main.go b/relay/adaptor/gemini/main.go
index 29637296..26788ae7 100644
--- a/relay/adaptor/gemini/main.go
+++ b/relay/adaptor/gemini/main.go
@@ -8,19 +8,18 @@ import (
 	"net/http"
 	"strings"
 
-	"github.com/songquanpeng/one-api/common/render"
-
+	"github.com/gin-gonic/gin"
 	"github.com/songquanpeng/one-api/common"
 	"github.com/songquanpeng/one-api/common/config"
 	"github.com/songquanpeng/one-api/common/helper"
 	"github.com/songquanpeng/one-api/common/image"
 	"github.com/songquanpeng/one-api/common/logger"
 	"github.com/songquanpeng/one-api/common/random"
+	"github.com/songquanpeng/one-api/common/render"
+	"github.com/songquanpeng/one-api/relay/adaptor/geminiv2"
 	"github.com/songquanpeng/one-api/relay/adaptor/openai"
 	"github.com/songquanpeng/one-api/relay/constant"
 	"github.com/songquanpeng/one-api/relay/model"
-
-	"github.com/gin-gonic/gin"
 )
 
 // https://ai.google.dev/docs/gemini_api_overview?hl=zh-cn
@@ -61,9 +60,10 @@ func ConvertRequest(textRequest model.GeneralOpenAIRequest) *ChatRequest {
 			},
 		},
 		GenerationConfig: ChatGenerationConfig{
-			Temperature:     textRequest.Temperature,
-			TopP:            textRequest.TopP,
-			MaxOutputTokens: textRequest.MaxTokens,
+			Temperature:        textRequest.Temperature,
+			TopP:               textRequest.TopP,
+			MaxOutputTokens:    textRequest.MaxTokens,
+			ResponseModalities: geminiv2.GetModelModalities(textRequest.Model),
 		},
 	}
 	if textRequest.ResponseFormat != nil {
@@ -106,9 +106,9 @@ func ConvertRequest(textRequest model.GeneralOpenAIRequest) *ChatRequest {
 		var parts []Part
 		imageNum := 0
 		for _, part := range openaiContent {
-			if part.Type == model.ContentTypeText {
+			if part.Type == model.ContentTypeText && part.Text != nil && *part.Text != "" {
 				parts = append(parts, Part{
-					Text: part.Text,
+					Text: *part.Text,
 				})
 			} else if part.Type == model.ContentTypeImageURL {
 				imageNum += 1
@@ -258,19 +258,52 @@ func responseGeminiChat2OpenAI(response *ChatResponse) *openai.TextResponse {
 			if candidate.Content.Parts[0].FunctionCall != nil {
 				choice.Message.ToolCalls = getToolCalls(&candidate)
 			} else {
+				// Handle text and image content
 				var builder strings.Builder
+				var contentItems []model.MessageContent
+
 				for _, part := range candidate.Content.Parts {
-					if i > 0 {
-						builder.WriteString("\n")
+					if part.Text != "" {
+						// For text parts
+						if i > 0 {
+							builder.WriteString("\n")
+						}
+						builder.WriteString(part.Text)
+
+						// Add to content items
+						contentItems = append(contentItems, model.MessageContent{
+							Type: model.ContentTypeText,
+							Text: &part.Text,
+						})
+					}
+
+					if part.InlineData != nil && part.InlineData.MimeType != "" && part.InlineData.Data != "" {
+						// For inline image data
+						imageURL := &model.ImageURL{
+							// The data is already base64 encoded
+							Url: fmt.Sprintf("data:%s;base64,%s", part.InlineData.MimeType, part.InlineData.Data),
+						}
+
+						contentItems = append(contentItems, model.MessageContent{
+							Type:     model.ContentTypeImageURL,
+							ImageURL: imageURL,
+						})
 					}
-					builder.WriteString(part.Text)
 				}
-				choice.Message.Content = builder.String()
+
+				// If we have multiple content types, use structured content format
+				if len(contentItems) > 1 || (len(contentItems) == 1 && contentItems[0].Type != model.ContentTypeText) {
+					choice.Message.Content = contentItems
+				} else {
+					// Otherwise use the simple string content format
+					choice.Message.Content = builder.String()
+				}
 			}
 		} else {
 			choice.Message.Content = ""
 			choice.FinishReason = candidate.FinishReason
 		}
+
 		fullTextResponse.Choices = append(fullTextResponse.Choices, choice)
 	}
 	return &fullTextResponse
@@ -278,14 +311,78 @@ func responseGeminiChat2OpenAI(response *ChatResponse) *openai.TextResponse {
 
 func streamResponseGeminiChat2OpenAI(geminiResponse *ChatResponse) *openai.ChatCompletionsStreamResponse {
 	var choice openai.ChatCompletionsStreamResponseChoice
-	choice.Delta.Content = geminiResponse.GetResponseText()
-	//choice.FinishReason = &constant.StopFinishReason
+	choice.Delta.Role = "assistant"
+
+	// Check if we have any candidates
+	if len(geminiResponse.Candidates) == 0 {
+		return nil
+	}
+
+	// Get the first candidate
+	candidate := geminiResponse.Candidates[0]
+
+	// Check if there are parts in the content
+	if len(candidate.Content.Parts) == 0 {
+		return nil
+	}
+
+	// Handle different content types in the parts
+	for _, part := range candidate.Content.Parts {
+		// Handle text content
+		if part.Text != "" {
+			// Store as string for simple text responses
+			textContent := part.Text
+			choice.Delta.Content = textContent
+		}
+
+		// Handle image content
+		if part.InlineData != nil && part.InlineData.MimeType != "" && part.InlineData.Data != "" {
+			// Create a structured response for image content
+			imageUrl := fmt.Sprintf("data:%s;base64,%s", part.InlineData.MimeType, part.InlineData.Data)
+
+			// If we already have text content, create a mixed content response
+			if strContent, ok := choice.Delta.Content.(string); ok && strContent != "" {
+				// Convert the existing text content and add the image
+				messageContents := []model.MessageContent{
+					{
+						Type: model.ContentTypeText,
+						Text: &strContent,
+					},
+					{
+						Type: model.ContentTypeImageURL,
+						ImageURL: &model.ImageURL{
+							Url: imageUrl,
+						},
+					},
+				}
+				choice.Delta.Content = messageContents
+			} else {
+				// Only have image content
+				choice.Delta.Content = []model.MessageContent{
+					{
+						Type: model.ContentTypeImageURL,
+						ImageURL: &model.ImageURL{
+							Url: imageUrl,
+						},
+					},
+				}
+			}
+		}
+
+		// Handle function calls (if present)
+		if part.FunctionCall != nil {
+			choice.Delta.ToolCalls = getToolCalls(&candidate)
+		}
+	}
+
+	// Create response
 	var response openai.ChatCompletionsStreamResponse
 	response.Id = fmt.Sprintf("chatcmpl-%s", random.GetUUID())
 	response.Created = helper.GetTimestamp()
 	response.Object = "chat.completion.chunk"
 	response.Model = "gemini"
 	response.Choices = []openai.ChatCompletionsStreamResponseChoice{choice}
+
 	return &response
 }
 
@@ -311,17 +408,23 @@ func StreamHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusC
 	scanner := bufio.NewScanner(resp.Body)
 	scanner.Split(bufio.ScanLines)
 
+	buffer := make([]byte, 1024*1024) // 1MB buffer
+	scanner.Buffer(buffer, len(buffer))
+
 	common.SetEventStreamHeaders(c)
 
 	for scanner.Scan() {
 		data := scanner.Text()
 		data = strings.TrimSpace(data)
+
 		if !strings.HasPrefix(data, "data: ") {
 			continue
 		}
 		data = strings.TrimPrefix(data, "data: ")
 		data = strings.TrimSuffix(data, "\"")
 
+		fmt.Printf(">> gemini response: %s\n", data)
+
 		var geminiResponse ChatResponse
 		err := json.Unmarshal([]byte(data), &geminiResponse)
 		if err != nil {
@@ -361,6 +464,7 @@ func Handler(c *gin.Context, resp *http.Response, promptTokens int, modelName st
 	if err != nil {
 		return openai.ErrorWrapper(err, "read_response_body_failed", http.StatusInternalServerError), nil
 	}
+
 	err = resp.Body.Close()
 	if err != nil {
 		return openai.ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), nil
diff --git a/relay/adaptor/gemini/model.go b/relay/adaptor/gemini/model.go
index c3acae60..3bf5fe8e 100644
--- a/relay/adaptor/gemini/model.go
+++ b/relay/adaptor/gemini/model.go
@@ -6,6 +6,19 @@ type ChatRequest struct {
 	GenerationConfig  ChatGenerationConfig `json:"generation_config,omitempty"`
 	Tools             []ChatTools          `json:"tools,omitempty"`
 	SystemInstruction *ChatContent         `json:"system_instruction,omitempty"`
+	ModelVersion      string               `json:"model_version,omitempty"`
+	UsageMetadata     *UsageMetadata       `json:"usage_metadata,omitempty"`
+}
+
+type UsageMetadata struct {
+	PromptTokenCount    int                   `json:"promptTokenCount,omitempty"`
+	TotalTokenCount     int                   `json:"totalTokenCount,omitempty"`
+	PromptTokensDetails []PromptTokensDetails `json:"promptTokensDetails,omitempty"`
+}
+
+type PromptTokensDetails struct {
+	Modality   string `json:"modality,omitempty"`
+	TokenCount int    `json:"tokenCount,omitempty"`
 }
 
 type EmbeddingRequest struct {
@@ -66,12 +79,13 @@ type ChatTools struct {
 }
 
 type ChatGenerationConfig struct {
-	ResponseMimeType string   `json:"responseMimeType,omitempty"`
-	ResponseSchema   any      `json:"responseSchema,omitempty"`
-	Temperature      *float64 `json:"temperature,omitempty"`
-	TopP             *float64 `json:"topP,omitempty"`
-	TopK             float64  `json:"topK,omitempty"`
-	MaxOutputTokens  int      `json:"maxOutputTokens,omitempty"`
-	CandidateCount   int      `json:"candidateCount,omitempty"`
-	StopSequences    []string `json:"stopSequences,omitempty"`
+	ResponseMimeType   string   `json:"responseMimeType,omitempty"`
+	ResponseSchema     any      `json:"responseSchema,omitempty"`
+	Temperature        *float64 `json:"temperature,omitempty"`
+	TopP               *float64 `json:"topP,omitempty"`
+	TopK               float64  `json:"topK,omitempty"`
+	MaxOutputTokens    int      `json:"maxOutputTokens,omitempty"`
+	CandidateCount     int      `json:"candidateCount,omitempty"`
+	StopSequences      []string `json:"stopSequences,omitempty"`
+	ResponseModalities []string `json:"responseModalities,omitempty"`
 }
diff --git a/relay/adaptor/geminiv2/constants.go b/relay/adaptor/geminiv2/constants.go
index 73e7ad7d..e2712ddd 100644
--- a/relay/adaptor/geminiv2/constants.go
+++ b/relay/adaptor/geminiv2/constants.go
@@ -1,15 +1,42 @@
 package geminiv2
 
+import "strings"
+
 // https://ai.google.dev/models/gemini
 
 var ModelList = []string{
 	"gemini-pro", "gemini-1.0-pro",
-	// "gemma-2-2b-it", "gemma-2-9b-it", "gemma-2-27b-it",
+	"gemma-2-2b-it", "gemma-2-9b-it", "gemma-2-27b-it",
+	"gemma-3-27b-it",
 	"gemini-1.5-flash", "gemini-1.5-flash-8b",
 	"gemini-1.5-pro", "gemini-1.5-pro-experimental",
 	"text-embedding-004", "aqa",
 	"gemini-2.0-flash", "gemini-2.0-flash-exp",
 	"gemini-2.0-flash-lite-preview-02-05",
 	"gemini-2.0-flash-thinking-exp-01-21",
+	"gemini-2.0-flash-exp-image-generation",
 	"gemini-2.0-pro-exp-02-05",
+	"gemini-2.5-pro-exp-03-25",
+}
+
+const (
+	ModalityText  = "TEXT"
+	ModalityImage = "IMAGE"
+)
+
+// GetModelModalities returns the modalities of the model.
+func GetModelModalities(model string) []string {
+	if strings.Contains(model, "-image-generation") {
+		return []string{ModalityText, ModalityImage}
+	}
+
+	// Until 2025-03-26, the following models do not accept the responseModalities field
+	if model == "gemini-2.5-pro-exp-03-25" ||
+		model == "aqa" ||
+		strings.HasPrefix(model, "gemma") ||
+		strings.HasPrefix(model, "text-embed") {
+		return nil
+	}
+
+	return []string{ModalityText}
 }
diff --git a/relay/adaptor/groq/constants.go b/relay/adaptor/groq/constants.go
index 2a26b28b..0b319bc4 100644
--- a/relay/adaptor/groq/constants.go
+++ b/relay/adaptor/groq/constants.go
@@ -1,27 +1,32 @@
 package groq
 
+// ModelList is a list of models that can be used with Groq.
+//
 // https://console.groq.com/docs/models
-
 var ModelList = []string{
+	// Regular Models
+	"distil-whisper-large-v3-en",
 	"gemma2-9b-it",
-	"llama-3.1-70b-versatile",
+	"llama-3.3-70b-versatile",
 	"llama-3.1-8b-instant",
-	"llama-3.2-11b-text-preview",
-	"llama-3.2-11b-vision-preview",
-	"llama-3.2-1b-preview",
-	"llama-3.2-3b-preview",
-	"llama-3.2-90b-text-preview",
-	"llama-3.2-90b-vision-preview",
 	"llama-guard-3-8b",
 	"llama3-70b-8192",
 	"llama3-8b-8192",
-	"llama3-groq-70b-8192-tool-use-preview",
-	"llama3-groq-8b-8192-tool-use-preview",
-	"llava-v1.5-7b-4096-preview",
 	"mixtral-8x7b-32768",
-	"distil-whisper-large-v3-en",
 	"whisper-large-v3",
 	"whisper-large-v3-turbo",
+
+	// Preview Models
+	"qwen-qwq-32b",
+	"mistral-saba-24b",
+	"qwen-2.5-coder-32b",
+	"qwen-2.5-32b",
+	"deepseek-r1-distill-qwen-32b",
 	"deepseek-r1-distill-llama-70b-specdec",
 	"deepseek-r1-distill-llama-70b",
+	"llama-3.2-1b-preview",
+	"llama-3.2-3b-preview",
+	"llama-3.2-11b-vision-preview",
+	"llama-3.2-90b-vision-preview",
+	"llama-3.3-70b-specdec",
 }
diff --git a/relay/adaptor/interface.go b/relay/adaptor/interface.go
index 01b2e2cb..88667561 100644
--- a/relay/adaptor/interface.go
+++ b/relay/adaptor/interface.go
@@ -13,7 +13,7 @@ type Adaptor interface {
 	GetRequestURL(meta *meta.Meta) (string, error)
 	SetupRequestHeader(c *gin.Context, req *http.Request, meta *meta.Meta) error
 	ConvertRequest(c *gin.Context, relayMode int, request *model.GeneralOpenAIRequest) (any, error)
-	ConvertImageRequest(request *model.ImageRequest) (any, error)
+	ConvertImageRequest(c *gin.Context, request *model.ImageRequest) (any, error)
 	DoRequest(c *gin.Context, meta *meta.Meta, requestBody io.Reader) (*http.Response, error)
 	DoResponse(c *gin.Context, resp *http.Response, meta *meta.Meta) (usage *model.Usage, err *model.ErrorWithStatusCode)
 	GetModelList() []string
diff --git a/relay/adaptor/ollama/adaptor.go b/relay/adaptor/ollama/adaptor.go
index ad1f8983..9305340d 100644
--- a/relay/adaptor/ollama/adaptor.go
+++ b/relay/adaptor/ollama/adaptor.go
@@ -48,7 +48,7 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	}
 }
 
-func (a *Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+func (a *Adaptor) ConvertImageRequest(_ *gin.Context, request *model.ImageRequest) (any, error) {
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
diff --git a/relay/adaptor/ollama/main.go b/relay/adaptor/ollama/main.go
index fa1b05f0..2bb49767 100644
--- a/relay/adaptor/ollama/main.go
+++ b/relay/adaptor/ollama/main.go
@@ -43,7 +43,9 @@ func ConvertRequest(request model.GeneralOpenAIRequest) *ChatRequest {
 		for _, part := range openaiContent {
 			switch part.Type {
 			case model.ContentTypeText:
-				contentText = part.Text
+				if part.Text != nil {
+					contentText = *part.Text
+				}
 			case model.ContentTypeImageURL:
 				_, data, _ := image.GetImageFromUrl(part.ImageURL.Url)
 				imageUrls = append(imageUrls, data)
diff --git a/relay/adaptor/openai/adaptor.go b/relay/adaptor/openai/adaptor.go
index 8faf90a5..184b6736 100644
--- a/relay/adaptor/openai/adaptor.go
+++ b/relay/adaptor/openai/adaptor.go
@@ -1,14 +1,18 @@
 package openai
 
 import (
-	"errors"
 	"fmt"
 	"io"
+	"math"
 	"net/http"
 	"strings"
 
 	"github.com/gin-gonic/gin"
+	"github.com/pkg/errors"
 
+	"github.com/songquanpeng/one-api/common/config"
+	"github.com/songquanpeng/one-api/common/ctxkey"
+	"github.com/songquanpeng/one-api/common/logger"
 	"github.com/songquanpeng/one-api/relay/adaptor"
 	"github.com/songquanpeng/one-api/relay/adaptor/alibailian"
 	"github.com/songquanpeng/one-api/relay/adaptor/baiduv2"
@@ -16,6 +20,8 @@ import (
 	"github.com/songquanpeng/one-api/relay/adaptor/geminiv2"
 	"github.com/songquanpeng/one-api/relay/adaptor/minimax"
 	"github.com/songquanpeng/one-api/relay/adaptor/novita"
+	"github.com/songquanpeng/one-api/relay/adaptor/openrouter"
+	"github.com/songquanpeng/one-api/relay/billing/ratio"
 	"github.com/songquanpeng/one-api/relay/channeltype"
 	"github.com/songquanpeng/one-api/relay/meta"
 	"github.com/songquanpeng/one-api/relay/model"
@@ -33,16 +39,24 @@ func (a *Adaptor) Init(meta *meta.Meta) {
 func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
 	switch meta.ChannelType {
 	case channeltype.Azure:
+		defaultVersion := meta.Config.APIVersion
+
+		// https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/reasoning?tabs=python#api--feature-support
+		if strings.HasPrefix(meta.ActualModelName, "o1") ||
+			strings.HasPrefix(meta.ActualModelName, "o3") {
+			defaultVersion = "2024-12-01-preview"
+		}
+
 		if meta.Mode == relaymode.ImagesGenerations {
 			// https://learn.microsoft.com/en-us/azure/ai-services/openai/dall-e-quickstart?tabs=dalle3%2Ccommand-line&pivots=rest-api
 			// https://{resource_name}.openai.azure.com/openai/deployments/dall-e-3/images/generations?api-version=2024-03-01-preview
-			fullRequestURL := fmt.Sprintf("%s/openai/deployments/%s/images/generations?api-version=%s", meta.BaseURL, meta.ActualModelName, meta.Config.APIVersion)
+			fullRequestURL := fmt.Sprintf("%s/openai/deployments/%s/images/generations?api-version=%s", meta.BaseURL, meta.ActualModelName, defaultVersion)
 			return fullRequestURL, nil
 		}
 
 		// https://learn.microsoft.com/en-us/azure/cognitive-services/openai/chatgpt-quickstart?pivots=rest-api&tabs=command-line#rest-api
 		requestURL := strings.Split(meta.RequestURLPath, "?")[0]
-		requestURL = fmt.Sprintf("%s?api-version=%s", requestURL, meta.Config.APIVersion)
+		requestURL = fmt.Sprintf("%s?api-version=%s", requestURL, defaultVersion)
 		task := strings.TrimPrefix(requestURL, "/v1/")
 		model_ := meta.ActualModelName
 		model_ = strings.Replace(model_, ".", "", -1)
@@ -85,28 +99,92 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
-	if request.Stream {
+
+	meta := meta.GetByContext(c)
+	switch meta.ChannelType {
+	case channeltype.OpenRouter:
+		includeReasoning := true
+		request.IncludeReasoning = &includeReasoning
+		if request.Provider == nil || request.Provider.Sort == "" &&
+			config.OpenrouterProviderSort != "" {
+			if request.Provider == nil {
+				request.Provider = &openrouter.RequestProvider{}
+			}
+
+			request.Provider.Sort = config.OpenrouterProviderSort
+		}
+	default:
+	}
+
+	if request.Stream && !config.EnforceIncludeUsage {
+		logger.Warn(c.Request.Context(),
+			"please set ENFORCE_INCLUDE_USAGE=true to ensure accurate billing in stream mode")
+	}
+
+	if config.EnforceIncludeUsage && request.Stream {
 		// always return usage in stream mode
 		if request.StreamOptions == nil {
 			request.StreamOptions = &model.StreamOptions{}
 		}
 		request.StreamOptions.IncludeUsage = true
 	}
+
+	// o1/o1-mini/o1-preview do not support system prompt/max_tokens/temperature
+	if strings.HasPrefix(meta.ActualModelName, "o1") ||
+		strings.HasPrefix(meta.ActualModelName, "o3") {
+		temperature := float64(1)
+		request.Temperature = &temperature // Only the default (1) value is supported
+
+		request.MaxTokens = 0
+		request.Messages = func(raw []model.Message) (filtered []model.Message) {
+			for i := range raw {
+				if raw[i].Role != "system" {
+					filtered = append(filtered, raw[i])
+				}
+			}
+
+			return
+		}(request.Messages)
+	}
+
+	// web search do not support system prompt/max_tokens/temperature
+	if strings.HasPrefix(meta.ActualModelName, "gpt-4o-search") ||
+		strings.HasPrefix(meta.ActualModelName, "gpt-4o-mini-search") {
+		request.Temperature = nil
+		request.TopP = nil
+		request.PresencePenalty = nil
+		request.N = nil
+		request.FrequencyPenalty = nil
+	}
+
+	if request.Stream && !config.EnforceIncludeUsage &&
+		(strings.HasPrefix(request.Model, "gpt-4o-audio") ||
+			strings.HasPrefix(request.Model, "gpt-4o-mini-audio")) {
+		// TODO: Since it is not clear how to implement billing in stream mode,
+		// it is temporarily not supported
+		return nil, errors.New("set ENFORCE_INCLUDE_USAGE=true to enable stream mode for gpt-4o-audio")
+	}
+
 	return request, nil
 }
 
-func (a *Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+func (a *Adaptor) ConvertImageRequest(_ *gin.Context, request *model.ImageRequest) (any, error) {
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
 	return request, nil
 }
 
-func (a *Adaptor) DoRequest(c *gin.Context, meta *meta.Meta, requestBody io.Reader) (*http.Response, error) {
+func (a *Adaptor) DoRequest(c *gin.Context,
+	meta *meta.Meta,
+	requestBody io.Reader) (*http.Response, error) {
 	return adaptor.DoRequestHelper(a, c, meta, requestBody)
 }
 
-func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *meta.Meta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
+func (a *Adaptor) DoResponse(c *gin.Context,
+	resp *http.Response,
+	meta *meta.Meta) (usage *model.Usage,
+	err *model.ErrorWithStatusCode) {
 	if meta.IsStream {
 		var responseText string
 		err, responseText, usage = StreamHandler(c, resp, meta.Mode)
@@ -121,10 +199,61 @@ func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *meta.Met
 		switch meta.Mode {
 		case relaymode.ImagesGenerations:
 			err, _ = ImageHandler(c, resp)
+		case relaymode.ImagesEdits:
+			err, _ = ImagesEditsHandler(c, resp)
 		default:
 			err, usage = Handler(c, resp, meta.PromptTokens, meta.ActualModelName)
 		}
 	}
+
+	// -------------------------------------
+	// calculate web-search tool cost
+	// -------------------------------------
+	if usage != nil {
+		searchContextSize := "medium"
+		var req *model.GeneralOpenAIRequest
+		if vi, ok := c.Get(ctxkey.ConvertedRequest); ok {
+			if req, ok = vi.(*model.GeneralOpenAIRequest); ok {
+				if req != nil &&
+					req.WebSearchOptions != nil &&
+					req.WebSearchOptions.SearchContextSize != nil {
+					searchContextSize = *req.WebSearchOptions.SearchContextSize
+				}
+
+				switch {
+				case strings.HasPrefix(meta.ActualModelName, "gpt-4o-search"):
+					switch searchContextSize {
+					case "low":
+						usage.ToolsCost += int64(math.Ceil(30 / 1000 * ratio.QuotaPerUsd))
+					case "medium":
+						usage.ToolsCost += int64(math.Ceil(35 / 1000 * ratio.QuotaPerUsd))
+					case "high":
+						usage.ToolsCost += int64(math.Ceil(40 / 1000 * ratio.QuotaPerUsd))
+					default:
+						return nil, ErrorWrapper(
+							errors.Errorf("invalid search context size %q", searchContextSize),
+							"invalid search context size: "+searchContextSize,
+							http.StatusBadRequest)
+					}
+				case strings.HasPrefix(meta.ActualModelName, "gpt-4o-mini-search"):
+					switch searchContextSize {
+					case "low":
+						usage.ToolsCost += int64(math.Ceil(25 / 1000 * ratio.QuotaPerUsd))
+					case "medium":
+						usage.ToolsCost += int64(math.Ceil(27.5 / 1000 * ratio.QuotaPerUsd))
+					case "high":
+						usage.ToolsCost += int64(math.Ceil(30 / 1000 * ratio.QuotaPerUsd))
+					default:
+						return nil, ErrorWrapper(
+							errors.Errorf("invalid search context size %q", searchContextSize),
+							"invalid search context size: "+searchContextSize,
+							http.StatusBadRequest)
+					}
+				}
+			}
+		}
+	}
+
 	return
 }
 
diff --git a/relay/adaptor/openai/constants.go b/relay/adaptor/openai/constants.go
index 8a643bc6..f7518894 100644
--- a/relay/adaptor/openai/constants.go
+++ b/relay/adaptor/openai/constants.go
@@ -7,11 +7,10 @@ var ModelList = []string{
 	"gpt-4", "gpt-4-0314", "gpt-4-0613", "gpt-4-1106-preview", "gpt-4-0125-preview",
 	"gpt-4-32k", "gpt-4-32k-0314", "gpt-4-32k-0613",
 	"gpt-4-turbo-preview", "gpt-4-turbo", "gpt-4-turbo-2024-04-09",
-	"gpt-4o", "gpt-4o-2024-05-13",
-	"gpt-4o-2024-08-06",
-	"gpt-4o-2024-11-20",
-	"chatgpt-4o-latest",
+	"gpt-4o", "gpt-4o-2024-05-13", "gpt-4o-2024-08-06", "gpt-4o-2024-11-20", "chatgpt-4o-latest",
 	"gpt-4o-mini", "gpt-4o-mini-2024-07-18",
+	"gpt-4o-mini-audio-preview", "gpt-4o-mini-audio-preview-2024-12-17",
+	"gpt-4o-audio-preview", "gpt-4o-audio-preview-2024-12-17", "gpt-4o-audio-preview-2024-10-01",
 	"gpt-4-vision-preview",
 	"text-embedding-ada-002", "text-embedding-3-small", "text-embedding-3-large",
 	"text-curie-001", "text-babbage-001", "text-ada-001", "text-davinci-002", "text-davinci-003",
@@ -24,4 +23,8 @@ var ModelList = []string{
 	"o1", "o1-2024-12-17",
 	"o1-preview", "o1-preview-2024-09-12",
 	"o1-mini", "o1-mini-2024-09-12",
+	"o3-mini", "o3-mini-2025-01-31",
+	"gpt-4.5-preview", "gpt-4.5-preview-2025-02-27",
+	// https://platform.openai.com/docs/guides/tools-web-search?api-mode=chat
+	"gpt-4o-search-preview", "gpt-4o-mini-search-preview",
 }
diff --git a/relay/adaptor/openai/image.go b/relay/adaptor/openai/image.go
index 0f89618a..433d9421 100644
--- a/relay/adaptor/openai/image.go
+++ b/relay/adaptor/openai/image.go
@@ -3,12 +3,30 @@ package openai
 import (
 	"bytes"
 	"encoding/json"
-	"github.com/gin-gonic/gin"
-	"github.com/songquanpeng/one-api/relay/model"
 	"io"
 	"net/http"
+
+	"github.com/gin-gonic/gin"
+	"github.com/songquanpeng/one-api/relay/model"
 )
 
+// ImagesEditsHandler just copy response body to client
+//
+// https://platform.openai.com/docs/api-reference/images/createEdit
+func ImagesEditsHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusCode, *model.Usage) {
+	c.Writer.WriteHeader(resp.StatusCode)
+	for k, v := range resp.Header {
+		c.Writer.Header().Set(k, v[0])
+	}
+
+	if _, err := io.Copy(c.Writer, resp.Body); err != nil {
+		return ErrorWrapper(err, "copy_response_body_failed", http.StatusInternalServerError), nil
+	}
+	defer resp.Body.Close()
+
+	return nil, nil
+}
+
 func ImageHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusCode, *model.Usage) {
 	var imageResponse ImageResponse
 	responseBody, err := io.ReadAll(resp.Body)
diff --git a/relay/adaptor/openai/main.go b/relay/adaptor/openai/main.go
index 97080738..1411980a 100644
--- a/relay/adaptor/openai/main.go
+++ b/relay/adaptor/openai/main.go
@@ -5,15 +5,16 @@ import (
 	"bytes"
 	"encoding/json"
 	"io"
+	"math"
 	"net/http"
 	"strings"
 
-	"github.com/songquanpeng/one-api/common/render"
-
 	"github.com/gin-gonic/gin"
 	"github.com/songquanpeng/one-api/common"
 	"github.com/songquanpeng/one-api/common/conv"
 	"github.com/songquanpeng/one-api/common/logger"
+	"github.com/songquanpeng/one-api/common/render"
+	"github.com/songquanpeng/one-api/relay/billing/ratio"
 	"github.com/songquanpeng/one-api/relay/model"
 	"github.com/songquanpeng/one-api/relay/relaymode"
 )
@@ -24,128 +25,300 @@ const (
 	dataPrefixLength = len(dataPrefix)
 )
 
+// StreamHandler processes streaming responses from OpenAI API
+// It handles incremental content delivery and accumulates the final response text
+// Returns error (if any), accumulated response text, and token usage information
 func StreamHandler(c *gin.Context, resp *http.Response, relayMode int) (*model.ErrorWithStatusCode, string, *model.Usage) {
+	// Initialize accumulators for the response
 	responseText := ""
-	scanner := bufio.NewScanner(resp.Body)
-	scanner.Split(bufio.ScanLines)
+	reasoningText := ""
 	var usage *model.Usage
 
+	// Set up scanner for reading the stream line by line
+	scanner := bufio.NewScanner(resp.Body)
+	buffer := make([]byte, 256*1024) // 256KB buffer for large messages
+	scanner.Buffer(buffer, len(buffer))
+	scanner.Split(bufio.ScanLines)
+
+	// Set response headers for SSE
 	common.SetEventStreamHeaders(c)
 
 	doneRendered := false
+
+	// Process each line from the stream
 	for scanner.Scan() {
-		data := scanner.Text()
-		if len(data) < dataPrefixLength { // ignore blank line or wrong format
-			continue
+		data := NormalizeDataLine(scanner.Text())
+
+		// logger.Debugf(c.Request.Context(), "stream response: %s", data)
+
+		// Skip lines that don't match expected format
+		if len(data) < dataPrefixLength {
+			continue // Ignore blank line or wrong format
 		}
+
+		// Verify line starts with expected prefix
 		if data[:dataPrefixLength] != dataPrefix && data[:dataPrefixLength] != done {
 			continue
 		}
+
+		// Check for stream termination
 		if strings.HasPrefix(data[dataPrefixLength:], done) {
 			render.StringData(c, data)
 			doneRendered = true
 			continue
 		}
+
+		// Process based on relay mode
 		switch relayMode {
 		case relaymode.ChatCompletions:
 			var streamResponse ChatCompletionsStreamResponse
+
+			// Parse the JSON response
 			err := json.Unmarshal([]byte(data[dataPrefixLength:]), &streamResponse)
 			if err != nil {
-				logger.SysError("error unmarshalling stream response: " + err.Error())
-				render.StringData(c, data) // if error happened, pass the data to client
-				continue                   // just ignore the error
+				logger.Errorf(c.Request.Context(), "unmarshalling stream data %q got %+v", data, err)
+				render.StringData(c, data) // Pass raw data to client if parsing fails
+				continue
 			}
+
+			// Skip empty choices (Azure specific behavior)
 			if len(streamResponse.Choices) == 0 && streamResponse.Usage == nil {
-				// but for empty choice and no usage, we should not pass it to client, this is for azure
-				continue // just ignore empty choice
+				continue
 			}
-			render.StringData(c, data)
+
+			// Process each choice in the response
 			for _, choice := range streamResponse.Choices {
+				// Extract reasoning content from different possible fields
+				currentReasoningChunk := extractReasoningContent(&choice.Delta)
+
+				// Update accumulated reasoning text
+				if currentReasoningChunk != "" {
+					reasoningText += currentReasoningChunk
+				}
+
+				// Set the reasoning content in the format requested by client
+				choice.Delta.SetReasoningContent(c.Query("reasoning_format"), currentReasoningChunk)
+
+				// Accumulate response content
 				responseText += conv.AsString(choice.Delta.Content)
 			}
+
+			// Send the processed data to the client
+			render.StringData(c, data)
+
+			// Update usage information if available
 			if streamResponse.Usage != nil {
 				usage = streamResponse.Usage
 			}
+
 		case relaymode.Completions:
+			// Send the data immediately for Completions mode
 			render.StringData(c, data)
+
 			var streamResponse CompletionsStreamResponse
 			err := json.Unmarshal([]byte(data[dataPrefixLength:]), &streamResponse)
 			if err != nil {
 				logger.SysError("error unmarshalling stream response: " + err.Error())
 				continue
 			}
+
+			// Accumulate text from all choices
 			for _, choice := range streamResponse.Choices {
 				responseText += choice.Text
 			}
 		}
 	}
 
+	// Check for scanner errors
 	if err := scanner.Err(); err != nil {
 		logger.SysError("error reading stream: " + err.Error())
 	}
 
+	// Ensure stream termination is sent to client
 	if !doneRendered {
 		render.Done(c)
 	}
 
-	err := resp.Body.Close()
-	if err != nil {
+	// Clean up resources
+	if err := resp.Body.Close(); err != nil {
 		return ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), "", nil
 	}
 
-	return nil, responseText, usage
+	// Return the complete response text (reasoning + content) and usage
+	return nil, reasoningText + responseText, usage
 }
 
+// Helper function to extract reasoning content from message delta
+func extractReasoningContent(delta *model.Message) string {
+	content := ""
+
+	// Extract reasoning from different possible fields
+	if delta.Reasoning != nil {
+		content += *delta.Reasoning
+		delta.Reasoning = nil
+	}
+
+	if delta.ReasoningContent != nil {
+		content += *delta.ReasoningContent
+		delta.ReasoningContent = nil
+	}
+
+	return content
+}
+
+// Handler processes non-streaming responses from OpenAI API
+// Returns error (if any) and token usage information
 func Handler(c *gin.Context, resp *http.Response, promptTokens int, modelName string) (*model.ErrorWithStatusCode, *model.Usage) {
-	var textResponse SlimTextResponse
+	// Read the entire response body
 	responseBody, err := io.ReadAll(resp.Body)
 	if err != nil {
 		return ErrorWrapper(err, "read_response_body_failed", http.StatusInternalServerError), nil
 	}
-	err = resp.Body.Close()
-	if err != nil {
+
+	// Close the original response body
+	if err = resp.Body.Close(); err != nil {
 		return ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), nil
 	}
-	err = json.Unmarshal(responseBody, &textResponse)
-	if err != nil {
+
+	// Parse the response JSON
+	var textResponse SlimTextResponse
+	if err = json.Unmarshal(responseBody, &textResponse); err != nil {
 		return ErrorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError), nil
 	}
+
+	// Check for API errors
 	if textResponse.Error.Type != "" {
 		return &model.ErrorWithStatusCode{
 			Error:      textResponse.Error,
 			StatusCode: resp.StatusCode,
 		}, nil
 	}
-	// Reset response body
-	resp.Body = io.NopCloser(bytes.NewBuffer(responseBody))
 
-	// We shouldn't set the header before we parse the response body, because the parse part may fail.
-	// And then we will have to send an error response, but in this case, the header has already been set.
-	// So the HTTPClient will be confused by the response.
-	// For example, Postman will report error, and we cannot check the response at all.
-	for k, v := range resp.Header {
-		c.Writer.Header().Set(k, v[0])
+	// Process reasoning content in each choice
+	for _, msg := range textResponse.Choices {
+		reasoningContent := processReasoningContent(&msg)
+
+		// Set reasoning in requested format if content exists
+		if reasoningContent != "" {
+			msg.SetReasoningContent(c.Query("reasoning_format"), reasoningContent)
+		}
 	}
+
+	// Reset response body for forwarding to client
+	resp.Body = io.NopCloser(bytes.NewBuffer(responseBody))
+	logger.Debugf(c.Request.Context(), "handler response: %s", string(responseBody))
+
+	// Forward all response headers (not just first value of each)
+	for k, values := range resp.Header {
+		for _, v := range values {
+			c.Writer.Header().Add(k, v)
+		}
+	}
+
+	// Set response status and copy body to client
 	c.Writer.WriteHeader(resp.StatusCode)
-	_, err = io.Copy(c.Writer, resp.Body)
-	if err != nil {
+	if _, err = io.Copy(c.Writer, resp.Body); err != nil {
 		return ErrorWrapper(err, "copy_response_body_failed", http.StatusInternalServerError), nil
 	}
-	err = resp.Body.Close()
-	if err != nil {
+
+	// Close the reset body
+	if err = resp.Body.Close(); err != nil {
 		return ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), nil
 	}
 
-	if textResponse.Usage.TotalTokens == 0 || (textResponse.Usage.PromptTokens == 0 && textResponse.Usage.CompletionTokens == 0) {
+	// Calculate token usage if not provided by API
+	calculateTokenUsage(&textResponse, promptTokens, modelName)
+
+	return nil, &textResponse.Usage
+}
+
+// processReasoningContent is a helper function to extract and process reasoning content from the message
+func processReasoningContent(msg *TextResponseChoice) string {
+	var reasoningContent string
+
+	// Check different locations for reasoning content
+	switch {
+	case msg.Reasoning != nil:
+		reasoningContent = *msg.Reasoning
+		msg.Reasoning = nil
+	case msg.ReasoningContent != nil:
+		reasoningContent = *msg.ReasoningContent
+		msg.ReasoningContent = nil
+	case msg.Message.Reasoning != nil:
+		reasoningContent = *msg.Message.Reasoning
+		msg.Message.Reasoning = nil
+	case msg.Message.ReasoningContent != nil:
+		reasoningContent = *msg.Message.ReasoningContent
+		msg.Message.ReasoningContent = nil
+	}
+
+	return reasoningContent
+}
+
+// Helper function to calculate token usage
+func calculateTokenUsage(response *SlimTextResponse, promptTokens int, modelName string) {
+	// Calculate tokens if not provided by the API
+	if response.Usage.TotalTokens == 0 ||
+		(response.Usage.PromptTokens == 0 && response.Usage.CompletionTokens == 0) {
+
 		completionTokens := 0
-		for _, choice := range textResponse.Choices {
+		for _, choice := range response.Choices {
+			// Count content tokens
 			completionTokens += CountTokenText(choice.Message.StringContent(), modelName)
+
+			// Count reasoning tokens in all possible locations
+			if choice.Message.Reasoning != nil {
+				completionTokens += CountToken(*choice.Message.Reasoning)
+			}
+			if choice.Message.ReasoningContent != nil {
+				completionTokens += CountToken(*choice.Message.ReasoningContent)
+			}
+			if choice.Reasoning != nil {
+				completionTokens += CountToken(*choice.Reasoning)
+			}
+			if choice.ReasoningContent != nil {
+				completionTokens += CountToken(*choice.ReasoningContent)
+			}
 		}
-		textResponse.Usage = model.Usage{
+
+		// Set usage values
+		response.Usage = model.Usage{
 			PromptTokens:     promptTokens,
 			CompletionTokens: completionTokens,
 			TotalTokens:      promptTokens + completionTokens,
 		}
+	} else if hasAudioTokens(response) {
+		// Handle audio tokens conversion
+		calculateAudioTokens(response, modelName)
 	}
-	return nil, &textResponse.Usage
+}
+
+// Helper function to check if response has audio tokens
+func hasAudioTokens(response *SlimTextResponse) bool {
+	return (response.PromptTokensDetails != nil && response.PromptTokensDetails.AudioTokens > 0) ||
+		(response.CompletionTokensDetails != nil && response.CompletionTokensDetails.AudioTokens > 0)
+}
+
+// Helper function to calculate audio token usage
+func calculateAudioTokens(response *SlimTextResponse, modelName string) {
+	// Convert audio tokens for prompt
+	if response.PromptTokensDetails != nil {
+		response.Usage.PromptTokens = response.PromptTokensDetails.TextTokens +
+			int(math.Ceil(
+				float64(response.PromptTokensDetails.AudioTokens)*
+					ratio.GetAudioPromptRatio(modelName),
+			))
+	}
+
+	// Convert audio tokens for completion
+	if response.CompletionTokensDetails != nil {
+		response.Usage.CompletionTokens = response.CompletionTokensDetails.TextTokens +
+			int(math.Ceil(
+				float64(response.CompletionTokensDetails.AudioTokens)*
+					ratio.GetAudioPromptRatio(modelName)*ratio.GetAudioCompletionRatio(modelName),
+			))
+	}
+
+	// Calculate total tokens
+	response.Usage.TotalTokens = response.Usage.PromptTokens + response.Usage.CompletionTokens
 }
diff --git a/relay/adaptor/openai/model.go b/relay/adaptor/openai/model.go
index 4c974de4..50a8e1da 100644
--- a/relay/adaptor/openai/model.go
+++ b/relay/adaptor/openai/model.go
@@ -1,6 +1,10 @@
 package openai
 
-import "github.com/songquanpeng/one-api/relay/model"
+import (
+	"mime/multipart"
+
+	"github.com/songquanpeng/one-api/relay/model"
+)
 
 type TextContent struct {
 	Type string `json:"type,omitempty"`
@@ -71,6 +75,24 @@ type TextToSpeechRequest struct {
 	ResponseFormat string  `json:"response_format"`
 }
 
+type AudioTranscriptionRequest struct {
+	File                 *multipart.FileHeader `form:"file" binding:"required"`
+	Model                string                `form:"model" binding:"required"`
+	Language             string                `form:"language"`
+	Prompt               string                `form:"prompt"`
+	ReponseFormat        string                `form:"response_format" binding:"oneof=json text srt verbose_json vtt"`
+	Temperature          float64               `form:"temperature"`
+	TimestampGranularity []string              `form:"timestamp_granularity"`
+}
+
+type AudioTranslationRequest struct {
+	File           *multipart.FileHeader `form:"file" binding:"required"`
+	Model          string                `form:"model" binding:"required"`
+	Prompt         string                `form:"prompt"`
+	ResponseFormat string                `form:"response_format" binding:"oneof=json text srt verbose_json vtt"`
+	Temperature    float64               `form:"temperature"`
+}
+
 type UsageOrResponseText struct {
 	*model.Usage
 	ResponseText string
@@ -110,12 +132,14 @@ type EmbeddingResponse struct {
 	model.Usage `json:"usage"`
 }
 
+// ImageData represents an image in the response
 type ImageData struct {
 	Url           string `json:"url,omitempty"`
 	B64Json       string `json:"b64_json,omitempty"`
 	RevisedPrompt string `json:"revised_prompt,omitempty"`
 }
 
+// ImageResponse represents the response structure for image generations
 type ImageResponse struct {
 	Created int64       `json:"created"`
 	Data    []ImageData `json:"data"`
diff --git a/relay/adaptor/openai/token.go b/relay/adaptor/openai/token.go
index b50220e7..e3194439 100644
--- a/relay/adaptor/openai/token.go
+++ b/relay/adaptor/openai/token.go
@@ -1,16 +1,20 @@
 package openai
 
 import (
-	"errors"
+	"bytes"
+	"context"
+	"encoding/base64"
 	"fmt"
 	"math"
 	"strings"
 
+	"github.com/pkg/errors"
 	"github.com/pkoukk/tiktoken-go"
-
 	"github.com/songquanpeng/one-api/common/config"
+	"github.com/songquanpeng/one-api/common/helper"
 	"github.com/songquanpeng/one-api/common/image"
 	"github.com/songquanpeng/one-api/common/logger"
+	"github.com/songquanpeng/one-api/relay/billing/ratio"
 	billingratio "github.com/songquanpeng/one-api/relay/billing/ratio"
 	"github.com/songquanpeng/one-api/relay/model"
 )
@@ -73,8 +77,10 @@ func getTokenNum(tokenEncoder *tiktoken.Tiktoken, text string) int {
 	return len(tokenEncoder.Encode(text, nil, nil))
 }
 
-func CountTokenMessages(messages []model.Message, model string) int {
-	tokenEncoder := getTokenEncoder(model)
+// CountTokenMessages counts the number of tokens in a list of messages.
+func CountTokenMessages(ctx context.Context,
+	messages []model.Message, actualModel string) int {
+	tokenEncoder := getTokenEncoder(actualModel)
 	// Reference:
 	// https://github.com/openai/openai-cookbook/blob/main/examples/How_to_count_tokens_with_tiktoken.ipynb
 	// https://github.com/pkoukk/tiktoken-go/issues/6
@@ -82,47 +88,54 @@ func CountTokenMessages(messages []model.Message, model string) int {
 	// Every message follows <|start|>{role/name}\n{content}<|end|>\n
 	var tokensPerMessage int
 	var tokensPerName int
-	if model == "gpt-3.5-turbo-0301" {
+	if actualModel == "gpt-3.5-turbo-0301" {
 		tokensPerMessage = 4
 		tokensPerName = -1 // If there's a name, the role is omitted
 	} else {
 		tokensPerMessage = 3
 		tokensPerName = 1
 	}
+
 	tokenNum := 0
+	var totalAudioTokens float64
 	for _, message := range messages {
 		tokenNum += tokensPerMessage
-		switch v := message.Content.(type) {
-		case string:
-			tokenNum += getTokenNum(tokenEncoder, v)
-		case []any:
-			for _, it := range v {
-				m := it.(map[string]any)
-				switch m["type"] {
-				case "text":
-					if textValue, ok := m["text"]; ok {
-						if textString, ok := textValue.(string); ok {
-							tokenNum += getTokenNum(tokenEncoder, textString)
-						}
-					}
-				case "image_url":
-					imageUrl, ok := m["image_url"].(map[string]any)
-					if ok {
-						url := imageUrl["url"].(string)
-						detail := ""
-						if imageUrl["detail"] != nil {
-							detail = imageUrl["detail"].(string)
-						}
-						imageTokens, err := countImageTokens(url, detail, model)
-						if err != nil {
-							logger.SysError("error counting image tokens: " + err.Error())
-						} else {
-							tokenNum += imageTokens
-						}
-					}
+		contents := message.ParseContent()
+		for _, content := range contents {
+			switch content.Type {
+			case model.ContentTypeText:
+				if content.Text != nil {
+					tokenNum += getTokenNum(tokenEncoder, *content.Text)
+				}
+			case model.ContentTypeImageURL:
+				imageTokens, err := countImageTokens(
+					content.ImageURL.Url,
+					content.ImageURL.Detail,
+					actualModel)
+				if err != nil {
+					logger.SysError("error counting image tokens: " + err.Error())
+				} else {
+					tokenNum += imageTokens
+				}
+			case model.ContentTypeInputAudio:
+				audioData, err := base64.StdEncoding.DecodeString(content.InputAudio.Data)
+				if err != nil {
+					logger.SysError("error decoding audio data: " + err.Error())
+				}
+
+				audioTokens, err := helper.GetAudioTokens(ctx,
+					bytes.NewReader(audioData),
+					ratio.GetAudioPromptTokensPerSecond(actualModel))
+				if err != nil {
+					logger.SysError("error counting audio tokens: " + err.Error())
+				} else {
+					totalAudioTokens += audioTokens
 				}
 			}
 		}
+
+		tokenNum += int(math.Ceil(totalAudioTokens))
+
 		tokenNum += getTokenNum(tokenEncoder, message.Role)
 		if message.Name != nil {
 			tokenNum += tokensPerName
diff --git a/relay/adaptor/openai/util.go b/relay/adaptor/openai/util.go
index 83beadba..ca5605c4 100644
--- a/relay/adaptor/openai/util.go
+++ b/relay/adaptor/openai/util.go
@@ -3,6 +3,7 @@ package openai
 import (
 	"context"
 	"fmt"
+	"strings"
 
 	"github.com/songquanpeng/one-api/common/logger"
 	"github.com/songquanpeng/one-api/relay/model"
@@ -21,3 +22,11 @@ func ErrorWrapper(err error, code string, statusCode int) *model.ErrorWithStatus
 		StatusCode: statusCode,
 	}
 }
+
+func NormalizeDataLine(data string) string {
+	if strings.HasPrefix(data, "data:") {
+		content := strings.TrimLeft(data[len("data:"):], " ")
+		return "data: " + content
+	}
+	return data
+}
diff --git a/relay/adaptor/openrouter/model.go b/relay/adaptor/openrouter/model.go
new file mode 100644
index 00000000..581bc2cc
--- /dev/null
+++ b/relay/adaptor/openrouter/model.go
@@ -0,0 +1,22 @@
+package openrouter
+
+// RequestProvider customize how your requests are routed using the provider object
+// in the request body for Chat Completions and Completions.
+//
+// https://openrouter.ai/docs/features/provider-routing
+type RequestProvider struct {
+	// Order is list of provider names to try in order (e.g. ["Anthropic", "OpenAI"]). Default: empty
+	Order []string `json:"order,omitempty"`
+	// AllowFallbacks is whether to allow backup providers when the primary is unavailable. Default: true
+	AllowFallbacks bool `json:"allow_fallbacks,omitempty"`
+	// RequireParameters is only use providers that support all parameters in your request. Default: false
+	RequireParameters bool `json:"require_parameters,omitempty"`
+	// DataCollection is control whether to use providers that may store data ("allow" or "deny"). Default: "allow"
+	DataCollection string `json:"data_collection,omitempty" binding:"omitempty,oneof=allow deny"`
+	// Ignore is list of provider names to skip for this request. Default: empty
+	Ignore []string `json:"ignore,omitempty"`
+	// Quantizations is list of quantization levels to filter by (e.g. ["int4", "int8"]). Default: empty
+	Quantizations []string `json:"quantizations,omitempty"`
+	// Sort is sort providers by price or throughput (e.g. "price" or "throughput"). Default: empty
+	Sort string `json:"sort,omitempty" binding:"omitempty,oneof=price throughput latency"`
+}
diff --git a/relay/adaptor/palm/adaptor.go b/relay/adaptor/palm/adaptor.go
index 98aa3e18..9b51562d 100644
--- a/relay/adaptor/palm/adaptor.go
+++ b/relay/adaptor/palm/adaptor.go
@@ -36,7 +36,7 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	return ConvertRequest(*request), nil
 }
 
-func (a *Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+func (a *Adaptor) ConvertImageRequest(_ *gin.Context, request *model.ImageRequest) (any, error) {
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
diff --git a/relay/adaptor/palm/palm.go b/relay/adaptor/palm/palm.go
index d31784ec..f3875417 100644
--- a/relay/adaptor/palm/palm.go
+++ b/relay/adaptor/palm/palm.go
@@ -25,11 +25,17 @@ func ConvertRequest(textRequest model.GeneralOpenAIRequest) *ChatRequest {
 		Prompt: Prompt{
 			Messages: make([]ChatMessage, 0, len(textRequest.Messages)),
 		},
-		Temperature:    textRequest.Temperature,
-		CandidateCount: textRequest.N,
-		TopP:           textRequest.TopP,
-		TopK:           textRequest.MaxTokens,
+		Temperature: textRequest.Temperature,
+		TopP:        textRequest.TopP,
+		TopK:        textRequest.MaxTokens,
 	}
+
+	if textRequest.N != nil {
+		palmRequest.CandidateCount = *textRequest.N
+	} else {
+		palmRequest.CandidateCount = 1
+	}
+
 	for _, message := range textRequest.Messages {
 		palmMessage := ChatMessage{
 			Content: message.StringContent(),
diff --git a/relay/adaptor/proxy/adaptor.go b/relay/adaptor/proxy/adaptor.go
index 670c7628..32984fc7 100644
--- a/relay/adaptor/proxy/adaptor.go
+++ b/relay/adaptor/proxy/adaptor.go
@@ -80,7 +80,7 @@ func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *me
 	return nil
 }
 
-func (a *Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+func (a *Adaptor) ConvertImageRequest(_ *gin.Context, request *model.ImageRequest) (any, error) {
 	return nil, errors.Errorf("not implement")
 }
 
diff --git a/relay/adaptor/replicate/adaptor.go b/relay/adaptor/replicate/adaptor.go
index a60a7de3..2b4c9af7 100644
--- a/relay/adaptor/replicate/adaptor.go
+++ b/relay/adaptor/replicate/adaptor.go
@@ -23,7 +23,7 @@ type Adaptor struct {
 }
 
 // ConvertImageRequest implements adaptor.Adaptor.
-func (*Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+func (*Adaptor) ConvertImageRequest(_ *gin.Context, request *model.ImageRequest) (any, error) {
 	return DrawImageRequest{
 		Input: ImageInput{
 			Steps:           25,
diff --git a/relay/adaptor/replicate/constant.go b/relay/adaptor/replicate/constant.go
index 989142c9..6ac06bd0 100644
--- a/relay/adaptor/replicate/constant.go
+++ b/relay/adaptor/replicate/constant.go
@@ -33,9 +33,16 @@ var ModelList = []string{
 	// -------------------------------------
 	// language model
 	// -------------------------------------
+	"anthropic/claude-3.5-haiku",
+	"anthropic/claude-3.5-sonnet",
+	"anthropic/claude-3.7-sonnet",
+	"deepseek-ai/deepseek-r1",
 	"ibm-granite/granite-20b-code-instruct-8k",
 	"ibm-granite/granite-3.0-2b-instruct",
 	"ibm-granite/granite-3.0-8b-instruct",
+	"ibm-granite/granite-3.1-2b-instruct",
+	"ibm-granite/granite-3.1-8b-instruct",
+	"ibm-granite/granite-3.2-8b-instruct",
 	"ibm-granite/granite-8b-code-instruct-128k",
 	"meta/llama-2-13b",
 	"meta/llama-2-13b-chat",
@@ -50,7 +57,6 @@ var ModelList = []string{
 	"meta/meta-llama-3-8b-instruct",
 	"mistralai/mistral-7b-instruct-v0.2",
 	"mistralai/mistral-7b-v0.1",
-	"mistralai/mixtral-8x7b-instruct-v0.1",
 	// -------------------------------------
 	// video model
 	// -------------------------------------
diff --git a/relay/adaptor/tencent/adaptor.go b/relay/adaptor/tencent/adaptor.go
index b20d4279..ce964dab 100644
--- a/relay/adaptor/tencent/adaptor.go
+++ b/relay/adaptor/tencent/adaptor.go
@@ -69,7 +69,7 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	return convertedRequest, nil
 }
 
-func (a *Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+func (a *Adaptor) ConvertImageRequest(_ *gin.Context, request *model.ImageRequest) (any, error) {
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
diff --git a/relay/adaptor/vertexai/adaptor.go b/relay/adaptor/vertexai/adaptor.go
index 3fab4a45..3b9e1a7f 100644
--- a/relay/adaptor/vertexai/adaptor.go
+++ b/relay/adaptor/vertexai/adaptor.go
@@ -105,7 +105,7 @@ func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *me
 	return nil
 }
 
-func (a *Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+func (a *Adaptor) ConvertImageRequest(_ *gin.Context, request *model.ImageRequest) (any, error) {
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
diff --git a/relay/adaptor/vertexai/claude/adapter.go b/relay/adaptor/vertexai/claude/adapter.go
index cb911cfe..f591e447 100644
--- a/relay/adaptor/vertexai/claude/adapter.go
+++ b/relay/adaptor/vertexai/claude/adapter.go
@@ -19,6 +19,7 @@ var ModelList = []string{
 	"claude-3-5-sonnet@20240620",
 	"claude-3-5-sonnet-v2@20241022",
 	"claude-3-5-haiku@20241022",
+	"claude-3-7-sonnet@20250219",
 }
 
 const anthropicVersion = "vertex-2023-10-16"
@@ -31,7 +32,11 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 		return nil, errors.New("request is nil")
 	}
 
-	claudeReq := anthropic.ConvertRequest(*request)
+	claudeReq, err := anthropic.ConvertRequest(c, *request)
+	if err != nil {
+		return nil, errors.Wrap(err, "convert request")
+	}
+
 	req := Request{
 		AnthropicVersion: anthropicVersion,
 		// Model:            claudeReq.Model,
diff --git a/relay/adaptor/xunfei/adaptor.go b/relay/adaptor/xunfei/adaptor.go
index b5967f26..404ec767 100644
--- a/relay/adaptor/xunfei/adaptor.go
+++ b/relay/adaptor/xunfei/adaptor.go
@@ -39,7 +39,7 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	return nil, nil
 }
 
-func (a *Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+func (a *Adaptor) ConvertImageRequest(_ *gin.Context, request *model.ImageRequest) (any, error) {
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
diff --git a/relay/adaptor/xunfei/main.go b/relay/adaptor/xunfei/main.go
index 9a8aef15..b02462fa 100644
--- a/relay/adaptor/xunfei/main.go
+++ b/relay/adaptor/xunfei/main.go
@@ -41,10 +41,15 @@ func requestOpenAI2Xunfei(request model.GeneralOpenAIRequest, xunfeiAppId string
 	xunfeiRequest.Header.AppId = xunfeiAppId
 	xunfeiRequest.Parameter.Chat.Domain = domain
 	xunfeiRequest.Parameter.Chat.Temperature = request.Temperature
-	xunfeiRequest.Parameter.Chat.TopK = request.N
 	xunfeiRequest.Parameter.Chat.MaxTokens = request.MaxTokens
 	xunfeiRequest.Payload.Message.Text = messages
 
+	if request.N != nil {
+		xunfeiRequest.Parameter.Chat.TopK = *request.N
+	} else {
+		xunfeiRequest.Parameter.Chat.TopK = 1
+	}
+
 	if strings.HasPrefix(domain, "generalv3") || domain == "4.0Ultra" {
 		functions := make([]model.Function, len(request.Tools))
 		for i, tool := range request.Tools {
diff --git a/relay/adaptor/zhipu/adaptor.go b/relay/adaptor/zhipu/adaptor.go
index 660bd379..1ae9e3f7 100644
--- a/relay/adaptor/zhipu/adaptor.go
+++ b/relay/adaptor/zhipu/adaptor.go
@@ -80,7 +80,7 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	}
 }
 
-func (a *Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error) {
+func (a *Adaptor) ConvertImageRequest(_ *gin.Context, request *model.ImageRequest) (any, error) {
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
diff --git a/relay/billing/ratio/model.go b/relay/billing/ratio/model.go
index e8b3b615..0c3fbb08 100644
--- a/relay/billing/ratio/model.go
+++ b/relay/billing/ratio/model.go
@@ -9,11 +9,20 @@ import (
 	"github.com/songquanpeng/one-api/common/logger"
 )
 
+// Constants defining currency conversion and token pricing
 const (
-	USD2RMB   = 7
-	USD       = 500 // $0.002 = 1 -> $1 = 500
-	MILLI_USD = 1.0 / 1000 * USD
-	RMB       = USD / USD2RMB
+	USD2RMB float64 = 7
+	// QuotaPerUsd is the number of tokens per USD
+	QuotaPerUsd float64 = 500000 // $0.002 / 1K tokens
+	// KiloTokensUsd multiply by the USD price per 1,000 tokens to get the quota cost per token
+	KiloTokensUsd float64 = QuotaPerUsd / 1000
+	// MilliTokensUsd multiply by the USD price per 1 million tokens to get the quota cost per token
+	MilliTokensUsd float64 = KiloTokensUsd / 1000
+	// KiloRmb multiply by the RMB price per 1,000 tokens to get the quota cost per token
+	KiloRmb float64 = KiloTokensUsd / USD2RMB
+	// MilliRmb multiply by the RMB price per 1 million tokens to get the quota cost per token
+	MilliRmb       float64 = MilliTokensUsd / USD2RMB
+	ImageUsdPerPic float64 = QuotaPerUsd / 1000
 )
 
 var modelRatioLock sync.RWMutex
@@ -25,305 +34,342 @@ var modelRatioLock sync.RWMutex
 // 1 === $0.002 / 1K tokens
 // 1 === ￥0.014 / 1k tokens
 var ModelRatio = map[string]float64{
+	// -------------------------------------
+	// OpenAI
 	// https://openai.com/pricing
-	"gpt-4":                   15,
-	"gpt-4-0314":              15,
-	"gpt-4-0613":              15,
-	"gpt-4-32k":               30,
-	"gpt-4-32k-0314":          30,
-	"gpt-4-32k-0613":          30,
-	"gpt-4-1106-preview":      5,     // $0.01 / 1K tokens
-	"gpt-4-0125-preview":      5,     // $0.01 / 1K tokens
-	"gpt-4-turbo-preview":     5,     // $0.01 / 1K tokens
-	"gpt-4-turbo":             5,     // $0.01 / 1K tokens
-	"gpt-4-turbo-2024-04-09":  5,     // $0.01 / 1K tokens
-	"gpt-4o":                  2.5,   // $0.005 / 1K tokens
-	"chatgpt-4o-latest":       2.5,   // $0.005 / 1K tokens
-	"gpt-4o-2024-05-13":       2.5,   // $0.005 / 1K tokens
-	"gpt-4o-2024-08-06":       1.25,  // $0.0025 / 1K tokens
-	"gpt-4o-2024-11-20":       1.25,  // $0.0025 / 1K tokens
-	"gpt-4o-mini":             0.075, // $0.00015 / 1K tokens
-	"gpt-4o-mini-2024-07-18":  0.075, // $0.00015 / 1K tokens
-	"gpt-4-vision-preview":    5,     // $0.01 / 1K tokens
-	"gpt-3.5-turbo":           0.25,  // $0.0005 / 1K tokens
-	"gpt-3.5-turbo-0301":      0.75,
-	"gpt-3.5-turbo-0613":      0.75,
-	"gpt-3.5-turbo-16k":       1.5, // $0.003 / 1K tokens
-	"gpt-3.5-turbo-16k-0613":  1.5,
-	"gpt-3.5-turbo-instruct":  0.75, // $0.0015 / 1K tokens
-	"gpt-3.5-turbo-1106":      0.5,  // $0.001 / 1K tokens
-	"gpt-3.5-turbo-0125":      0.25, // $0.0005 / 1K tokens
-	"o1":                      7.5,  // $15.00 / 1M input tokens
-	"o1-2024-12-17":           7.5,
-	"o1-preview":              7.5, // $15.00 / 1M input tokens
-	"o1-preview-2024-09-12":   7.5,
-	"o1-mini":                 1.5, // $3.00 / 1M input tokens
-	"o1-mini-2024-09-12":      1.5,
-	"o3-mini":                 1.5, // $3.00 / 1M input tokens
-	"o3-mini-2025-01-31":      1.5,
-	"davinci-002":             1,   // $0.002 / 1K tokens
-	"babbage-002":             0.2, // $0.0004 / 1K tokens
-	"text-ada-001":            0.2,
-	"text-babbage-001":        0.25,
-	"text-curie-001":          1,
-	"text-davinci-002":        10,
-	"text-davinci-003":        10,
-	"text-davinci-edit-001":   10,
-	"code-davinci-edit-001":   10,
-	"whisper-1":               15,  // $0.006 / minute -> $0.006 / 150 words -> $0.006 / 200 tokens -> $0.03 / 1k tokens
-	"tts-1":                   7.5, // $0.015 / 1K characters
-	"tts-1-1106":              7.5,
-	"tts-1-hd":                15, // $0.030 / 1K characters
-	"tts-1-hd-1106":           15,
-	"davinci":                 10,
-	"curie":                   10,
-	"babbage":                 10,
-	"ada":                     10,
-	"text-embedding-ada-002":  0.05,
-	"text-embedding-3-small":  0.01,
-	"text-embedding-3-large":  0.065,
-	"text-search-ada-doc-001": 10,
-	"text-moderation-stable":  0.1,
-	"text-moderation-latest":  0.1,
-	"dall-e-2":                0.02 * USD, // $0.016 - $0.020 / image
-	"dall-e-3":                0.04 * USD, // $0.040 - $0.120 / image
-	// https://docs.anthropic.com/en/docs/about-claude/models
-	"claude-instant-1.2":         0.8 / 1000 * USD,
-	"claude-2.0":                 8.0 / 1000 * USD,
-	"claude-2.1":                 8.0 / 1000 * USD,
-	"claude-3-haiku-20240307":    0.25 / 1000 * USD,
-	"claude-3-5-haiku-20241022":  1.0 / 1000 * USD,
-	"claude-3-5-haiku-latest":    1.0 / 1000 * USD,
-	"claude-3-sonnet-20240229":   3.0 / 1000 * USD,
-	"claude-3-5-sonnet-20240620": 3.0 / 1000 * USD,
-	"claude-3-5-sonnet-20241022": 3.0 / 1000 * USD,
-	"claude-3-5-sonnet-latest":   3.0 / 1000 * USD,
-	"claude-3-opus-20240229":     15.0 / 1000 * USD,
+	// -------------------------------------
+	"gpt-4.5-preview":            75 * MilliTokensUsd,
+	"gpt-4.5-preview-2025-02-27": 75 * MilliTokensUsd,
+	"gpt-4":                      30 * MilliTokensUsd,
+	"gpt-4-0314":                 30 * MilliTokensUsd,
+	"gpt-4-0613":                 30 * MilliTokensUsd,
+	"gpt-4-32k":                  60 * MilliTokensUsd,
+	"gpt-4-32k-0314":             60 * MilliTokensUsd,
+	"gpt-4-32k-0613":             60 * MilliTokensUsd,
+	"gpt-4-1106-preview":         10 * MilliTokensUsd,
+	"gpt-4-0125-preview":         10 * MilliTokensUsd,
+	"gpt-4-turbo-preview":        10 * MilliTokensUsd,
+	"gpt-4-turbo":                10 * MilliTokensUsd,
+	"gpt-4-turbo-2024-04-09":     10 * MilliTokensUsd,
+	"gpt-4o":                     2.5 * MilliTokensUsd,
+	"chatgpt-4o-latest":          5 * MilliTokensUsd,
+	"gpt-4o-2024-05-13":          5 * MilliTokensUsd,
+	"gpt-4o-2024-08-06":          2.5 * MilliTokensUsd,
+	"gpt-4o-2024-11-20":          2.5 * MilliTokensUsd,
+	"gpt-4o-search-preview":      5 * MilliTokensUsd,
+	"gpt-4o-mini":                0.15 * MilliTokensUsd,
+	"gpt-4o-mini-2024-07-18":     0.15 * MilliTokensUsd,
+	"gpt-4o-mini-search-preview": 0.15 * MilliTokensUsd,
+	"gpt-4-vision-preview":       10 * MilliTokensUsd,
+	// Audio billing will mix text and audio tokens, the unit price is different.
+	// Here records the cost of text, the cost multiplier of audio
+	// relative to text is in AudioRatio
+	"gpt-4o-audio-preview":                 2.5 * MilliTokensUsd,
+	"gpt-4o-audio-preview-2024-12-17":      2.5 * MilliTokensUsd,
+	"gpt-4o-audio-preview-2024-10-01":      2.5 * MilliTokensUsd,
+	"gpt-4o-mini-audio-preview":            0.15 * MilliTokensUsd,
+	"gpt-4o-mini-audio-preview-2024-12-17": 0.15 * MilliTokensUsd,
+	"gpt-3.5-turbo":                        0.5 * MilliTokensUsd,
+	"gpt-3.5-turbo-0301":                   1.5 * MilliTokensUsd,
+	"gpt-3.5-turbo-0613":                   1.5 * MilliTokensUsd,
+	"gpt-3.5-turbo-16k":                    3 * MilliTokensUsd,
+	"gpt-3.5-turbo-16k-0613":               3 * MilliTokensUsd,
+	"gpt-3.5-turbo-instruct":               1.5 * MilliTokensUsd,
+	"gpt-3.5-turbo-1106":                   1 * MilliTokensUsd,
+	"gpt-3.5-turbo-0125":                   0.5 * MilliTokensUsd,
+	"o1":                                   15 * MilliTokensUsd,
+	"o1-2024-12-17":                        15 * MilliTokensUsd,
+	"o1-preview":                           15 * MilliTokensUsd,
+	"o1-preview-2024-09-12":                15 * MilliTokensUsd,
+	"o1-mini":                              1.1 * MilliTokensUsd,
+	"o1-mini-2024-09-12":                   1.1 * MilliTokensUsd,
+	"o3-mini":                              1.1 * MilliTokensUsd,
+	"o3-mini-2025-01-31":                   1.1 * MilliTokensUsd,
+	"davinci-002":                          2 * MilliTokensUsd,
+	"babbage-002":                          0.4 * MilliTokensUsd,
+	"text-ada-001":                         0.4 * MilliTokensUsd,
+	"text-babbage-001":                     0.5 * MilliTokensUsd,
+	"text-curie-001":                       2 * MilliTokensUsd,
+	"text-davinci-002":                     20 * MilliTokensUsd,
+	"text-davinci-003":                     20 * MilliTokensUsd,
+	"text-davinci-edit-001":                20 * MilliTokensUsd,
+	"code-davinci-edit-001":                20 * MilliTokensUsd,
+	"whisper-1":                            30 * MilliTokensUsd,
+	"tts-1":                                15 * MilliTokensUsd,
+	"tts-1-1106":                           15 * MilliTokensUsd,
+	"tts-1-hd":                             30 * MilliTokensUsd,
+	"tts-1-hd-1106":                        30 * MilliTokensUsd,
+	"davinci":                              20 * MilliTokensUsd,
+	"curie":                                20 * MilliTokensUsd,
+	"babbage":                              20 * MilliTokensUsd,
+	"ada":                                  20 * MilliTokensUsd,
+	"text-embedding-ada-002":               0.1 * MilliTokensUsd,
+	"text-embedding-3-small":               0.02 * MilliTokensUsd,
+	"text-embedding-3-large":               0.13 * MilliTokensUsd,
+	"text-search-ada-doc-001":              20 * MilliTokensUsd,
+	"text-moderation-stable":               0.2 * MilliTokensUsd,
+	"text-moderation-latest":               0.2 * MilliTokensUsd,
+	"dall-e-2":                             0.02 * ImageUsdPerPic,
+	"dall-e-3":                             0.04 * ImageUsdPerPic,
+	// https://www.anthropic.com/api#pricing
+	"claude-instant-1.2":         0.8 * MilliTokensUsd,
+	"claude-2.0":                 8.0 * MilliTokensUsd,
+	"claude-2.1":                 8.0 * MilliTokensUsd,
+	"claude-3-haiku-20240307":    0.25 * MilliTokensUsd,
+	"claude-3-5-haiku-20241022":  1.0 * MilliTokensUsd,
+	"claude-3-5-haiku-latest":    1.0 * MilliTokensUsd,
+	"claude-3-sonnet-20240229":   3.0 * MilliTokensUsd,
+	"claude-3-5-sonnet-20240620": 3.0 * MilliTokensUsd,
+	"claude-3-5-sonnet-20241022": 3.0 * MilliTokensUsd,
+	"claude-3-5-sonnet-latest":   3.0 * MilliTokensUsd,
+	"claude-3-7-sonnet-20250219": 3.0 * MilliTokensUsd,
+	"claude-3-7-sonnet-latest":   3.0 * MilliTokensUsd,
+	"claude-3-opus-20240229":     15.0 * MilliTokensUsd,
 	// https://cloud.baidu.com/doc/WENXINWORKSHOP/s/hlrk4akp7
-	"ERNIE-4.0-8K":       0.120 * RMB,
-	"ERNIE-3.5-8K":       0.012 * RMB,
-	"ERNIE-3.5-8K-0205":  0.024 * RMB,
-	"ERNIE-3.5-8K-1222":  0.012 * RMB,
-	"ERNIE-Bot-8K":       0.024 * RMB,
-	"ERNIE-3.5-4K-0205":  0.012 * RMB,
-	"ERNIE-Speed-8K":     0.004 * RMB,
-	"ERNIE-Speed-128K":   0.004 * RMB,
-	"ERNIE-Lite-8K-0922": 0.008 * RMB,
-	"ERNIE-Lite-8K-0308": 0.003 * RMB,
-	"ERNIE-Tiny-8K":      0.001 * RMB,
-	"BLOOMZ-7B":          0.004 * RMB,
-	"Embedding-V1":       0.002 * RMB,
-	"bge-large-zh":       0.002 * RMB,
-	"bge-large-en":       0.002 * RMB,
-	"tao-8k":             0.002 * RMB,
+	"ERNIE-4.0-8K":       0.120 * KiloRmb,
+	"ERNIE-3.5-8K":       0.012 * KiloRmb,
+	"ERNIE-3.5-8K-0205":  0.024 * KiloRmb,
+	"ERNIE-3.5-8K-1222":  0.012 * KiloRmb,
+	"ERNIE-Bot-8K":       0.024 * KiloRmb,
+	"ERNIE-3.5-4K-0205":  0.012 * KiloRmb,
+	"ERNIE-Speed-8K":     0.004 * KiloRmb,
+	"ERNIE-Speed-128K":   0.004 * KiloRmb,
+	"ERNIE-Lite-8K-0922": 0.008 * KiloRmb,
+	"ERNIE-Lite-8K-0308": 0.003 * KiloRmb,
+	"ERNIE-Tiny-8K":      0.001 * KiloRmb,
+	"BLOOMZ-7B":          0.004 * KiloRmb,
+	"Embedding-V1":       0.002 * KiloRmb,
+	"bge-large-zh":       0.002 * KiloRmb,
+	"bge-large-en":       0.002 * KiloRmb,
+	"tao-8k":             0.002 * KiloRmb,
 	// https://ai.google.dev/pricing
 	// https://cloud.google.com/vertex-ai/generative-ai/pricing
-	// "gemma-2-2b-it":                       0,
-	// "gemma-2-9b-it":                       0,
-	// "gemma-2-27b-it":                      0,
-	"gemini-pro":                          0.25 * MILLI_USD, // $0.00025 / 1k characters -> $0.001 / 1k tokens
-	"gemini-1.0-pro":                      0.125 * MILLI_USD,
-	"gemini-1.5-pro":                      1.25 * MILLI_USD,
-	"gemini-1.5-pro-001":                  1.25 * MILLI_USD,
-	"gemini-1.5-pro-experimental":         1.25 * MILLI_USD,
-	"gemini-1.5-flash":                    0.075 * MILLI_USD,
-	"gemini-1.5-flash-001":                0.075 * MILLI_USD,
-	"gemini-1.5-flash-8b":                 0.0375 * MILLI_USD,
-	"gemini-2.0-flash-exp":                0.075 * MILLI_USD,
-	"gemini-2.0-flash":                    0.15 * MILLI_USD,
-	"gemini-2.0-flash-001":                0.15 * MILLI_USD,
-	"gemini-2.0-flash-lite-preview-02-05": 0.075 * MILLI_USD,
-	"gemini-2.0-flash-thinking-exp-01-21": 0.075 * MILLI_USD,
-	"gemini-2.0-pro-exp-02-05":            1.25 * MILLI_USD,
-	"aqa":                                 1,
+	"gemma-2-2b-it":                         0,
+	"gemma-2-9b-it":                         0,
+	"gemma-2-27b-it":                        0,
+	"gemma-3-27b-it":                        0,
+	"gemini-pro":                            0.25 * MilliTokensUsd, // $0.00025 / 1k characters -> $0.001 / 1k tokens
+	"gemini-1.0-pro":                        0.125 * MilliTokensUsd,
+	"gemini-1.0-pro-vision":                 0.125 * MilliTokensUsd,
+	"gemini-1.5-pro":                        1.25 * MilliTokensUsd,
+	"gemini-1.5-pro-001":                    1.25 * MilliTokensUsd,
+	"gemini-1.5-pro-002":                    1.25 * MilliTokensUsd,
+	"gemini-1.5-pro-experimental":           1.25 * MilliTokensUsd,
+	"gemini-1.5-flash":                      0.075 * MilliTokensUsd,
+	"gemini-1.5-flash-001":                  0.075 * MilliTokensUsd,
+	"gemini-1.5-flash-002":                  0.075 * MilliTokensUsd,
+	"gemini-1.5-flash-8b":                   0.0375 * MilliTokensUsd,
+	"gemini-2.0-flash":                      0.15 * MilliTokensUsd,
+	"gemini-2.0-flash-exp":                  0.075 * MilliTokensUsd,
+	"gemini-2.0-flash-001":                  0.15 * MilliTokensUsd,
+	"gemini-2.0-flash-lite":                 0.075 * MilliTokensUsd,
+	"gemini-2.0-flash-lite-001":             0.075 * MilliTokensUsd,
+	"gemini-2.0-flash-lite-preview-02-05":   0.075 * MilliTokensUsd,
+	"gemini-2.0-flash-thinking-exp-01-21":   0.075 * MilliTokensUsd,
+	"gemini-2.0-flash-exp-image-generation": 0.075 * MilliTokensUsd,
+	"gemini-2.0-pro-exp-02-05":              1.25 * MilliTokensUsd,
+	"gemini-2.5-pro-exp-03-25":              1.25 * MilliTokensUsd,
+	"aqa":                                   1,
 	// https://open.bigmodel.cn/pricing
-	"glm-zero-preview": 0.01 * RMB,
-	"glm-4-plus":       0.05 * RMB,
-	"glm-4-0520":       0.1 * RMB,
-	"glm-4-airx":       0.01 * RMB,
-	"glm-4-air":        0.0005 * RMB,
-	"glm-4-long":       0.001 * RMB,
-	"glm-4-flashx":     0.0001 * RMB,
+	"glm-zero-preview": 0.01 * KiloRmb,
+	"glm-4-plus":       0.05 * KiloRmb,
+	"glm-4-0520":       0.1 * KiloRmb,
+	"glm-4-airx":       0.01 * KiloRmb,
+	"glm-4-air":        0.0005 * KiloRmb,
+	"glm-4-long":       0.001 * KiloRmb,
+	"glm-4-flashx":     0.0001 * KiloRmb,
 	"glm-4-flash":      0,
-	"glm-4":            0.1 * RMB,   // deprecated model, available until 2025/06
-	"glm-3-turbo":      0.001 * RMB, // deprecated model, available until 2025/06
-	"glm-4v-plus":      0.004 * RMB,
-	"glm-4v":           0.05 * RMB,
+	"glm-4":            0.1 * KiloRmb,   // deprecated model, available until 2025/06
+	"glm-3-turbo":      0.001 * KiloRmb, // deprecated model, available until 2025/06
+	"glm-4v-plus":      0.004 * KiloRmb,
+	"glm-4v":           0.05 * KiloRmb,
 	"glm-4v-flash":     0,
-	"cogview-3-plus":   0.06 * RMB,
-	"cogview-3":        0.1 * RMB,
+	"cogview-3-plus":   0.06 * KiloRmb,
+	"cogview-3":        0.1 * KiloRmb,
 	"cogview-3-flash":  0,
-	"cogviewx":         0.5 * RMB,
+	"cogviewx":         0.5 * KiloRmb,
 	"cogviewx-flash":   0,
-	"charglm-4":        0.001 * RMB,
-	"emohaa":           0.015 * RMB,
-	"codegeex-4":       0.0001 * RMB,
-	"embedding-2":      0.0005 * RMB,
-	"embedding-3":      0.0005 * RMB,
+	"charglm-4":        0.001 * KiloRmb,
+	"emohaa":           0.015 * KiloRmb,
+	"codegeex-4":       0.0001 * KiloRmb,
+	"embedding-2":      0.0005 * KiloRmb,
+	"embedding-3":      0.0005 * KiloRmb,
 	// https://help.aliyun.com/zh/dashscope/developer-reference/tongyi-thousand-questions-metering-and-billing
-	"qwen-turbo":                    0.0003 * RMB,
-	"qwen-turbo-latest":             0.0003 * RMB,
-	"qwen-plus":                     0.0008 * RMB,
-	"qwen-plus-latest":              0.0008 * RMB,
-	"qwen-max":                      0.0024 * RMB,
-	"qwen-max-latest":               0.0024 * RMB,
-	"qwen-max-longcontext":          0.0005 * RMB,
-	"qwen-vl-max":                   0.003 * RMB,
-	"qwen-vl-max-latest":            0.003 * RMB,
-	"qwen-vl-plus":                  0.0015 * RMB,
-	"qwen-vl-plus-latest":           0.0015 * RMB,
-	"qwen-vl-ocr":                   0.005 * RMB,
-	"qwen-vl-ocr-latest":            0.005 * RMB,
+	"qwen-turbo":                    0.0003 * KiloRmb,
+	"qwen-turbo-latest":             0.0003 * KiloRmb,
+	"qwen-plus":                     0.0008 * KiloRmb,
+	"qwen-plus-latest":              0.0008 * KiloRmb,
+	"qwen-max":                      0.0024 * KiloRmb,
+	"qwen-max-latest":               0.0024 * KiloRmb,
+	"qwen-max-longcontext":          0.0005 * KiloRmb,
+	"qwen-vl-max":                   0.003 * KiloRmb,
+	"qwen-vl-max-latest":            0.003 * KiloRmb,
+	"qwen-vl-plus":                  0.0015 * KiloRmb,
+	"qwen-vl-plus-latest":           0.0015 * KiloRmb,
+	"qwen-vl-ocr":                   0.005 * KiloRmb,
+	"qwen-vl-ocr-latest":            0.005 * KiloRmb,
 	"qwen-audio-turbo":              1.4286,
-	"qwen-math-plus":                0.004 * RMB,
-	"qwen-math-plus-latest":         0.004 * RMB,
-	"qwen-math-turbo":               0.002 * RMB,
-	"qwen-math-turbo-latest":        0.002 * RMB,
-	"qwen-coder-plus":               0.0035 * RMB,
-	"qwen-coder-plus-latest":        0.0035 * RMB,
-	"qwen-coder-turbo":              0.002 * RMB,
-	"qwen-coder-turbo-latest":       0.002 * RMB,
-	"qwen-mt-plus":                  0.015 * RMB,
-	"qwen-mt-turbo":                 0.001 * RMB,
-	"qwq-32b-preview":               0.002 * RMB,
-	"qwen2.5-72b-instruct":          0.004 * RMB,
-	"qwen2.5-32b-instruct":          0.03 * RMB,
-	"qwen2.5-14b-instruct":          0.001 * RMB,
-	"qwen2.5-7b-instruct":           0.0005 * RMB,
-	"qwen2.5-3b-instruct":           0.006 * RMB,
-	"qwen2.5-1.5b-instruct":         0.0003 * RMB,
-	"qwen2.5-0.5b-instruct":         0.0003 * RMB,
-	"qwen2-72b-instruct":            0.004 * RMB,
-	"qwen2-57b-a14b-instruct":       0.0035 * RMB,
-	"qwen2-7b-instruct":             0.001 * RMB,
-	"qwen2-1.5b-instruct":           0.001 * RMB,
-	"qwen2-0.5b-instruct":           0.001 * RMB,
-	"qwen1.5-110b-chat":             0.007 * RMB,
-	"qwen1.5-72b-chat":              0.005 * RMB,
-	"qwen1.5-32b-chat":              0.0035 * RMB,
-	"qwen1.5-14b-chat":              0.002 * RMB,
-	"qwen1.5-7b-chat":               0.001 * RMB,
-	"qwen1.5-1.8b-chat":             0.001 * RMB,
-	"qwen1.5-0.5b-chat":             0.001 * RMB,
-	"qwen-72b-chat":                 0.02 * RMB,
-	"qwen-14b-chat":                 0.008 * RMB,
-	"qwen-7b-chat":                  0.006 * RMB,
-	"qwen-1.8b-chat":                0.006 * RMB,
-	"qwen-1.8b-longcontext-chat":    0.006 * RMB,
-	"qvq-72b-preview":               0.012 * RMB,
-	"qwen2.5-vl-72b-instruct":       0.016 * RMB,
-	"qwen2.5-vl-7b-instruct":        0.002 * RMB,
-	"qwen2.5-vl-3b-instruct":        0.0012 * RMB,
-	"qwen2-vl-7b-instruct":          0.016 * RMB,
-	"qwen2-vl-2b-instruct":          0.002 * RMB,
-	"qwen-vl-v1":                    0.002 * RMB,
-	"qwen-vl-chat-v1":               0.002 * RMB,
-	"qwen2-audio-instruct":          0.002 * RMB,
-	"qwen-audio-chat":               0.002 * RMB,
-	"qwen2.5-math-72b-instruct":     0.004 * RMB,
-	"qwen2.5-math-7b-instruct":      0.001 * RMB,
-	"qwen2.5-math-1.5b-instruct":    0.001 * RMB,
-	"qwen2-math-72b-instruct":       0.004 * RMB,
-	"qwen2-math-7b-instruct":        0.001 * RMB,
-	"qwen2-math-1.5b-instruct":      0.001 * RMB,
-	"qwen2.5-coder-32b-instruct":    0.002 * RMB,
-	"qwen2.5-coder-14b-instruct":    0.002 * RMB,
-	"qwen2.5-coder-7b-instruct":     0.001 * RMB,
-	"qwen2.5-coder-3b-instruct":     0.001 * RMB,
-	"qwen2.5-coder-1.5b-instruct":   0.001 * RMB,
-	"qwen2.5-coder-0.5b-instruct":   0.001 * RMB,
-	"text-embedding-v1":             0.0007 * RMB, // ￥0.0007 / 1k tokens
-	"text-embedding-v3":             0.0007 * RMB,
-	"text-embedding-v2":             0.0007 * RMB,
-	"text-embedding-async-v2":       0.0007 * RMB,
-	"text-embedding-async-v1":       0.0007 * RMB,
-	"ali-stable-diffusion-xl":       8.00,
-	"ali-stable-diffusion-v1.5":     8.00,
-	"wanx-v1":                       8.00,
-	"deepseek-r1":                   0.002 * RMB,
-	"deepseek-v3":                   0.001 * RMB,
-	"deepseek-r1-distill-qwen-1.5b": 0.001 * RMB,
-	"deepseek-r1-distill-qwen-7b":   0.0005 * RMB,
-	"deepseek-r1-distill-qwen-14b":  0.001 * RMB,
-	"deepseek-r1-distill-qwen-32b":  0.002 * RMB,
-	"deepseek-r1-distill-llama-8b":  0.0005 * RMB,
-	"deepseek-r1-distill-llama-70b": 0.004 * RMB,
-	"SparkDesk":                     1.2858, // ￥0.018 / 1k tokens
-	"SparkDesk-v1.1":                1.2858, // ￥0.018 / 1k tokens
-	"SparkDesk-v2.1":                1.2858, // ￥0.018 / 1k tokens
-	"SparkDesk-v3.1":                1.2858, // ￥0.018 / 1k tokens
-	"SparkDesk-v3.1-128K":           1.2858, // ￥0.018 / 1k tokens
-	"SparkDesk-v3.5":                1.2858, // ￥0.018 / 1k tokens
-	"SparkDesk-v3.5-32K":            1.2858, // ￥0.018 / 1k tokens
-	"SparkDesk-v4.0":                1.2858, // ￥0.018 / 1k tokens
-	"360GPT_S2_V9":                  0.8572, // ¥0.012 / 1k tokens
-	"embedding-bert-512-v1":         0.0715, // ¥0.001 / 1k tokens
-	"embedding_s1_v1":               0.0715, // ¥0.001 / 1k tokens
-	"semantic_similarity_s1_v1":     0.0715, // ¥0.001 / 1k tokens
+	"qwen-math-plus":                0.004 * KiloRmb,
+	"qwen-math-plus-latest":         0.004 * KiloRmb,
+	"qwen-math-turbo":               0.002 * KiloRmb,
+	"qwen-math-turbo-latest":        0.002 * KiloRmb,
+	"qwen-coder-plus":               0.0035 * KiloRmb,
+	"qwen-coder-plus-latest":        0.0035 * KiloRmb,
+	"qwen-coder-turbo":              0.002 * KiloRmb,
+	"qwen-coder-turbo-latest":       0.002 * KiloRmb,
+	"qwen-mt-plus":                  0.015 * KiloRmb,
+	"qwen-mt-turbo":                 0.001 * KiloRmb,
+	"qwq-32b-preview":               0.002 * KiloRmb,
+	"qwen2.5-72b-instruct":          0.004 * KiloRmb,
+	"qwen2.5-32b-instruct":          0.03 * KiloRmb,
+	"qwen2.5-14b-instruct":          0.001 * KiloRmb,
+	"qwen2.5-7b-instruct":           0.0005 * KiloRmb,
+	"qwen2.5-3b-instruct":           0.006 * KiloRmb,
+	"qwen2.5-1.5b-instruct":         0.0003 * KiloRmb,
+	"qwen2.5-0.5b-instruct":         0.0003 * KiloRmb,
+	"qwen2-72b-instruct":            0.004 * KiloRmb,
+	"qwen2-57b-a14b-instruct":       0.0035 * KiloRmb,
+	"qwen2-7b-instruct":             0.001 * KiloRmb,
+	"qwen2-1.5b-instruct":           0.001 * KiloRmb,
+	"qwen2-0.5b-instruct":           0.001 * KiloRmb,
+	"qwen1.5-110b-chat":             0.007 * KiloRmb,
+	"qwen1.5-72b-chat":              0.005 * KiloRmb,
+	"qwen1.5-32b-chat":              0.0035 * KiloRmb,
+	"qwen1.5-14b-chat":              0.002 * KiloRmb,
+	"qwen1.5-7b-chat":               0.001 * KiloRmb,
+	"qwen1.5-1.8b-chat":             0.001 * KiloRmb,
+	"qwen1.5-0.5b-chat":             0.001 * KiloRmb,
+	"qwen-72b-chat":                 0.02 * KiloRmb,
+	"qwen-14b-chat":                 0.008 * KiloRmb,
+	"qwen-7b-chat":                  0.006 * KiloRmb,
+	"qwen-1.8b-chat":                0.006 * KiloRmb,
+	"qwen-1.8b-longcontext-chat":    0.006 * KiloRmb,
+	"qvq-72b-preview":               0.012 * KiloRmb,
+	"qwen2.5-vl-72b-instruct":       0.016 * KiloRmb,
+	"qwen2.5-vl-7b-instruct":        0.002 * KiloRmb,
+	"qwen2.5-vl-3b-instruct":        0.0012 * KiloRmb,
+	"qwen2-vl-7b-instruct":          0.016 * KiloRmb,
+	"qwen2-vl-2b-instruct":          0.002 * KiloRmb,
+	"qwen-vl-v1":                    0.002 * KiloRmb,
+	"qwen-vl-chat-v1":               0.002 * KiloRmb,
+	"qwen2-audio-instruct":          0.002 * KiloRmb,
+	"qwen-audio-chat":               0.002 * KiloRmb,
+	"qwen2.5-math-72b-instruct":     0.004 * KiloRmb,
+	"qwen2.5-math-7b-instruct":      0.001 * KiloRmb,
+	"qwen2.5-math-1.5b-instruct":    0.001 * KiloRmb,
+	"qwen2-math-72b-instruct":       0.004 * KiloRmb,
+	"qwen2-math-7b-instruct":        0.001 * KiloRmb,
+	"qwen2-math-1.5b-instruct":      0.001 * KiloRmb,
+	"qwen2.5-coder-32b-instruct":    0.002 * KiloRmb,
+	"qwen2.5-coder-14b-instruct":    0.002 * KiloRmb,
+	"qwen2.5-coder-7b-instruct":     0.001 * KiloRmb,
+	"qwen2.5-coder-3b-instruct":     0.001 * KiloRmb,
+	"qwen2.5-coder-1.5b-instruct":   0.001 * KiloRmb,
+	"qwen2.5-coder-0.5b-instruct":   0.001 * KiloRmb,
+	"text-embedding-v1":             0.0007 * KiloRmb, // ￥0.0007 / 1k tokens
+	"text-embedding-v3":             0.0007 * KiloRmb,
+	"text-embedding-v2":             0.0007 * KiloRmb,
+	"text-embedding-async-v2":       0.0007 * KiloRmb,
+	"text-embedding-async-v1":       0.0007 * KiloRmb,
+	"ali-stable-diffusion-xl":       0.016 * ImageUsdPerPic,
+	"ali-stable-diffusion-v1.5":     0.016 * ImageUsdPerPic,
+	"wanx-v1":                       0.016 * ImageUsdPerPic,
+	"deepseek-r1":                   0.002 * KiloRmb,
+	"deepseek-v3":                   0.001 * KiloRmb,
+	"deepseek-r1-distill-qwen-1.5b": 0.001 * KiloRmb,
+	"deepseek-r1-distill-qwen-7b":   0.0005 * KiloRmb,
+	"deepseek-r1-distill-qwen-14b":  0.001 * KiloRmb,
+	// "deepseek-r1-distill-qwen-32b":  0.002 * KiloRmb,
+	"deepseek-r1-distill-llama-8b": 0.0005 * KiloRmb,
+	// "deepseek-r1-distill-llama-70b": 0.004 * KiloRmb,
+	"SparkDesk":                 1.2858, // ￥0.018 / 1k tokens
+	"SparkDesk-v1.1":            1.2858, // ￥0.018 / 1k tokens
+	"SparkDesk-v2.1":            1.2858, // ￥0.018 / 1k tokens
+	"SparkDesk-v3.1":            1.2858, // ￥0.018 / 1k tokens
+	"SparkDesk-v3.1-128K":       1.2858, // ￥0.018 / 1k tokens
+	"SparkDesk-v3.5":            1.2858, // ￥0.018 / 1k tokens
+	"SparkDesk-v3.5-32K":        1.2858, // ￥0.018 / 1k tokens
+	"SparkDesk-v4.0":            1.2858, // ￥0.018 / 1k tokens
+	"360GPT_S2_V9":              0.8572, // ¥0.012 / 1k tokens
+	"embedding-bert-512-v1":     0.0715, // ¥0.001 / 1k tokens
+	"embedding_s1_v1":           0.0715, // ¥0.001 / 1k tokens
+	"semantic_similarity_s1_v1": 0.0715, // ¥0.001 / 1k tokens
 	// https://cloud.tencent.com/document/product/1729/97731#e0e6be58-60c8-469f-bdeb-6c264ce3b4d0
-	"hunyuan-turbo":             0.015 * RMB,
-	"hunyuan-large":             0.004 * RMB,
-	"hunyuan-large-longcontext": 0.006 * RMB,
-	"hunyuan-standard":          0.0008 * RMB,
-	"hunyuan-standard-256K":     0.0005 * RMB,
-	"hunyuan-translation-lite":  0.005 * RMB,
-	"hunyuan-role":              0.004 * RMB,
-	"hunyuan-functioncall":      0.004 * RMB,
-	"hunyuan-code":              0.004 * RMB,
-	"hunyuan-turbo-vision":      0.08 * RMB,
-	"hunyuan-vision":            0.018 * RMB,
-	"hunyuan-embedding":         0.0007 * RMB,
+	"hunyuan-turbo":             0.015 * KiloRmb,
+	"hunyuan-large":             0.004 * KiloRmb,
+	"hunyuan-large-longcontext": 0.006 * KiloRmb,
+	"hunyuan-standard":          0.0008 * KiloRmb,
+	"hunyuan-standard-256K":     0.0005 * KiloRmb,
+	"hunyuan-translation-lite":  0.005 * KiloRmb,
+	"hunyuan-role":              0.004 * KiloRmb,
+	"hunyuan-functioncall":      0.004 * KiloRmb,
+	"hunyuan-code":              0.004 * KiloRmb,
+	"hunyuan-turbo-vision":      0.08 * KiloRmb,
+	"hunyuan-vision":            0.018 * KiloRmb,
+	"hunyuan-embedding":         0.0007 * KiloRmb,
 	// https://platform.moonshot.cn/pricing
-	"moonshot-v1-8k":   0.012 * RMB,
-	"moonshot-v1-32k":  0.024 * RMB,
-	"moonshot-v1-128k": 0.06 * RMB,
+	"moonshot-v1-8k":   0.012 * KiloRmb,
+	"moonshot-v1-32k":  0.024 * KiloRmb,
+	"moonshot-v1-128k": 0.06 * KiloRmb,
 	// https://platform.baichuan-ai.com/price
-	"Baichuan2-Turbo":      0.008 * RMB,
-	"Baichuan2-Turbo-192k": 0.016 * RMB,
-	"Baichuan2-53B":        0.02 * RMB,
+	"Baichuan2-Turbo":      0.008 * KiloRmb,
+	"Baichuan2-Turbo-192k": 0.016 * KiloRmb,
+	"Baichuan2-53B":        0.02 * KiloRmb,
 	// https://api.minimax.chat/document/price
-	"abab6.5-chat":  0.03 * RMB,
-	"abab6.5s-chat": 0.01 * RMB,
-	"abab6-chat":    0.1 * RMB,
-	"abab5.5-chat":  0.015 * RMB,
-	"abab5.5s-chat": 0.005 * RMB,
+	"abab6.5-chat":  0.03 * KiloRmb,
+	"abab6.5s-chat": 0.01 * KiloRmb,
+	"abab6-chat":    0.1 * KiloRmb,
+	"abab5.5-chat":  0.015 * KiloRmb,
+	"abab5.5s-chat": 0.005 * KiloRmb,
 	// https://docs.mistral.ai/platform/pricing/
-	"open-mistral-7b":       0.25 / 1000 * USD,
-	"open-mixtral-8x7b":     0.7 / 1000 * USD,
-	"mistral-small-latest":  2.0 / 1000 * USD,
-	"mistral-medium-latest": 2.7 / 1000 * USD,
-	"mistral-large-latest":  8.0 / 1000 * USD,
-	"mistral-embed":         0.1 / 1000 * USD,
-	// https://wow.groq.com/#:~:text=inquiries%C2%A0here.-,Model,-Current%20Speed
-	"gemma-7b-it":                           0.07 / 1000000 * USD,
-	"gemma2-9b-it":                          0.20 / 1000000 * USD,
-	"llama-3.1-70b-versatile":               0.59 / 1000000 * USD,
-	"llama-3.1-8b-instant":                  0.05 / 1000000 * USD,
-	"llama-3.2-11b-text-preview":            0.05 / 1000000 * USD,
-	"llama-3.2-11b-vision-preview":          0.05 / 1000000 * USD,
-	"llama-3.2-1b-preview":                  0.05 / 1000000 * USD,
-	"llama-3.2-3b-preview":                  0.05 / 1000000 * USD,
-	"llama-3.2-90b-text-preview":            0.59 / 1000000 * USD,
-	"llama-guard-3-8b":                      0.05 / 1000000 * USD,
-	"llama3-70b-8192":                       0.59 / 1000000 * USD,
-	"llama3-8b-8192":                        0.05 / 1000000 * USD,
-	"llama3-groq-70b-8192-tool-use-preview": 0.89 / 1000000 * USD,
-	"llama3-groq-8b-8192-tool-use-preview":  0.19 / 1000000 * USD,
-	"mixtral-8x7b-32768":                    0.24 / 1000000 * USD,
-
+	"open-mistral-7b":       0.25 * MilliTokensUsd,
+	"open-mixtral-8x7b":     0.7 * MilliTokensUsd,
+	"mistral-small-latest":  2.0 * MilliTokensUsd,
+	"mistral-medium-latest": 2.7 * MilliTokensUsd,
+	"mistral-large-latest":  8.0 * MilliTokensUsd,
+	"mistral-embed":         0.1 * MilliTokensUsd,
+	// -------------------------------------
+	// https://groq.com/pricing/
+	// -------------------------------------
+	"gemma2-9b-it":                          0.20 * MilliTokensUsd,
+	"llama-3.1-8b-instant":                  0.05 * MilliTokensUsd,
+	"llama-3.2-11b-text-preview":            0.18 * MilliTokensUsd,
+	"llama-3.2-11b-vision-preview":          0.18 * MilliTokensUsd,
+	"llama-3.2-1b-preview":                  0.04 * MilliTokensUsd,
+	"llama-3.2-3b-preview":                  0.06 * MilliTokensUsd,
+	"llama-3.2-90b-text-preview":            0.90 * MilliTokensUsd,
+	"llama-3.2-90b-vision-preview":          0.90 * MilliTokensUsd,
+	"llama-3.3-70b-versatile":               0.59 * MilliTokensUsd,
+	"llama-guard-3-8b":                      0.20 * MilliTokensUsd,
+	"llama3-70b-8192":                       0.59 * MilliTokensUsd,
+	"llama3-8b-8192":                        0.05 * MilliTokensUsd,
+	"llama3-groq-70b-8192-tool-use-preview": 0.59 * MilliTokensUsd,
+	"llama3-groq-8b-8192-tool-use-preview":  0.05 * MilliTokensUsd,
+	"llama-3.3-70b-specdec":                 0.59 * MilliTokensUsd,
+	"mistral-saba-24b":                      0.79 * MilliTokensUsd,
+	"qwen-qwq-32b":                          0.29 * MilliTokensUsd,
+	"qwen-2.5-coder-32b":                    0.79 * MilliTokensUsd,
+	"qwen-2.5-32b":                          0.79 * MilliTokensUsd,
+	"mixtral-8x7b-32768":                    0.24 * MilliTokensUsd,
+	"whisper-large-v3":                      0.111 * MilliTokensUsd,
+	"whisper-large-v3-turbo":                0.04 * MilliTokensUsd,
+	"distil-whisper-large-v3-en":            0.02 * MilliTokensUsd,
+	"deepseek-r1-distill-qwen-32b":          0.69 * MilliTokensUsd,
+	"deepseek-r1-distill-llama-70b-specdec": 0.75 * MilliTokensUsd,
+	"deepseek-r1-distill-llama-70b":         0.75 * MilliTokensUsd,
 	// https://platform.lingyiwanwu.com/docs#-计费单元
-	"yi-34b-chat-0205": 2.5 / 1000 * RMB,
-	"yi-34b-chat-200k": 12.0 / 1000 * RMB,
-	"yi-vl-plus":       6.0 / 1000 * RMB,
+	"yi-34b-chat-0205": 2.5 * MilliRmb,
+	"yi-34b-chat-200k": 12.0 * MilliRmb,
+	"yi-vl-plus":       6.0 * MilliRmb,
 	// https://platform.stepfun.com/docs/pricing/details
-	"step-1-8k":    0.005 / 1000 * RMB,
-	"step-1-32k":   0.015 / 1000 * RMB,
-	"step-1-128k":  0.040 / 1000 * RMB,
-	"step-1-256k":  0.095 / 1000 * RMB,
-	"step-1-flash": 0.001 / 1000 * RMB,
-	"step-2-16k":   0.038 / 1000 * RMB,
-	"step-1v-8k":   0.005 / 1000 * RMB,
-	"step-1v-32k":  0.015 / 1000 * RMB,
+	"step-1-8k":    0.005 * MilliRmb,
+	"step-1-32k":   0.015 * MilliRmb,
+	"step-1-128k":  0.040 * MilliRmb,
+	"step-1-256k":  0.095 * MilliRmb,
+	"step-1-flash": 0.001 * MilliRmb,
+	"step-2-16k":   0.038 * MilliRmb,
+	"step-1v-8k":   0.005 * MilliRmb,
+	"step-1v-32k":  0.015 * MilliRmb,
 	// aws llama3 https://aws.amazon.com/cn/bedrock/pricing/
 	"llama3-8b-8192(33)":  0.0003 / 0.002,  // $0.0003 / 1K tokens
 	"llama3-70b-8192(33)": 0.00265 / 0.002, // $0.00265 / 1K tokens
@@ -332,100 +378,116 @@ var ModelRatio = map[string]float64{
 	"command-nightly":       0.5,
 	"command-light":         0.5,
 	"command-light-nightly": 0.5,
-	"command-r":             0.5 / 1000 * USD,
-	"command-r-plus":        3.0 / 1000 * USD,
+	"command-r":             0.5 * MilliTokensUsd,
+	"command-r-plus":        3.0 * MilliTokensUsd,
 	// https://platform.deepseek.com/api-docs/pricing/
-	"deepseek-chat":     0.14 * MILLI_USD,
-	"deepseek-reasoner": 0.55 * MILLI_USD,
+	"deepseek-chat":     0.27 * MilliTokensUsd,
+	"deepseek-reasoner": 0.55 * MilliTokensUsd,
 	// https://www.deepl.com/pro?cta=header-prices
-	"deepl-zh": 25.0 / 1000 * USD,
-	"deepl-en": 25.0 / 1000 * USD,
-	"deepl-ja": 25.0 / 1000 * USD,
+	"deepl-zh": 25.0 * MilliTokensUsd,
+	"deepl-en": 25.0 * MilliTokensUsd,
+	"deepl-ja": 25.0 * MilliTokensUsd,
 	// https://console.x.ai/
-	"grok-beta": 5.0 / 1000 * USD,
+	"grok-beta": 5.0 * MilliTokensUsd,
+	// vertex imagen3
+	// https://cloud.google.com/vertex-ai/generative-ai/pricing#imagen-models
+	"imagen-3.0-generate-001":      0.04 * ImageUsdPerPic,
+	"imagen-3.0-generate-002":      0.04 * ImageUsdPerPic,
+	"imagen-3.0-fast-generate-001": 0.02 * ImageUsdPerPic,
+	"imagen-3.0-capability-001":    0.04 * ImageUsdPerPic,
+	// -------------------------------------
 	// replicate charges based on the number of generated images
 	// https://replicate.com/pricing
-	"black-forest-labs/flux-1.1-pro":                0.04 * USD,
-	"black-forest-labs/flux-1.1-pro-ultra":          0.06 * USD,
-	"black-forest-labs/flux-canny-dev":              0.025 * USD,
-	"black-forest-labs/flux-canny-pro":              0.05 * USD,
-	"black-forest-labs/flux-depth-dev":              0.025 * USD,
-	"black-forest-labs/flux-depth-pro":              0.05 * USD,
-	"black-forest-labs/flux-dev":                    0.025 * USD,
-	"black-forest-labs/flux-dev-lora":               0.032 * USD,
-	"black-forest-labs/flux-fill-dev":               0.04 * USD,
-	"black-forest-labs/flux-fill-pro":               0.05 * USD,
-	"black-forest-labs/flux-pro":                    0.055 * USD,
-	"black-forest-labs/flux-redux-dev":              0.025 * USD,
-	"black-forest-labs/flux-redux-schnell":          0.003 * USD,
-	"black-forest-labs/flux-schnell":                0.003 * USD,
-	"black-forest-labs/flux-schnell-lora":           0.02 * USD,
-	"ideogram-ai/ideogram-v2":                       0.08 * USD,
-	"ideogram-ai/ideogram-v2-turbo":                 0.05 * USD,
-	"recraft-ai/recraft-v3":                         0.04 * USD,
-	"recraft-ai/recraft-v3-svg":                     0.08 * USD,
-	"stability-ai/stable-diffusion-3":               0.035 * USD,
-	"stability-ai/stable-diffusion-3.5-large":       0.065 * USD,
-	"stability-ai/stable-diffusion-3.5-large-turbo": 0.04 * USD,
-	"stability-ai/stable-diffusion-3.5-medium":      0.035 * USD,
+	// -------------------------------------
+	"black-forest-labs/flux-1.1-pro":                0.04 * ImageUsdPerPic,
+	"black-forest-labs/flux-1.1-pro-ultra":          0.06 * ImageUsdPerPic,
+	"black-forest-labs/flux-canny-dev":              0.025 * ImageUsdPerPic,
+	"black-forest-labs/flux-canny-pro":              0.05 * ImageUsdPerPic,
+	"black-forest-labs/flux-depth-dev":              0.025 * ImageUsdPerPic,
+	"black-forest-labs/flux-depth-pro":              0.05 * ImageUsdPerPic,
+	"black-forest-labs/flux-dev":                    0.025 * ImageUsdPerPic,
+	"black-forest-labs/flux-dev-lora":               0.032 * ImageUsdPerPic,
+	"black-forest-labs/flux-fill-dev":               0.04 * ImageUsdPerPic,
+	"black-forest-labs/flux-fill-pro":               0.05 * ImageUsdPerPic,
+	"black-forest-labs/flux-pro":                    0.055 * ImageUsdPerPic,
+	"black-forest-labs/flux-redux-dev":              0.025 * ImageUsdPerPic,
+	"black-forest-labs/flux-redux-schnell":          0.003 * ImageUsdPerPic,
+	"black-forest-labs/flux-schnell":                0.003 * ImageUsdPerPic,
+	"black-forest-labs/flux-schnell-lora":           0.02 * ImageUsdPerPic,
+	"ideogram-ai/ideogram-v2":                       0.08 * ImageUsdPerPic,
+	"ideogram-ai/ideogram-v2-turbo":                 0.05 * ImageUsdPerPic,
+	"recraft-ai/recraft-v3":                         0.04 * ImageUsdPerPic,
+	"recraft-ai/recraft-v3-svg":                     0.08 * ImageUsdPerPic,
+	"stability-ai/stable-diffusion-3":               0.035 * ImageUsdPerPic,
+	"stability-ai/stable-diffusion-3.5-large":       0.065 * ImageUsdPerPic,
+	"stability-ai/stable-diffusion-3.5-large-turbo": 0.04 * ImageUsdPerPic,
+	"stability-ai/stable-diffusion-3.5-medium":      0.035 * ImageUsdPerPic,
 	// replicate chat models
-	"ibm-granite/granite-20b-code-instruct-8k":  0.100 * USD,
-	"ibm-granite/granite-3.0-2b-instruct":       0.030 * USD,
-	"ibm-granite/granite-3.0-8b-instruct":       0.050 * USD,
-	"ibm-granite/granite-8b-code-instruct-128k": 0.050 * USD,
-	"meta/llama-2-13b":                          0.100 * USD,
-	"meta/llama-2-13b-chat":                     0.100 * USD,
-	"meta/llama-2-70b":                          0.650 * USD,
-	"meta/llama-2-70b-chat":                     0.650 * USD,
-	"meta/llama-2-7b":                           0.050 * USD,
-	"meta/llama-2-7b-chat":                      0.050 * USD,
-	"meta/meta-llama-3.1-405b-instruct":         9.500 * USD,
-	"meta/meta-llama-3-70b":                     0.650 * USD,
-	"meta/meta-llama-3-70b-instruct":            0.650 * USD,
-	"meta/meta-llama-3-8b":                      0.050 * USD,
-	"meta/meta-llama-3-8b-instruct":             0.050 * USD,
-	"mistralai/mistral-7b-instruct-v0.2":        0.050 * USD,
-	"mistralai/mistral-7b-v0.1":                 0.050 * USD,
-	"mistralai/mixtral-8x7b-instruct-v0.1":      0.300 * USD,
+	"anthropic/claude-3.5-haiku":                1.0 * MilliTokensUsd,
+	"anthropic/claude-3.5-sonnet":               3.75 * MilliTokensUsd,
+	"anthropic/claude-3.7-sonnet":               3.0 * MilliTokensUsd,
+	"deepseek-ai/deepseek-r1":                   10.0 * MilliTokensUsd,
+	"ibm-granite/granite-20b-code-instruct-8k":  0.100 * MilliTokensUsd,
+	"ibm-granite/granite-3.0-2b-instruct":       0.030 * MilliTokensUsd,
+	"ibm-granite/granite-3.0-8b-instruct":       0.050 * MilliTokensUsd,
+	"ibm-granite/granite-3.1-2b-instruct":       0.030 * MilliTokensUsd,
+	"ibm-granite/granite-3.1-8b-instruct":       0.030 * MilliTokensUsd,
+	"ibm-granite/granite-3.2-8b-instruct":       0.030 * MilliTokensUsd,
+	"ibm-granite/granite-8b-code-instruct-128k": 0.050 * MilliTokensUsd,
+	"meta/llama-2-13b":                          0.100 * MilliTokensUsd,
+	"meta/llama-2-13b-chat":                     0.100 * MilliTokensUsd,
+	"meta/llama-2-70b":                          0.650 * MilliTokensUsd,
+	"meta/llama-2-70b-chat":                     0.650 * MilliTokensUsd,
+	"meta/llama-2-7b":                           0.050 * MilliTokensUsd,
+	"meta/llama-2-7b-chat":                      0.050 * MilliTokensUsd,
+	"meta/meta-llama-3.1-405b-instruct":         9.500 * MilliTokensUsd,
+	"meta/meta-llama-3-70b":                     0.650 * MilliTokensUsd,
+	"meta/meta-llama-3-70b-instruct":            0.650 * MilliTokensUsd,
+	"meta/meta-llama-3-8b":                      0.050 * MilliTokensUsd,
+	"meta/meta-llama-3-8b-instruct":             0.050 * MilliTokensUsd,
+	"mistralai/mistral-7b-instruct-v0.2":        0.050 * MilliTokensUsd,
+	"mistralai/mistral-7b-v0.1":                 0.050 * MilliTokensUsd,
+	// -------------------------------------
 	//https://openrouter.ai/models
-	"01-ai/yi-large":                                  1.5,
-	"aetherwiing/mn-starcannon-12b":                   0.6,
-	"ai21/jamba-1-5-large":                            4.0,
-	"ai21/jamba-1-5-mini":                             0.2,
-	"ai21/jamba-instruct":                             0.35,
-	"aion-labs/aion-1.0":                              6.0,
-	"aion-labs/aion-1.0-mini":                         1.2,
-	"aion-labs/aion-rp-llama-3.1-8b":                  0.1,
-	"allenai/llama-3.1-tulu-3-405b":                   5.0,
-	"alpindale/goliath-120b":                          4.6875,
-	"alpindale/magnum-72b":                            1.125,
-	"amazon/nova-lite-v1":                             0.12,
-	"amazon/nova-micro-v1":                            0.07,
-	"amazon/nova-pro-v1":                              1.6,
-	"anthracite-org/magnum-v2-72b":                    1.5,
-	"anthracite-org/magnum-v4-72b":                    1.125,
-	"anthropic/claude-2":                              12.0,
-	"anthropic/claude-2.0":                            12.0,
-	"anthropic/claude-2.0:beta":                       12.0,
-	"anthropic/claude-2.1":                            12.0,
-	"anthropic/claude-2.1:beta":                       12.0,
-	"anthropic/claude-2:beta":                         12.0,
-	"anthropic/claude-3-haiku":                        0.625,
-	"anthropic/claude-3-haiku:beta":                   0.625,
-	"anthropic/claude-3-opus":                         37.5,
-	"anthropic/claude-3-opus:beta":                    37.5,
-	"anthropic/claude-3-sonnet":                       7.5,
-	"anthropic/claude-3-sonnet:beta":                  7.5,
-	"anthropic/claude-3.5-haiku":                      2.0,
-	"anthropic/claude-3.5-haiku-20241022":             2.0,
-	"anthropic/claude-3.5-haiku-20241022:beta":        2.0,
-	"anthropic/claude-3.5-haiku:beta":                 2.0,
-	"anthropic/claude-3.5-sonnet":                     7.5,
-	"anthropic/claude-3.5-sonnet-20240620":            7.5,
-	"anthropic/claude-3.5-sonnet-20240620:beta":       7.5,
-	"anthropic/claude-3.5-sonnet:beta":                7.5,
-	"cognitivecomputations/dolphin-mixtral-8x22b":     0.45,
-	"cognitivecomputations/dolphin-mixtral-8x7b":      0.25,
+	// -------------------------------------
+	"01-ai/yi-large":                 1.5,
+	"aetherwiing/mn-starcannon-12b":  0.6,
+	"ai21/jamba-1-5-large":           4.0,
+	"ai21/jamba-1-5-mini":            0.2,
+	"ai21/jamba-instruct":            0.35,
+	"aion-labs/aion-1.0":             6.0,
+	"aion-labs/aion-1.0-mini":        1.2,
+	"aion-labs/aion-rp-llama-3.1-8b": 0.1,
+	"allenai/llama-3.1-tulu-3-405b":  5.0,
+	"alpindale/goliath-120b":         4.6875,
+	"alpindale/magnum-72b":           1.125,
+	"amazon/nova-lite-v1":            0.12,
+	"amazon/nova-micro-v1":           0.07,
+	"amazon/nova-pro-v1":             1.6,
+	"anthracite-org/magnum-v2-72b":   1.5,
+	"anthracite-org/magnum-v4-72b":   1.125,
+	"anthropic/claude-2":             12.0,
+	"anthropic/claude-2.0":           12.0,
+	"anthropic/claude-2.0:beta":      12.0,
+	"anthropic/claude-2.1":           12.0,
+	"anthropic/claude-2.1:beta":      12.0,
+	"anthropic/claude-2:beta":        12.0,
+	"anthropic/claude-3-haiku":       0.625,
+	"anthropic/claude-3-haiku:beta":  0.625,
+	"anthropic/claude-3-opus":        37.5,
+	"anthropic/claude-3-opus:beta":   37.5,
+	"anthropic/claude-3-sonnet":      7.5,
+	"anthropic/claude-3-sonnet:beta": 7.5,
+	// "anthropic/claude-3.5-haiku":                      2.0,
+	"anthropic/claude-3.5-haiku-20241022":      2.0,
+	"anthropic/claude-3.5-haiku-20241022:beta": 2.0,
+	"anthropic/claude-3.5-haiku:beta":          2.0,
+	// "anthropic/claude-3.5-sonnet":                     7.5,
+	"anthropic/claude-3.5-sonnet-20240620":        7.5,
+	"anthropic/claude-3.5-sonnet-20240620:beta":   7.5,
+	"anthropic/claude-3.5-sonnet:beta":            7.5,
+	"cognitivecomputations/dolphin-mixtral-8x22b": 0.45,
+	"cognitivecomputations/dolphin-mixtral-8x7b":  0.25,
 	"cohere/command":                                  0.95,
 	"cohere/command-r":                                0.7125,
 	"cohere/command-r-03-2024":                        0.7125,
@@ -435,10 +497,10 @@ var ModelRatio = map[string]float64{
 	"cohere/command-r-plus-08-2024":                   4.75,
 	"cohere/command-r7b-12-2024":                      0.075,
 	"databricks/dbrx-instruct":                        0.6,
-	"deepseek/deepseek-chat":                          0.445,
+	"deepseek/deepseek-chat":                          1.25,
 	"deepseek/deepseek-chat-v2.5":                     1.0,
 	"deepseek/deepseek-chat:free":                     0.0,
-	"deepseek/deepseek-r1":                            1.2,
+	"deepseek/deepseek-r1":                            7,
 	"deepseek/deepseek-r1-distill-llama-70b":          0.345,
 	"deepseek/deepseek-r1-distill-llama-70b:free":     0.0,
 	"deepseek/deepseek-r1-distill-llama-8b":           0.02,
@@ -566,6 +628,7 @@ var ModelRatio = map[string]float64{
 	"openai/gpt-4o-mini":                              0.3,
 	"openai/gpt-4o-mini-2024-07-18":                   0.3,
 	"openai/gpt-4o:extended":                          9.0,
+	"openai/gpt-4.5-preview":                          75,
 	"openai/o1":                                       30.0,
 	"openai/o1-mini":                                  2.2,
 	"openai/o1-mini-2024-09-12":                       2.2,
@@ -575,61 +638,199 @@ var ModelRatio = map[string]float64{
 	"openai/o3-mini-high":                             2.2,
 	"openchat/openchat-7b":                            0.0275,
 	"openchat/openchat-7b:free":                       0.0,
-	"openrouter/auto":                                 -500000.0,
-	"perplexity/llama-3.1-sonar-huge-128k-online":     2.5,
-	"perplexity/llama-3.1-sonar-large-128k-chat":      0.5,
-	"perplexity/llama-3.1-sonar-large-128k-online":    0.5,
-	"perplexity/llama-3.1-sonar-small-128k-chat":      0.1,
-	"perplexity/llama-3.1-sonar-small-128k-online":    0.1,
-	"perplexity/sonar":                                0.5,
-	"perplexity/sonar-reasoning":                      2.5,
-	"pygmalionai/mythalion-13b":                       0.6,
-	"qwen/qvq-72b-preview":                            0.25,
-	"qwen/qwen-2-72b-instruct":                        0.45,
-	"qwen/qwen-2-7b-instruct":                         0.027,
-	"qwen/qwen-2-7b-instruct:free":                    0.0,
-	"qwen/qwen-2-vl-72b-instruct":                     0.2,
-	"qwen/qwen-2-vl-7b-instruct":                      0.05,
-	"qwen/qwen-2.5-72b-instruct":                      0.2,
-	"qwen/qwen-2.5-7b-instruct":                       0.025,
-	"qwen/qwen-2.5-coder-32b-instruct":                0.08,
-	"qwen/qwen-max":                                   3.2,
-	"qwen/qwen-plus":                                  0.6,
-	"qwen/qwen-turbo":                                 0.1,
-	"qwen/qwen-vl-plus:free":                          0.0,
-	"qwen/qwen2.5-vl-72b-instruct:free":               0.0,
-	"qwen/qwq-32b-preview":                            0.09,
-	"raifle/sorcererlm-8x22b":                         2.25,
-	"sao10k/fimbulvetr-11b-v2":                        0.6,
-	"sao10k/l3-euryale-70b":                           0.4,
-	"sao10k/l3-lunaris-8b":                            0.03,
-	"sao10k/l3.1-70b-hanami-x1":                       1.5,
-	"sao10k/l3.1-euryale-70b":                         0.4,
-	"sao10k/l3.3-euryale-70b":                         0.4,
-	"sophosympatheia/midnight-rose-70b":               0.4,
-	"sophosympatheia/rogue-rose-103b-v0.2:free":       0.0,
-	"teknium/openhermes-2.5-mistral-7b":               0.085,
-	"thedrummer/rocinante-12b":                        0.25,
-	"thedrummer/unslopnemo-12b":                       0.25,
-	"undi95/remm-slerp-l2-13b":                        0.6,
-	"undi95/toppy-m-7b":                               0.035,
-	"undi95/toppy-m-7b:free":                          0.0,
-	"x-ai/grok-2-1212":                                5.0,
-	"x-ai/grok-2-vision-1212":                         5.0,
-	"x-ai/grok-beta":                                  7.5,
-	"x-ai/grok-vision-beta":                           7.5,
-	"xwin-lm/xwin-lm-70b":                             1.875,
+	// "openrouter/auto":                                 -500000.0,
+	"perplexity/llama-3.1-sonar-huge-128k-online":  2.5,
+	"perplexity/llama-3.1-sonar-large-128k-chat":   0.5,
+	"perplexity/llama-3.1-sonar-large-128k-online": 0.5,
+	"perplexity/llama-3.1-sonar-small-128k-chat":   0.1,
+	"perplexity/llama-3.1-sonar-small-128k-online": 0.1,
+	"perplexity/sonar":                             0.5,
+	"perplexity/sonar-reasoning":                   2.5,
+	"pygmalionai/mythalion-13b":                    0.6,
+	"qwen/qvq-72b-preview":                         0.25,
+	"qwen/qwen-2-72b-instruct":                     0.45,
+	"qwen/qwen-2-7b-instruct":                      0.027,
+	"qwen/qwen-2-7b-instruct:free":                 0.0,
+	"qwen/qwen-2-vl-72b-instruct":                  0.2,
+	"qwen/qwen-2-vl-7b-instruct":                   0.05,
+	"qwen/qwen-2.5-72b-instruct":                   0.2,
+	"qwen/qwen-2.5-7b-instruct":                    0.025,
+	"qwen/qwen-2.5-coder-32b-instruct":             0.08,
+	"qwen/qwen-max":                                3.2,
+	"qwen/qwen-plus":                               0.6,
+	"qwen/qwen-turbo":                              0.1,
+	"qwen/qwen-vl-plus:free":                       0.0,
+	"qwen/qwen2.5-vl-72b-instruct:free":            0.0,
+	"qwen/qwq-32b-preview":                         0.09,
+	"raifle/sorcererlm-8x22b":                      2.25,
+	"sao10k/fimbulvetr-11b-v2":                     0.6,
+	"sao10k/l3-euryale-70b":                        0.4,
+	"sao10k/l3-lunaris-8b":                         0.03,
+	"sao10k/l3.1-70b-hanami-x1":                    1.5,
+	"sao10k/l3.1-euryale-70b":                      0.4,
+	"sao10k/l3.3-euryale-70b":                      0.4,
+	"sophosympatheia/midnight-rose-70b":            0.4,
+	"sophosympatheia/rogue-rose-103b-v0.2:free":    0.0,
+	"teknium/openhermes-2.5-mistral-7b":            0.085,
+	"thedrummer/rocinante-12b":                     0.25,
+	"thedrummer/unslopnemo-12b":                    0.25,
+	"undi95/remm-slerp-l2-13b":                     0.6,
+	"undi95/toppy-m-7b":                            0.035,
+	"undi95/toppy-m-7b:free":                       0.0,
+	"x-ai/grok-2-1212":                             5.0,
+	"x-ai/grok-2-vision-1212":                      5.0,
+	"x-ai/grok-beta":                               7.5,
+	"x-ai/grok-vision-beta":                        7.5,
+	"xwin-lm/xwin-lm-70b":                          1.875,
 }
 
+// CompletionRatio is the price ratio between completion tokens and prompt tokens
 var CompletionRatio = map[string]float64{
+	// -------------------------------------
 	// aws llama3
+	// -------------------------------------
 	"llama3-8b-8192(33)":  0.0006 / 0.0003,
 	"llama3-70b-8192(33)": 0.0035 / 0.00265,
+	// -------------------------------------
 	// whisper
-	"whisper-1": 0, // only count input tokens
+	// -------------------------------------
+	"whisper-1":                  0, // only count input tokens
+	"whisper-large-v3":           0, // only count input tokens
+	"whisper-large-v3-turbo":     0, // only count input tokens
+	"distil-whisper-large-v3-en": 0, // only count input tokens
+	// -------------------------------------
 	// deepseek
-	"deepseek-chat":     0.28 / 0.14,
+	// -------------------------------------
+	"deepseek-chat":     1.1 / 0.27,
 	"deepseek-reasoner": 2.19 / 0.55,
+	// -------------------------------------
+	// openrouter
+	// -------------------------------------
+	"deepseek/deepseek-chat": 1,
+	"deepseek/deepseek-r1":   1,
+	// -------------------------------------
+	// groq
+	// -------------------------------------
+	"llama-3.3-70b-versatile":               0.79 / 0.59,
+	"llama-3.1-8b-instant":                  0.08 / 0.05,
+	"llama3-70b-8192":                       0.79 / 0.59,
+	"llama3-8b-8192":                        0.08 / 0.05,
+	"gemma2-9b-it":                          1.0,
+	"llama-3.2-11b-text-preview":            1.0,
+	"llama-3.2-11b-vision-preview":          1.0,
+	"llama-3.2-1b-preview":                  1.0,
+	"llama-3.2-3b-preview":                  1.0,
+	"llama-3.2-90b-text-preview":            1.0,
+	"llama-3.2-90b-vision-preview":          1.0,
+	"llama-guard-3-8b":                      1.0,
+	"llama3-groq-70b-8192-tool-use-preview": 0.79 / 0.59,
+	"llama3-groq-8b-8192-tool-use-preview":  0.08 / 0.05,
+	"mixtral-8x7b-32768":                    1.0,
+	"deepseek-r1-distill-qwen-32b":          1.0,
+	"deepseek-r1-distill-llama-70b-specdec": 0.99 / 0.75,
+	"deepseek-r1-distill-llama-70b":         0.99 / 0.75,
+	"llama-3.3-70b-specdec":                 0.99 / 0.59,
+	"mistral-saba-24b":                      1.0,
+	"qwen-qwq-32b":                          0.39 / 0.29,
+	"qwen-2.5-coder-32b":                    1.0,
+	"qwen-2.5-32b":                          1.0,
+	// -------------------------------------
+	// Replicate
+	// -------------------------------------
+	"anthropic/claude-3.5-haiku":                5.0 / 1.0,
+	"anthropic/claude-3.5-sonnet":               18.75 / 3.75,
+	"anthropic/claude-3.7-sonnet":               15.0 / 3.0,
+	"deepseek-ai/deepseek-r1":                   10.0 / 10.0,
+	"ibm-granite/granite-20b-code-instruct-8k":  0.5 / 0.1,
+	"ibm-granite/granite-3.0-2b-instruct":       0.25 / 0.03,
+	"ibm-granite/granite-3.0-8b-instruct":       0.25 / 0.05,
+	"ibm-granite/granite-3.1-2b-instruct":       0.25 / 0.03,
+	"ibm-granite/granite-3.1-8b-instruct":       0.25 / 0.03,
+	"ibm-granite/granite-3.2-8b-instruct":       0.25 / 0.03,
+	"ibm-granite/granite-8b-code-instruct-128k": 0.25 / 0.05,
+	"meta/llama-2-13b":                          0.5 / 0.1,
+	"meta/llama-2-13b-chat":                     0.5 / 0.1,
+	"meta/llama-2-70b":                          2.75 / 0.65,
+	"meta/llama-2-70b-chat":                     2.75 / 0.65,
+	"meta/llama-2-7b":                           0.25 / 0.05,
+	"meta/llama-2-7b-chat":                      0.25 / 0.05,
+	"meta/meta-llama-3.1-405b-instruct":         9.5 / 9.5,
+	"meta/meta-llama-3-70b":                     2.75 / 0.65,
+	"meta/meta-llama-3-70b-instruct":            2.75 / 0.65,
+	"meta/meta-llama-3-8b":                      0.25 / 0.05,
+	"meta/meta-llama-3-8b-instruct":             0.25 / 0.05,
+	"mistralai/mistral-7b-instruct-v0.2":        0.25 / 0.05,
+	"mistralai/mistral-7b-v0.1":                 0.25 / 0.05,
+}
+
+// AudioRatio represents the price ratio between audio tokens and text tokens
+var AudioRatio = map[string]float64{
+	"gpt-4o-audio-preview":                 16,
+	"gpt-4o-audio-preview-2024-12-17":      16,
+	"gpt-4o-audio-preview-2024-10-01":      40,
+	"gpt-4o-mini-audio-preview":            10 / 0.15,
+	"gpt-4o-mini-audio-preview-2024-12-17": 10 / 0.15,
+}
+
+// GetAudioPromptRatio returns the audio prompt ratio for the given model.
+func GetAudioPromptRatio(actualModelName string) float64 {
+	var v float64
+	if ratio, ok := AudioRatio[actualModelName]; ok {
+		v = ratio
+	} else {
+		v = 16
+	}
+
+	return v
+}
+
+// AudioCompletionRatio is the completion ratio for audio models.
+var AudioCompletionRatio = map[string]float64{
+	"whisper-1":                            0,
+	"gpt-4o-audio-preview":                 2,
+	"gpt-4o-audio-preview-2024-12-17":      2,
+	"gpt-4o-audio-preview-2024-10-01":      2,
+	"gpt-4o-mini-audio-preview":            2,
+	"gpt-4o-mini-audio-preview-2024-12-17": 2,
+}
+
+// GetAudioCompletionRatio returns the completion ratio for audio models.
+func GetAudioCompletionRatio(actualModelName string) float64 {
+	var v float64
+	if ratio, ok := AudioCompletionRatio[actualModelName]; ok {
+		v = ratio
+	} else {
+		v = 2
+	}
+
+	return v
+}
+
+// AudioTokensPerSecond is the number of audio tokens per second for each model.
+var AudioPromptTokensPerSecond = map[string]float64{
+	// Whisper API price is $0.0001/sec. One-api's historical ratio is 15,
+	// corresponding to $0.03/kilo_tokens.
+	// After conversion, tokens per second should be 0.0001/0.03*1000 = 3.3333.
+	"whisper-1": 0.0001 / 0.03 * 1000,
+	// gpt-4o-audio series processes 10 tokens per second
+	"gpt-4o-audio-preview":                 10,
+	"gpt-4o-audio-preview-2024-12-17":      10,
+	"gpt-4o-audio-preview-2024-10-01":      10,
+	"gpt-4o-mini-audio-preview":            10,
+	"gpt-4o-mini-audio-preview-2024-12-17": 10,
+}
+
+// GetAudioPromptTokensPerSecond returns the number of audio tokens per second
+// for the given model.
+func GetAudioPromptTokensPerSecond(actualModelName string) float64 {
+	var v float64
+	if tokensPerSecond, ok := AudioPromptTokensPerSecond[actualModelName]; ok {
+		v = tokensPerSecond
+	} else {
+		v = 10
+	}
+
+	return v
 }
 
 var (
@@ -676,11 +877,18 @@ func ModelRatio2JSONString() string {
 	return string(jsonBytes)
 }
 
+// UpdateModelRatioByJSONString updates the ModelRatio map with the given JSON string.
 func UpdateModelRatioByJSONString(jsonStr string) error {
 	modelRatioLock.Lock()
 	defer modelRatioLock.Unlock()
 	ModelRatio = make(map[string]float64)
-	return json.Unmarshal([]byte(jsonStr), &ModelRatio)
+	err := json.Unmarshal([]byte(jsonStr), &ModelRatio)
+	if err != nil {
+		logger.SysError("error unmarshalling model ratio: " + err.Error())
+		return err
+	}
+
+	return nil
 }
 
 func GetModelRatio(name string, channelType int) float64 {
@@ -692,23 +900,26 @@ func GetModelRatio(name string, channelType int) float64 {
 	if strings.HasPrefix(name, "command-") && strings.HasSuffix(name, "-internet") {
 		name = strings.TrimSuffix(name, "-internet")
 	}
+
 	model := fmt.Sprintf("%s(%d)", name, channelType)
-	if ratio, ok := ModelRatio[model]; ok {
-		return ratio
-	}
-	if ratio, ok := DefaultModelRatio[model]; ok {
-		return ratio
-	}
-	if ratio, ok := ModelRatio[name]; ok {
-		return ratio
-	}
-	if ratio, ok := DefaultModelRatio[name]; ok {
-		return ratio
+
+	for _, targetName := range []string{model, name} {
+		for _, ratioMap := range []map[string]float64{
+			ModelRatio,
+			DefaultModelRatio,
+			AudioRatio,
+		} {
+			if ratio, ok := ratioMap[targetName]; ok {
+				return ratio
+			}
+		}
 	}
+
 	logger.SysError("model ratio not found: " + name)
-	return 30
+	return 2.5 * MilliTokensUsd
 }
 
+// CompletionRatio2JSONString returns the CompletionRatio map as a JSON string.
 func CompletionRatio2JSONString() string {
 	jsonBytes, err := json.Marshal(CompletionRatio)
 	if err != nil {
@@ -717,59 +928,79 @@ func CompletionRatio2JSONString() string {
 	return string(jsonBytes)
 }
 
+// completionRatioLock is a mutex for synchronizing access to the CompletionRatio map.
+var completionRatioLock sync.RWMutex
+
+// UpdateCompletionRatioByJSONString updates the CompletionRatio map with the given JSON string.
 func UpdateCompletionRatioByJSONString(jsonStr string) error {
+	completionRatioLock.Lock()
+	defer completionRatioLock.Unlock()
 	CompletionRatio = make(map[string]float64)
 	return json.Unmarshal([]byte(jsonStr), &CompletionRatio)
 }
 
+// GetCompletionRatio returns the completion ratio for the given model name and channel type.
 func GetCompletionRatio(name string, channelType int) float64 {
+	completionRatioLock.RLock()
+	defer completionRatioLock.RUnlock()
 	if strings.HasPrefix(name, "qwen-") && strings.HasSuffix(name, "-internet") {
 		name = strings.TrimSuffix(name, "-internet")
 	}
 	model := fmt.Sprintf("%s(%d)", name, channelType)
-	if ratio, ok := CompletionRatio[model]; ok {
-		return ratio
+
+	name = strings.TrimPrefix(name, "openai/")
+	for _, targetName := range []string{model, name} {
+		for _, ratioMap := range []map[string]float64{
+			CompletionRatio,
+			DefaultCompletionRatio,
+			AudioCompletionRatio,
+		} {
+			// first try the model name
+			if ratio, ok := ratioMap[targetName]; ok {
+				return ratio
+			}
+
+			// then try the model name without some special prefix
+			normalizedTargetName := strings.TrimPrefix(targetName, "openai/")
+			if ratio, ok := ratioMap[normalizedTargetName]; ok {
+				return ratio
+			}
+		}
 	}
-	if ratio, ok := DefaultCompletionRatio[model]; ok {
-		return ratio
-	}
-	if ratio, ok := CompletionRatio[name]; ok {
-		return ratio
-	}
-	if ratio, ok := DefaultCompletionRatio[name]; ok {
-		return ratio
-	}
-	if strings.HasPrefix(name, "gpt-3.5") {
-		if name == "gpt-3.5-turbo" || strings.HasSuffix(name, "0125") {
+
+	// openai
+	switch {
+	case strings.HasPrefix(name, "gpt-3.5"):
+		switch {
+		case name == "gpt-3.5-turbo" || strings.HasSuffix(name, "0125"):
 			// https://openai.com/blog/new-embedding-models-and-api-updates
 			// Updated GPT-3.5 Turbo model and lower pricing
 			return 3
-		}
-		if strings.HasSuffix(name, "1106") {
+		case strings.HasSuffix(name, "1106"):
 			return 2
+		default:
+			return 4.0 / 3.0
 		}
-		return 4.0 / 3.0
-	}
-	if strings.HasPrefix(name, "gpt-4") {
-		if strings.HasPrefix(name, "gpt-4o") {
+	case name == "chatgpt-4o-latest":
+		return 3
+	case strings.HasPrefix(name, "gpt-4"):
+		switch {
+		case strings.HasPrefix(name, "gpt-4o"):
 			if name == "gpt-4o-2024-05-13" {
 				return 3
 			}
 			return 4
-		}
-		if strings.HasPrefix(name, "gpt-4-turbo") ||
-			strings.HasSuffix(name, "preview") {
+		case strings.HasPrefix(name, "gpt-4-"):
 			return 3
+		default:
+			return 2
 		}
-		return 2
-	}
-	// including o1, o1-preview, o1-mini
-	if strings.HasPrefix(name, "o1") {
+	// including o1/o1-preview/o1-mini
+	case strings.HasPrefix(name, "o1") ||
+		strings.HasPrefix(name, "o3"):
 		return 4
 	}
-	if name == "chatgpt-4o-latest" {
-		return 3
-	}
+
 	if strings.HasPrefix(name, "claude-3") {
 		return 5
 	}
@@ -780,10 +1011,7 @@ func GetCompletionRatio(name string, channelType int) float64 {
 		return 3
 	}
 	if strings.HasPrefix(name, "gemini-") {
-		return 3
-	}
-	if strings.HasPrefix(name, "deepseek-") {
-		return 2
+		return 4
 	}
 
 	switch name {
@@ -831,5 +1059,6 @@ func GetCompletionRatio(name string, channelType int) float64 {
 		return 1.000 / 0.300 // ≈3.333333
 	}
 
+	logger.SysWarn(fmt.Sprintf("completion ratio not found for model: %s (channel type: %d), using default value 1", name, channelType))
 	return 1
 }
diff --git a/relay/controller/helper.go b/relay/controller/helper.go
index 5b6f023f..1f352bea 100644
--- a/relay/controller/helper.go
+++ b/relay/controller/helper.go
@@ -8,18 +8,16 @@ import (
 	"net/http"
 	"strings"
 
-	"github.com/songquanpeng/one-api/common/helper"
-	"github.com/songquanpeng/one-api/relay/constant/role"
-
 	"github.com/gin-gonic/gin"
-
 	"github.com/songquanpeng/one-api/common"
 	"github.com/songquanpeng/one-api/common/config"
+	"github.com/songquanpeng/one-api/common/helper"
 	"github.com/songquanpeng/one-api/common/logger"
 	"github.com/songquanpeng/one-api/model"
 	"github.com/songquanpeng/one-api/relay/adaptor/openai"
 	billingratio "github.com/songquanpeng/one-api/relay/billing/ratio"
 	"github.com/songquanpeng/one-api/relay/channeltype"
+	"github.com/songquanpeng/one-api/relay/constant/role"
 	"github.com/songquanpeng/one-api/relay/controller/validator"
 	"github.com/songquanpeng/one-api/relay/meta"
 	relaymodel "github.com/songquanpeng/one-api/relay/model"
@@ -45,10 +43,10 @@ func getAndValidateTextRequest(c *gin.Context, relayMode int) (*relaymodel.Gener
 	return textRequest, nil
 }
 
-func getPromptTokens(textRequest *relaymodel.GeneralOpenAIRequest, relayMode int) int {
+func getPromptTokens(ctx context.Context, textRequest *relaymodel.GeneralOpenAIRequest, relayMode int) int {
 	switch relayMode {
 	case relaymode.ChatCompletions:
-		return openai.CountTokenMessages(textRequest.Messages, textRequest.Model)
+		return openai.CountTokenMessages(ctx, textRequest.Messages, textRequest.Model)
 	case relaymode.Completions:
 		return openai.CountTokenInput(textRequest.Prompt, textRequest.Model)
 	case relaymode.Moderations:
@@ -94,19 +92,30 @@ func preConsumeQuota(ctx context.Context, textRequest *relaymodel.GeneralOpenAIR
 	return preConsumedQuota, nil
 }
 
-func postConsumeQuota(ctx context.Context, usage *relaymodel.Usage, meta *meta.Meta, textRequest *relaymodel.GeneralOpenAIRequest, ratio float64, preConsumedQuota int64, modelRatio float64, groupRatio float64, systemPromptReset bool) {
+func postConsumeQuota(ctx context.Context,
+	usage *relaymodel.Usage,
+	meta *meta.Meta,
+	textRequest *relaymodel.GeneralOpenAIRequest,
+	ratio float64,
+	preConsumedQuota int64,
+	modelRatio float64,
+	groupRatio float64,
+	systemPromptReset bool) (quota int64) {
 	if usage == nil {
 		logger.Error(ctx, "usage is nil, which is unexpected")
 		return
 	}
-	var quota int64
 	completionRatio := billingratio.GetCompletionRatio(textRequest.Model, meta.ChannelType)
 	promptTokens := usage.PromptTokens
+	// It appears that DeepSeek's official service automatically merges ReasoningTokens into CompletionTokens,
+	// but the behavior of third-party providers may differ, so for now we do not add them manually.
+	// completionTokens := usage.CompletionTokens + usage.CompletionTokensDetails.ReasoningTokens
 	completionTokens := usage.CompletionTokens
-	quota = int64(math.Ceil((float64(promptTokens) + float64(completionTokens)*completionRatio) * ratio))
+	quota = int64(math.Ceil((float64(promptTokens)+float64(completionTokens)*completionRatio)*ratio)) + usage.ToolsCost
 	if ratio != 0 && quota <= 0 {
 		quota = 1
 	}
+
 	totalTokens := promptTokens + completionTokens
 	if totalTokens == 0 {
 		// in this case, must be some error happened
@@ -122,7 +131,13 @@ func postConsumeQuota(ctx context.Context, usage *relaymodel.Usage, meta *meta.M
 	if err != nil {
 		logger.Error(ctx, "error update user quota cache: "+err.Error())
 	}
-	logContent := fmt.Sprintf("倍率：%.2f × %.2f × %.2f", modelRatio, groupRatio, completionRatio)
+
+	var logContent string
+	if usage.ToolsCost == 0 {
+		logContent = fmt.Sprintf("倍率：%.2f × %.2f × %.2f", modelRatio, groupRatio, completionRatio)
+	} else {
+		logContent = fmt.Sprintf("倍率：%.2f × %.2f × %.2f, tools cost %d", modelRatio, groupRatio, completionRatio, usage.ToolsCost)
+	}
 	model.RecordConsumeLog(ctx, &model.Log{
 		UserId:            meta.UserId,
 		ChannelId:         meta.ChannelId,
@@ -138,6 +153,8 @@ func postConsumeQuota(ctx context.Context, usage *relaymodel.Usage, meta *meta.M
 	})
 	model.UpdateUserUsedQuotaAndRequestCount(meta.UserId, quota)
 	model.UpdateChannelUsedQuota(meta.ChannelId, quota)
+
+	return quota
 }
 
 func getMappedModelName(modelName string, mapping map[string]string) (string, bool) {
diff --git a/relay/controller/image.go b/relay/controller/image.go
index 9a980a14..b1344116 100644
--- a/relay/controller/image.go
+++ b/relay/controller/image.go
@@ -157,7 +157,7 @@ func RelayImageHelper(c *gin.Context, relayMode int) *relaymodel.ErrorWithStatus
 		channeltype.Ali,
 		channeltype.Replicate,
 		channeltype.Baidu:
-		finalRequest, err := adaptor.ConvertImageRequest(imageRequest)
+		finalRequest, err := adaptor.ConvertImageRequest(c, imageRequest)
 		if err != nil {
 			return openai.ErrorWrapper(err, "convert_image_request_failed", http.StatusInternalServerError)
 		}
diff --git a/relay/controller/text.go b/relay/controller/text.go
index f912498a..dbb47115 100644
--- a/relay/controller/text.go
+++ b/relay/controller/text.go
@@ -10,6 +10,7 @@ import (
 	"github.com/gin-gonic/gin"
 
 	"github.com/songquanpeng/one-api/common/config"
+	"github.com/songquanpeng/one-api/common/ctxkey"
 	"github.com/songquanpeng/one-api/common/logger"
 	"github.com/songquanpeng/one-api/relay"
 	"github.com/songquanpeng/one-api/relay/adaptor"
@@ -44,7 +45,7 @@ func RelayTextHelper(c *gin.Context) *model.ErrorWithStatusCode {
 	groupRatio := billingratio.GetGroupRatio(meta.Group)
 	ratio := modelRatio * groupRatio
 	// pre-consume quota
-	promptTokens := getPromptTokens(textRequest, meta.Mode)
+	promptTokens := getPromptTokens(c.Request.Context(), textRequest, meta.Mode)
 	meta.PromptTokens = promptTokens
 	preConsumedQuota, bizErr := preConsumeQuota(ctx, textRequest, promptTokens, ratio, meta)
 	if bizErr != nil {
@@ -104,6 +105,8 @@ func getRequestBody(c *gin.Context, meta *meta.Meta, textRequest *model.GeneralO
 		logger.Debugf(c.Request.Context(), "converted request failed: %s\n", err.Error())
 		return nil, err
 	}
+	c.Set(ctxkey.ConvertedRequest, convertedRequest)
+
 	jsonData, err := json.Marshal(convertedRequest)
 	if err != nil {
 		logger.Debugf(c.Request.Context(), "converted request json_marshal_failed: %s\n", err.Error())
diff --git a/relay/model/general.go b/relay/model/general.go
index 5f5968c8..f264cf07 100644
--- a/relay/model/general.go
+++ b/relay/model/general.go
@@ -1,5 +1,7 @@
 package model
 
+import "github.com/songquanpeng/one-api/relay/adaptor/openrouter"
+
 type ResponseFormat struct {
 	Type       string      `json:"type,omitempty"`
 	JsonSchema *JSONSchema `json:"json_schema,omitempty"`
@@ -23,49 +25,103 @@ type StreamOptions struct {
 
 type GeneralOpenAIRequest struct {
 	// https://platform.openai.com/docs/api-reference/chat/create
-	Messages            []Message       `json:"messages,omitempty"`
-	Model               string          `json:"model,omitempty"`
-	Store               *bool           `json:"store,omitempty"`
-	ReasoningEffort     *string         `json:"reasoning_effort,omitempty"`
-	Metadata            any             `json:"metadata,omitempty"`
-	FrequencyPenalty    *float64        `json:"frequency_penalty,omitempty"`
-	LogitBias           any             `json:"logit_bias,omitempty"`
-	Logprobs            *bool           `json:"logprobs,omitempty"`
-	TopLogprobs         *int            `json:"top_logprobs,omitempty"`
-	MaxTokens           int             `json:"max_tokens,omitempty"`
-	MaxCompletionTokens *int            `json:"max_completion_tokens,omitempty"`
-	N                   int             `json:"n,omitempty"`
-	Modalities          []string        `json:"modalities,omitempty"`
-	Prediction          any             `json:"prediction,omitempty"`
-	Audio               *Audio          `json:"audio,omitempty"`
-	PresencePenalty     *float64        `json:"presence_penalty,omitempty"`
-	ResponseFormat      *ResponseFormat `json:"response_format,omitempty"`
-	Seed                float64         `json:"seed,omitempty"`
-	ServiceTier         *string         `json:"service_tier,omitempty"`
-	Stop                any             `json:"stop,omitempty"`
-	Stream              bool            `json:"stream,omitempty"`
-	StreamOptions       *StreamOptions  `json:"stream_options,omitempty"`
-	Temperature         *float64        `json:"temperature,omitempty"`
-	TopP                *float64        `json:"top_p,omitempty"`
-	TopK                int             `json:"top_k,omitempty"`
-	Tools               []Tool          `json:"tools,omitempty"`
-	ToolChoice          any             `json:"tool_choice,omitempty"`
-	ParallelTooCalls    *bool           `json:"parallel_tool_calls,omitempty"`
-	User                string          `json:"user,omitempty"`
-	FunctionCall        any             `json:"function_call,omitempty"`
-	Functions           any             `json:"functions,omitempty"`
+	Messages []Message `json:"messages,omitempty"`
+	Model    string    `json:"model,omitempty"`
+	Store    *bool     `json:"store,omitempty"`
+	Metadata any       `json:"metadata,omitempty"`
+	// FrequencyPenalty is a number between -2.0 and 2.0 that penalizes
+	// new tokens based on their existing frequency in the text so far,
+	// default is 0.
+	FrequencyPenalty    *float64 `json:"frequency_penalty,omitempty" binding:"omitempty,min=-2,max=2"`
+	LogitBias           any      `json:"logit_bias,omitempty"`
+	Logprobs            *bool    `json:"logprobs,omitempty"`
+	TopLogprobs         *int     `json:"top_logprobs,omitempty"`
+	MaxTokens           int      `json:"max_tokens,omitempty"`
+	MaxCompletionTokens *int     `json:"max_completion_tokens,omitempty"`
+	// N is how many chat completion choices to generate for each input message,
+	// default to 1.
+	N *int `json:"n,omitempty" binding:"omitempty,min=0"`
+	// ReasoningEffort constrains effort on reasoning for reasoning models, reasoning models only.
+	ReasoningEffort *string `json:"reasoning_effort,omitempty" binding:"omitempty,oneof=low medium high"`
+	// Modalities currently the model only programmatically allows modalities = [“text”, “audio”]
+	Modalities []string `json:"modalities,omitempty"`
+	Prediction any      `json:"prediction,omitempty"`
+	Audio      *Audio   `json:"audio,omitempty"`
+	// PresencePenalty is a number between -2.0 and 2.0 that penalizes
+	// new tokens based on whether they appear in the text so far, default is 0.
+	PresencePenalty  *float64        `json:"presence_penalty,omitempty" binding:"omitempty,min=-2,max=2"`
+	ResponseFormat   *ResponseFormat `json:"response_format,omitempty"`
+	Seed             float64         `json:"seed,omitempty"`
+	ServiceTier      *string         `json:"service_tier,omitempty" binding:"omitempty,oneof=default auto"`
+	Stop             any             `json:"stop,omitempty"`
+	Stream           bool            `json:"stream,omitempty"`
+	StreamOptions    *StreamOptions  `json:"stream_options,omitempty"`
+	Temperature      *float64        `json:"temperature,omitempty"`
+	TopP             *float64        `json:"top_p,omitempty"`
+	TopK             int             `json:"top_k,omitempty"`
+	Tools            []Tool          `json:"tools,omitempty"`
+	ToolChoice       any             `json:"tool_choice,omitempty"`
+	ParallelTooCalls *bool           `json:"parallel_tool_calls,omitempty"`
+	User             string          `json:"user,omitempty"`
+	FunctionCall     any             `json:"function_call,omitempty"`
+	Functions        any             `json:"functions,omitempty"`
 	// https://platform.openai.com/docs/api-reference/embeddings/create
 	Input          any    `json:"input,omitempty"`
 	EncodingFormat string `json:"encoding_format,omitempty"`
 	Dimensions     int    `json:"dimensions,omitempty"`
 	// https://platform.openai.com/docs/api-reference/images/create
-	Prompt  any     `json:"prompt,omitempty"`
-	Quality *string `json:"quality,omitempty"`
-	Size    string  `json:"size,omitempty"`
-	Style   *string `json:"style,omitempty"`
+	Prompt           string            `json:"prompt,omitempty"`
+	Quality          *string           `json:"quality,omitempty"`
+	Size             string            `json:"size,omitempty"`
+	Style            *string           `json:"style,omitempty"`
+	WebSearchOptions *WebSearchOptions `json:"web_search_options,omitempty"`
+
 	// Others
 	Instruction string `json:"instruction,omitempty"`
 	NumCtx      int    `json:"num_ctx,omitempty"`
+	// -------------------------------------
+	// Openrouter
+	// -------------------------------------
+	Provider         *openrouter.RequestProvider `json:"provider,omitempty"`
+	IncludeReasoning *bool                       `json:"include_reasoning,omitempty"`
+	// -------------------------------------
+	// Anthropic
+	// -------------------------------------
+	Thinking *Thinking `json:"thinking,omitempty"`
+}
+
+// WebSearchOptions is the tool searches the web for relevant results to use in a response.
+type WebSearchOptions struct {
+	// SearchContextSize is the high level guidance for the amount of context window space to use for the search,
+	// default is "medium".
+	SearchContextSize *string       `json:"search_context_size,omitempty" binding:"omitempty,oneof=low medium high"`
+	UserLocation      *UserLocation `json:"user_location,omitempty"`
+}
+
+// UserLocation is a struct that contains the location of the user.
+type UserLocation struct {
+	// Approximate is the approximate location parameters for the search.
+	Approximate UserLocationApproximate `json:"approximate" binding:"required"`
+	// Type is the type of location approximation.
+	Type string `json:"type" binding:"required,oneof=approximate"`
+}
+
+// UserLocationApproximate is a struct that contains the approximate location of the user.
+type UserLocationApproximate struct {
+	// City is the city of the user, e.g. San Francisco.
+	City *string `json:"city,omitempty"`
+	// Country is the country of the user, e.g. US.
+	Country *string `json:"country,omitempty"`
+	// Region is the region of the user, e.g. California.
+	Region *string `json:"region,omitempty"`
+	// Timezone is the IANA timezone of the user, e.g. America/Los_Angeles.
+	Timezone *string `json:"timezone,omitempty"`
+}
+
+// https://docs.anthropic.com/en/docs/build-with-claude/extended-thinking#implementing-extended-thinking
+type Thinking struct {
+	Type         string `json:"type"`
+	BudgetTokens int    `json:"budget_tokens" binding:"omitempty,min=1024"`
 }
 
 func (r GeneralOpenAIRequest) ParseInput() []string {
diff --git a/relay/model/message.go b/relay/model/message.go
index 5ff7b7ae..597ee693 100644
--- a/relay/model/message.go
+++ b/relay/model/message.go
@@ -1,12 +1,106 @@
 package model
 
+import (
+	"context"
+	"strings"
+
+	"github.com/songquanpeng/one-api/common/logger"
+)
+
+// ReasoningFormat is the format of reasoning content,
+// can be set by the reasoning_format parameter in the request url.
+type ReasoningFormat string
+
+const (
+	ReasoningFormatUnspecified ReasoningFormat = ""
+	// ReasoningFormatReasoningContent is the reasoning format used by deepseek official API
+	ReasoningFormatReasoningContent ReasoningFormat = "reasoning_content"
+	// ReasoningFormatReasoning is the reasoning format used by openrouter
+	ReasoningFormatReasoning ReasoningFormat = "reasoning"
+
+	// ReasoningFormatThinkTag is the reasoning format used by 3rd party deepseek-r1 providers.
+	//
+	// Deprecated: I believe <think> is a very poor format, especially in stream mode, it is difficult to extract and convert.
+	// Considering that only a few deepseek-r1 third-party providers use this format, it has been decided to no longer support it.
+	// ReasoningFormatThinkTag ReasoningFormat = "think-tag"
+
+	// ReasoningFormatThinking is the reasoning format used by anthropic
+	ReasoningFormatThinking ReasoningFormat = "thinking"
+)
+
 type Message struct {
-	Role             string  `json:"role,omitempty"`
-	Content          any     `json:"content,omitempty"`
-	ReasoningContent any     `json:"reasoning_content,omitempty"`
-	Name             *string `json:"name,omitempty"`
-	ToolCalls        []Tool  `json:"tool_calls,omitempty"`
-	ToolCallId       string  `json:"tool_call_id,omitempty"`
+	Role string `json:"role,omitempty"`
+	// Content is a string or a list of objects
+	Content    any              `json:"content,omitempty"`
+	Name       *string          `json:"name,omitempty"`
+	ToolCalls  []Tool           `json:"tool_calls,omitempty"`
+	ToolCallId string           `json:"tool_call_id,omitempty"`
+	Audio      *messageAudio    `json:"audio,omitempty"`
+	Annotation []AnnotationItem `json:"annotation,omitempty"`
+
+	// -------------------------------------
+	// Deepseek 专有的一些字段
+	// https://api-docs.deepseek.com/api/create-chat-completion
+	// -------------------------------------
+	// Prefix forces the model to begin its answer with the supplied prefix in the assistant message.
+	// To enable this feature, set base_url to "https://api.deepseek.com/beta".
+	Prefix *bool `json:"prefix,omitempty"` // ReasoningContent is Used for the deepseek-reasoner model in the Chat
+	// Prefix Completion feature as the input for the CoT in the last assistant message.
+	// When using this feature, the prefix parameter must be set to true.
+	ReasoningContent *string `json:"reasoning_content,omitempty"`
+
+	// -------------------------------------
+	// Openrouter
+	// -------------------------------------
+	Reasoning *string `json:"reasoning,omitempty"`
+	Refusal   *bool   `json:"refusal,omitempty"`
+
+	// -------------------------------------
+	// Anthropic
+	// -------------------------------------
+	Thinking  *string `json:"thinking,omitempty"`
+	Signature *string `json:"signature,omitempty"`
+}
+
+type AnnotationItem struct {
+	Type        string      `json:"type" binding:"oneof=url_citation"`
+	UrlCitation UrlCitation `json:"url_citation"`
+}
+
+// UrlCitation is a URL citation when using web search.
+type UrlCitation struct {
+	// Endpoint is the index of the last character of the URL citation in the message.
+	EndIndex int `json:"end_index"`
+	// StartIndex is the index of the first character of the URL citation in the message.
+	StartIndex int `json:"start_index"`
+	// Title is the title of the web resource.
+	Title string `json:"title"`
+	// Url is the URL of the web resource.
+	Url string `json:"url"`
+}
+
+// SetReasoningContent sets the reasoning content based on the format
+func (m *Message) SetReasoningContent(format string, reasoningContent string) {
+	switch ReasoningFormat(strings.ToLower(strings.TrimSpace(format))) {
+	case ReasoningFormatReasoningContent:
+		m.ReasoningContent = &reasoningContent
+		// case ReasoningFormatThinkTag:
+		// 	m.Content = fmt.Sprintf("<think>%s</think>%s", reasoningContent, m.Content)
+	case ReasoningFormatThinking:
+		m.Thinking = &reasoningContent
+	case ReasoningFormatReasoning,
+		ReasoningFormatUnspecified:
+		m.Reasoning = &reasoningContent
+	default:
+		logger.Warnf(context.TODO(), "unknown reasoning format: %q", format)
+	}
+}
+
+type messageAudio struct {
+	Id         string `json:"id"`
+	Data       string `json:"data,omitempty"`
+	ExpiredAt  int    `json:"expired_at,omitempty"`
+	Transcript string `json:"transcript,omitempty"`
 }
 
 func (m Message) IsStringContent() bool {
@@ -27,6 +121,7 @@ func (m Message) StringContent() string {
 			if !ok {
 				continue
 			}
+
 			if contentMap["type"] == ContentTypeText {
 				if subStr, ok := contentMap["text"].(string); ok {
 					contentStr += subStr
@@ -35,6 +130,7 @@ func (m Message) StringContent() string {
 		}
 		return contentStr
 	}
+
 	return ""
 }
 
@@ -44,10 +140,11 @@ func (m Message) ParseContent() []MessageContent {
 	if ok {
 		contentList = append(contentList, MessageContent{
 			Type: ContentTypeText,
-			Text: content,
+			Text: &content,
 		})
 		return contentList
 	}
+
 	anyList, ok := m.Content.([]any)
 	if ok {
 		for _, contentItem := range anyList {
@@ -60,7 +157,7 @@ func (m Message) ParseContent() []MessageContent {
 				if subStr, ok := contentMap["text"].(string); ok {
 					contentList = append(contentList, MessageContent{
 						Type: ContentTypeText,
-						Text: subStr,
+						Text: &subStr,
 					})
 				}
 			case ContentTypeImageURL:
@@ -72,8 +169,21 @@ func (m Message) ParseContent() []MessageContent {
 						},
 					})
 				}
+			case ContentTypeInputAudio:
+				if subObj, ok := contentMap["input_audio"].(map[string]any); ok {
+					contentList = append(contentList, MessageContent{
+						Type: ContentTypeInputAudio,
+						InputAudio: &InputAudio{
+							Data:   subObj["data"].(string),
+							Format: subObj["format"].(string),
+						},
+					})
+				}
+			default:
+				logger.Warnf(context.TODO(), "unknown content type: %s", contentMap["type"])
 			}
 		}
+
 		return contentList
 	}
 	return nil
@@ -85,7 +195,23 @@ type ImageURL struct {
 }
 
 type MessageContent struct {
-	Type     string    `json:"type,omitempty"`
-	Text     string    `json:"text"`
-	ImageURL *ImageURL `json:"image_url,omitempty"`
+	// Type should be one of the following: text/input_audio
+	Type       string      `json:"type,omitempty"`
+	Text       *string     `json:"text,omitempty"`
+	ImageURL   *ImageURL   `json:"image_url,omitempty"`
+	InputAudio *InputAudio `json:"input_audio,omitempty"`
+	// -------------------------------------
+	// Anthropic
+	// -------------------------------------
+	Thinking  *string `json:"thinking,omitempty"`
+	Signature *string `json:"signature,omitempty"`
+}
+
+type InputAudio struct {
+	// Data is the base64 encoded audio data
+	Data string `json:"data" binding:"required"`
+	// Format is the audio format, should be one of the
+	// following: mp3/mp4/mpeg/mpga/m4a/wav/webm/pcm16.
+	// When stream=true, format should be pcm16
+	Format string `json:"format"`
 }
diff --git a/relay/model/misc.go b/relay/model/misc.go
index fdba01ea..0f58bdd3 100644
--- a/relay/model/misc.go
+++ b/relay/model/misc.go
@@ -1,17 +1,22 @@
 package model
 
+// Usage is the token usage information returned by OpenAI API.
 type Usage struct {
 	PromptTokens     int `json:"prompt_tokens"`
 	CompletionTokens int `json:"completion_tokens"`
 	TotalTokens      int `json:"total_tokens"`
+	// PromptTokensDetails may be empty for some models
+	PromptTokensDetails *usagePromptTokensDetails `json:"prompt_tokens_details,omitempty"`
+	// CompletionTokensDetails may be empty for some models
+	CompletionTokensDetails *usageCompletionTokensDetails `json:"completion_tokens_details,omitempty"`
+	ServiceTier             string                        `json:"service_tier,omitempty"`
+	SystemFingerprint       string                        `json:"system_fingerprint,omitempty"`
 
-	CompletionTokensDetails *CompletionTokensDetails `json:"completion_tokens_details,omitempty"`
-}
-
-type CompletionTokensDetails struct {
-	ReasoningTokens          int `json:"reasoning_tokens"`
-	AcceptedPredictionTokens int `json:"accepted_prediction_tokens"`
-	RejectedPredictionTokens int `json:"rejected_prediction_tokens"`
+	// -------------------------------------
+	// Custom fields
+	// -------------------------------------
+	// ToolsCost is the cost of using tools, in quota.
+	ToolsCost int64 `json:"tools_cost,omitempty"`
 }
 
 type Error struct {
@@ -25,3 +30,20 @@ type ErrorWithStatusCode struct {
 	Error
 	StatusCode int `json:"status_code"`
 }
+
+type usagePromptTokensDetails struct {
+	CachedTokens int `json:"cached_tokens"`
+	AudioTokens  int `json:"audio_tokens"`
+	// TextTokens could be zero for pure text chats
+	TextTokens  int `json:"text_tokens"`
+	ImageTokens int `json:"image_tokens"`
+}
+
+type usageCompletionTokensDetails struct {
+	ReasoningTokens          int `json:"reasoning_tokens"`
+	AudioTokens              int `json:"audio_tokens"`
+	AcceptedPredictionTokens int `json:"accepted_prediction_tokens"`
+	RejectedPredictionTokens int `json:"rejected_prediction_tokens"`
+	// TextTokens could be zero for pure text chats
+	TextTokens int `json:"text_tokens"`
+}
diff --git a/relay/relaymode/define.go b/relay/relaymode/define.go
index aa771205..12acb940 100644
--- a/relay/relaymode/define.go
+++ b/relay/relaymode/define.go
@@ -13,4 +13,5 @@ const (
 	AudioTranslation
 	// Proxy is a special relay mode for proxying requests to custom upstream
 	Proxy
+	ImagesEdits
 )
diff --git a/web/air/src/components/ChannelsTable.js b/web/air/src/components/ChannelsTable.js
index c384d50c..1d312790 100644
--- a/web/air/src/components/ChannelsTable.js
+++ b/web/air/src/components/ChannelsTable.js
@@ -327,7 +327,7 @@ const ChannelsTable = () => {
     let res;
     switch (action) {
       case 'delete':
-        res = await API.delete(`/api/channel/${id}/`);
+        res = await API.delete(`/api/channel/${id}`);
         break;
       case 'enable':
         data.status = 1;
diff --git a/web/air/src/components/LogsTable.js b/web/air/src/components/LogsTable.js
index 7d372d49..7981eb5e 100644
--- a/web/air/src/components/LogsTable.js
+++ b/web/air/src/components/LogsTable.js
@@ -245,7 +245,7 @@ const LogsTable = () => {
     if (isAdminUser) {
       url = `/api/log/?p=${startIdx}&page_size=${pageSize}&type=${logType}&username=${username}&token_name=${token_name}&model_name=${model_name}&start_timestamp=${localStartTimestamp}&end_timestamp=${localEndTimestamp}&channel=${channel}`;
     } else {
-      url = `/api/log/self/?p=${startIdx}&page_size=${pageSize}&type=${logType}&token_name=${token_name}&model_name=${model_name}&start_timestamp=${localStartTimestamp}&end_timestamp=${localEndTimestamp}`;
+      url = `/api/log/self?p=${startIdx}&page_size=${pageSize}&type=${logType}&token_name=${token_name}&model_name=${model_name}&start_timestamp=${localStartTimestamp}&end_timestamp=${localEndTimestamp}`;
     }
     const res = await API.get(url);
     const { success, message, data } = res.data;
diff --git a/web/air/src/components/RedemptionsTable.js b/web/air/src/components/RedemptionsTable.js
index 89e4ce20..fdfd2f17 100644
--- a/web/air/src/components/RedemptionsTable.js
+++ b/web/air/src/components/RedemptionsTable.js
@@ -250,7 +250,7 @@ const RedemptionsTable = () => {
     let res;
     switch (action) {
       case 'delete':
-        res = await API.delete(`/api/redemption/${id}/`);
+        res = await API.delete(`/api/redemption/${id}`);
         break;
       case 'enable':
         data.status = 1;
diff --git a/web/default/src/components/ChannelsTable.js b/web/default/src/components/ChannelsTable.js
index ef2aec89..f54a223b 100644
--- a/web/default/src/components/ChannelsTable.js
+++ b/web/default/src/components/ChannelsTable.js
@@ -165,7 +165,7 @@ const ChannelsTable = () => {
     let res;
     switch (action) {
       case 'delete':
-        res = await API.delete(`/api/channel/${id}/`);
+        res = await API.delete(`/api/channel/${id}`);
         break;
       case 'enable':
         data.status = 1;
@@ -360,7 +360,7 @@ const ChannelsTable = () => {
   };
 
   const updateChannelBalance = async (id, name, idx) => {
-    const res = await API.get(`/api/channel/update_balance/${id}/`);
+    const res = await API.get(`/api/channel/update_balance/${id}`);
     const { success, message, balance } = res.data;
     if (success) {
       let newChannels = [...channels];
diff --git a/web/default/src/components/LogsTable.js b/web/default/src/components/LogsTable.js
index 8af285a1..8e106e42 100644
--- a/web/default/src/components/LogsTable.js
+++ b/web/default/src/components/LogsTable.js
@@ -225,7 +225,7 @@ const LogsTable = () => {
     if (isAdminUser) {
       url = `/api/log/?p=${startIdx}&type=${logType}&username=${username}&token_name=${token_name}&model_name=${model_name}&start_timestamp=${localStartTimestamp}&end_timestamp=${localEndTimestamp}&channel=${channel}`;
     } else {
-      url = `/api/log/self/?p=${startIdx}&type=${logType}&token_name=${token_name}&model_name=${model_name}&start_timestamp=${localStartTimestamp}&end_timestamp=${localEndTimestamp}`;
+      url = `/api/log/self?p=${startIdx}&type=${logType}&token_name=${token_name}&model_name=${model_name}&start_timestamp=${localStartTimestamp}&end_timestamp=${localEndTimestamp}`;
     }
     const res = await API.get(url);
     const { success, message, data } = res.data;
diff --git a/web/default/src/components/RedemptionsTable.js b/web/default/src/components/RedemptionsTable.js
index 3b24759d..a0185727 100644
--- a/web/default/src/components/RedemptionsTable.js
+++ b/web/default/src/components/RedemptionsTable.js
@@ -103,7 +103,7 @@ const RedemptionsTable = () => {
     let res;
     switch (action) {
       case 'delete':
-        res = await API.delete(`/api/redemption/${id}/`);
+        res = await API.delete(`/api/redemption/${id}`);
         break;
       case 'enable':
         data.status = 1;