Merge 969fdca9ef into 8df4a2670b

fix: update model ratio calculations to use multiplication instead of division for cost values
fix: refactor cost calculation logic for web-search tools and improve quota handling
2025-11-17 13:43:42 +08:00 · 2025-03-13 10:07:03 +00:00 · 2025-03-13 10:06:55 +00:00 · 2025-03-13 09:52:13 +00:00 · 2025-03-13 09:50:35 +00:00 · 2025-03-13 08:23:07 +00:00
29 changed files with 1025 additions and 417 deletions
--- a/.gitignore
+++ b/.gitignore
@@ -12,4 +12,5 @@ cmd.md
 .env
 /one-api
 temp
-.DS_Store
+.DS_Store
+/__debug_bin*
--- a/2
+++ b/2
@@ -44,4 +44,4 @@ COPY --from=builder2 /build/one-api /

 EXPOSE 3000
 WORKDIR /data
-ENTRYPOINT ["/one-api"]
+ENTRYPOINT ["/one-api"]
--- 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`
--- a/common/config/config.go
+++ b/common/config/config.go
@@ -164,3 +164,6 @@ 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", "")
--- 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 ""
 }
--- 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
--- 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
@@ -47,7 +47,7 @@ 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) {
--- 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",
 }
--- 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"`
@@ -142,13 +174,14 @@ 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 +191,10 @@ 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 +209,10 @@ 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 +230,20 @@ 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 +261,23 @@ 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 +290,13 @@ 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 +343,8 @@ 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 +412,9 @@ 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 {
--- 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 {
--- 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
--- 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"),
 	}
--- 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"`
 }
--- 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"),
 	}
--- a/relay/adaptor/aws/utils/consts.go
+++ 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
+}
--- 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,
--- 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"
@@ -85,13 +91,72 @@ 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
 }

@@ -102,11 +167,16 @@ func (a *Adaptor) ConvertImageRequest(request *model.ImageRequest) (any, error)
 	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)
@@ -125,6 +195,55 @@ func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *meta.Met
 			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
 }

--- a/relay/adaptor/openai/constants.go
+++ b/relay/adaptor/openai/constants.go
@@ -24,4 +24,8 @@ 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",
 }
--- a/relay/adaptor/openai/main.go
+++ b/relay/adaptor/openai/main.go
@@ -8,12 +8,11 @@ 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/conv"
 	"github.com/songquanpeng/one-api/common/logger"
+	"github.com/songquanpeng/one-api/common/render"
 	"github.com/songquanpeng/one-api/relay/model"
 	"github.com/songquanpeng/one-api/relay/relaymode"
 )
@@ -26,6 +25,7 @@ const (

 func StreamHandler(c *gin.Context, resp *http.Response, relayMode int) (*model.ErrorWithStatusCode, string, *model.Usage) {
 	responseText := ""
+	reasoningText := ""
 	scanner := bufio.NewScanner(resp.Body)
 	scanner.Split(bufio.ScanLines)
 	var usage *model.Usage
@@ -61,6 +61,13 @@ func StreamHandler(c *gin.Context, resp *http.Response, relayMode int) (*model.E
 			}
 			render.StringData(c, data)
 			for _, choice := range streamResponse.Choices {
+				if choice.Delta.Reasoning != nil {
+					reasoningText += *choice.Delta.Reasoning
+				}
+				if choice.Delta.ReasoningContent != nil {
+					reasoningText += *choice.Delta.ReasoningContent
+				}
+
 				responseText += conv.AsString(choice.Delta.Content)
 			}
 			if streamResponse.Usage != nil {
@@ -93,7 +100,7 @@ func StreamHandler(c *gin.Context, resp *http.Response, relayMode int) (*model.E
 		return ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), "", nil
 	}

-	return nil, responseText, usage
+	return nil, reasoningText + responseText, usage
 }

 func Handler(c *gin.Context, resp *http.Response, promptTokens int, modelName string) (*model.ErrorWithStatusCode, *model.Usage) {
@@ -136,10 +143,17 @@ func Handler(c *gin.Context, resp *http.Response, promptTokens int, modelName st
 		return ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), nil
 	}

-	if textResponse.Usage.TotalTokens == 0 || (textResponse.Usage.PromptTokens == 0 && textResponse.Usage.CompletionTokens == 0) {
+	if textResponse.Usage.TotalTokens == 0 ||
+		(textResponse.Usage.PromptTokens == 0 && textResponse.Usage.CompletionTokens == 0) {
 		completionTokens := 0
 		for _, choice := range textResponse.Choices {
 			completionTokens += CountTokenText(choice.Message.StringContent(), modelName)
+			if choice.Message.Reasoning != nil {
+				completionTokens += CountToken(*choice.Message.Reasoning)
+			}
+			if choice.ReasoningContent != nil {
+				completionTokens += CountToken(*choice.ReasoningContent)
+			}
 		}
 		textResponse.Usage = model.Usage{
 			PromptTokens:     promptTokens,
@@ -147,5 +161,6 @@ func Handler(c *gin.Context, resp *http.Response, promptTokens int, modelName st
 			TotalTokens:      promptTokens + completionTokens,
 		}
 	}
+
 	return nil, &textResponse.Usage
 }
--- a/relay/adaptor/openrouter/model.go
+++ 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"`
+}
--- 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(),
--- 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,
--- 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 {
--- a/relay/billing/ratio/model.go
+++ b/relay/billing/ratio/model.go
@@ -9,11 +9,19 @@ 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
 )

 var modelRatioLock sync.RWMutex
@@ -26,228 +34,247 @@ var modelRatioLock sync.RWMutex
 // 1 === ￥0.014 / 1k tokens
 var ModelRatio = map[string]float64{
 	// 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":                      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-search-preview":      2.5,   // $0.005 / 1K tokens
+	"gpt-4o-mini":                0.075, // $0.00015 / 1K tokens
+	"gpt-4o-mini-2024-07-18":     0.075, // $0.00015 / 1K tokens
+	"gpt-4o-mini-search-preview": 0.075, // $0.00015 / 1K tokens
+	"gpt-4-vision-preview":       5,     // $0.01 / 1K tokens
+	// 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":                 1.25,                  // $0.0025 / 1K tokens
+	"gpt-4o-audio-preview-2024-12-17":      1.25,                  // $0.0025 / 1K tokens
+	"gpt-4o-audio-preview-2024-10-01":      1.25,                  // $0.0025 / 1K tokens
+	"gpt-4o-mini-audio-preview":            0.15 * MilliTokensUsd, // $0.15/1M tokens
+	"gpt-4o-mini-audio-preview-2024-12-17": 0.15 * MilliTokensUsd, // $0.15/1M 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":                                   15 * MilliTokensUsd, // $15.00 / 1M input tokens
+	"o1-2024-12-17":                        15 * MilliTokensUsd,
+	"o1-preview":                           15 * MilliTokensUsd, // $15.00 / 1M input tokens
+	"o1-preview-2024-09-12":                15 * MilliTokensUsd,
+	"o1-mini":                              1.1 * MilliTokensUsd, // $3.00 / 1M input tokens
+	"o1-mini-2024-09-12":                   1.1 * MilliTokensUsd,
+	"o3-mini":                              1.1 * MilliTokensUsd,
+	"o3-mini-2025-01-31":                   1.1 * MilliTokensUsd,
+	"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,
+	"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":                             20 * MilliTokensUsd, // $0.016 - $0.020 / image
+	"dall-e-3":                             40 * MilliTokensUsd, // $0.040 - $0.120 / image
+	// 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,
+	"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-pro-exp-02-05":            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,
+	"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":       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,
+	"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
@@ -261,69 +288,68 @@ var ModelRatio = map[string]float64{
 	"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,
+	"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://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,
-
+	"gemma-7b-it":                           0.07 / 1000 * MilliTokensUsd,
+	"gemma2-9b-it":                          0.20 / 1000 * MilliTokensUsd,
+	"llama-3.1-70b-versatile":               0.59 / 1000 * MilliTokensUsd,
+	"llama-3.1-8b-instant":                  0.05 / 1000 * MilliTokensUsd,
+	"llama-3.2-11b-text-preview":            0.05 / 1000 * MilliTokensUsd,
+	"llama-3.2-11b-vision-preview":          0.05 / 1000 * MilliTokensUsd,
+	"llama-3.2-1b-preview":                  0.05 / 1000 * MilliTokensUsd,
+	"llama-3.2-3b-preview":                  0.05 / 1000 * MilliTokensUsd,
+	"llama-3.2-90b-text-preview":            0.59 / 1000 * MilliTokensUsd,
+	"llama-guard-3-8b":                      0.05 / 1000 * MilliTokensUsd,
+	"llama3-70b-8192":                       0.59 / 1000 * MilliTokensUsd,
+	"llama3-8b-8192":                        0.05 / 1000 * MilliTokensUsd,
+	"llama3-groq-70b-8192-tool-use-preview": 0.89 / 1000 * MilliTokensUsd,
+	"llama3-groq-8b-8192-tool-use-preview":  0.19 / 1000 * MilliTokensUsd,
+	"mixtral-8x7b-32768":                    0.24 / 1000 * 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
@@ -332,61 +358,64 @@ 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": 0.02 * KiloTokensUsd,
 	// 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 * KiloTokensUsd,
+	"black-forest-labs/flux-1.1-pro-ultra":          0.06 * KiloTokensUsd,
+	"black-forest-labs/flux-canny-dev":              0.025 * KiloTokensUsd,
+	"black-forest-labs/flux-canny-pro":              0.05 * KiloTokensUsd,
+	"black-forest-labs/flux-depth-dev":              0.025 * KiloTokensUsd,
+	"black-forest-labs/flux-depth-pro":              0.05 * KiloTokensUsd,
+	"black-forest-labs/flux-dev":                    0.025 * KiloTokensUsd,
+	"black-forest-labs/flux-dev-lora":               0.032 * KiloTokensUsd,
+	"black-forest-labs/flux-fill-dev":               0.04 * KiloTokensUsd,
+	"black-forest-labs/flux-fill-pro":               0.05 * KiloTokensUsd,
+	"black-forest-labs/flux-pro":                    0.055 * KiloTokensUsd,
+	"black-forest-labs/flux-redux-dev":              0.025 * KiloTokensUsd,
+	"black-forest-labs/flux-redux-schnell":          0.003 * KiloTokensUsd,
+	"black-forest-labs/flux-schnell":                0.003 * KiloTokensUsd,
+	"black-forest-labs/flux-schnell-lora":           0.02 * KiloTokensUsd,
+	"ideogram-ai/ideogram-v2":                       0.08 * KiloTokensUsd,
+	"ideogram-ai/ideogram-v2-turbo":                 0.05 * KiloTokensUsd,
+	"recraft-ai/recraft-v3":                         0.04 * KiloTokensUsd,
+	"recraft-ai/recraft-v3-svg":                     0.08 * KiloTokensUsd,
+	"stability-ai/stable-diffusion-3":               0.035 * KiloTokensUsd,
+	"stability-ai/stable-diffusion-3.5-large":       0.065 * KiloTokensUsd,
+	"stability-ai/stable-diffusion-3.5-large-turbo": 0.04 * KiloTokensUsd,
+	"stability-ai/stable-diffusion-3.5-medium":      0.035 * KiloTokensUsd,
 	// 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,
+	"ibm-granite/granite-20b-code-instruct-8k":  0.100 * KiloTokensUsd,
+	"ibm-granite/granite-3.0-2b-instruct":       0.030 * KiloTokensUsd,
+	"ibm-granite/granite-3.0-8b-instruct":       0.050 * KiloTokensUsd,
+	"ibm-granite/granite-8b-code-instruct-128k": 0.050 * KiloTokensUsd,
+	"meta/llama-2-13b":                          0.100 * KiloTokensUsd,
+	"meta/llama-2-13b-chat":                     0.100 * KiloTokensUsd,
+	"meta/llama-2-70b":                          0.650 * KiloTokensUsd,
+	"meta/llama-2-70b-chat":                     0.650 * KiloTokensUsd,
+	"meta/llama-2-7b":                           0.050 * KiloTokensUsd,
+	"meta/llama-2-7b-chat":                      0.050 * KiloTokensUsd,
+	"meta/meta-llama-3.1-405b-instruct":         9.500 * KiloTokensUsd,
+	"meta/meta-llama-3-70b":                     0.650 * KiloTokensUsd,
+	"meta/meta-llama-3-70b-instruct":            0.650 * KiloTokensUsd,
+	"meta/meta-llama-3-8b":                      0.050 * KiloTokensUsd,
+	"meta/meta-llama-3-8b-instruct":             0.050 * KiloTokensUsd,
+	"mistralai/mistral-7b-instruct-v0.2":        0.050 * KiloTokensUsd,
+	"mistralai/mistral-7b-v0.1":                 0.050 * KiloTokensUsd,
+	"mistralai/mixtral-8x7b-instruct-v0.1":      0.300 * KiloTokensUsd,
 	//https://openrouter.ai/models
 	"01-ai/yi-large":                                  1.5,
 	"aetherwiing/mn-starcannon-12b":                   0.6,
@@ -435,10 +464,10 @@ 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 * MilliTokensUsd,
 	"deepseek/deepseek-chat-v2.5":                     1.0,
 	"deepseek/deepseek-chat:free":                     0.0,
-	"deepseek/deepseek-r1":                            1.2,
+	"deepseek/deepseek-r1":                            7 * MilliTokensUsd,
 	"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 +595,7 @@ 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 * MilliTokensUsd,
 	"openai/o1":                                       30.0,
 	"openai/o1-mini":                                  2.2,
 	"openai/o1-mini-2024-09-12":                       2.2,
--- 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"
@@ -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) {
--- 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"
@@ -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())
--- 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=1"`
+	// 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 {
--- 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 ""
 }

@@ -48,6 +144,7 @@ func (m Message) ParseContent() []MessageContent {
 		})
 		return contentList
 	}
+
 	anyList, ok := m.Content.([]any)
 	if ok {
 		for _, contentItem := range anyList {
@@ -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"`
+	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"`
 }
--- 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"`
+}
Author	SHA1	Message	Date
Laisky.Cai	78de70ff5c	Merge `969fdca9ef` into `8df4a2670b`	2025-03-13 10:07:03 +00:00
Laisky.Cai	969fdca9ef	fix: update model ratio calculations to use multiplication instead of division for cost values	2025-03-13 10:06:55 +00:00
Laisky.Cai	6708eed8a0	fix: refactor cost calculation logic for web-search tools and improve quota handling	2025-03-13 09:52:13 +00:00
Laisky.Cai	ad63c9e66f	fix: update cost calculation to use QuotaPerUsd for search context sizes	2025-03-13 09:50:35 +00:00
Laisky.Cai	76e8199026	fix: add support for OpenAI web search models in documentation and request handling	2025-03-13 08:23:07 +00:00
Laisky.Cai	413fcde382	feat: support openai websearch models	2025-03-13 03:46:15 +00:00
Laisky.Cai	6e634b85cf	fix: update StreamHandler to support cross-region model IDs for AWS	2025-03-12 00:34:15 +00:00
Laisky.Cai	a0d7d5a965	fix: support thinking for aws claude	2025-03-10 07:00:45 +00:00
Laisky.Cai	de10e102bd	feat: add support for aws's cross region inferences closes #2024, closes #2145	2025-03-10 06:43:40 +00:00
Laisky.Cai	c61d6440f9	fix: claude thinking for non-stream mode	2025-02-25 03:14:18 +00:00
Laisky.Cai	3a8924d7af	feat: add support for extended reasoning in Claude 3.7 model	2025-02-25 03:02:51 +00:00
Laisky.Cai	95527d76ef	feat: update model list and pricing for Claude 3.7 versions	2025-02-25 03:02:24 +00:00
Laisky.Cai	7ec33793b7	feat: add OpenrouterProviderSort configuration for provider sorting	2025-02-20 01:51:45 +00:00
Laisky.Cai	1a6812182b	fix: improve reasoning token counting in OpenAI adaptor	2025-02-19 09:13:24 +00:00
Laisky.Cai	5ba60433d7	feat: enhance reasoning token handling in OpenAI adaptor	2025-02-19 08:10:19 +00:00
Laisky.Cai	480f248a3d	feat: support OpenRouter reasoning	2025-02-19 01:20:14 +00:00