Merge 290931b506 into 8df4a2670b

- Add comprehensive npm mirror configuration for faster package downloads
- Use parallel npm install and build processes for better performance
- Configure Go proxy and Alpine package mirrors for network reliability
- Fix frontend dependency issues that caused blank page display

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>

Dockerfile

@@ -1,14 +1,28 @@
 FROM --platform=$BUILDPLATFORM node:16 AS builder
 
+# 配置npm使用淘宝镜像源
+RUN npm config set registry https://registry.npmmirror.com && \
+    npm config set disturl https://npmmirror.com/dist && \
+    npm config set sass_binary_site https://npmmirror.com/mirrors/node-sass/ && \
+    npm config set electron_mirror https://npmmirror.com/mirrors/electron/ && \
+    npm config set puppeteer_download_host https://npmmirror.com/mirrors && \
+    npm config set chromedriver_cdnurl https://npmmirror.com/mirrors/chromedriver && \
+    npm config set operadriver_cdnurl https://npmmirror.com/mirrors/operadriver && \
+    npm config set phantomjs_cdnurl https://npmmirror.com/mirrors/phantomjs && \
+    npm config set selenium_cdnurl https://npmmirror.com/mirrors/selenium && \
+    npm config set node_inspector_cdnurl https://npmmirror.com/mirrors/node-inspector
+
 WORKDIR /web
 COPY ./VERSION .
 COPY ./web .
 
-RUN npm install --prefix /web/default & \
+# 并行安装npm依赖，提高构建速度
-    npm install --prefix /web/berry & \
+RUN npm install --prefix /web/default --prefer-offline --no-audit & \
-    npm install --prefix /web/air & \
+    npm install --prefix /web/berry --prefer-offline --no-audit & \
+    npm install --prefix /web/air --prefer-offline --no-audit & \
     wait
 
+# 并行构建前端项目，提高构建速度
 RUN DISABLE_ESLINT_PLUGIN='true' REACT_APP_VERSION=$(cat ./VERSION) npm run build --prefix /web/default & \
     DISABLE_ESLINT_PLUGIN='true' REACT_APP_VERSION=$(cat ./VERSION) npm run build --prefix /web/berry & \
     DISABLE_ESLINT_PLUGIN='true' REACT_APP_VERSION=$(cat ./VERSION) npm run build --prefix /web/air & \
@@ -16,16 +30,22 @@ RUN DISABLE_ESLINT_PLUGIN='true' REACT_APP_VERSION=$(cat ./VERSION) npm run buil
 
 FROM golang:alpine AS builder2
 
-RUN apk add --no-cache \
+# 配置Go使用国内镜像源
+ENV GOPROXY=https://goproxy.cn,direct \
+    GOSUMDB=sum.golang.google.cn \
+    GO111MODULE=on \
+    CGO_ENABLED=1 \
+    GOOS=linux
+
+# 使用阿里云 Alpine 源以加速 apk 包安装
+RUN sed -i 's/dl-cdn.alpinelinux.org/mirrors.aliyun.com/g' /etc/apk/repositories && \
+    apk update && \
+    apk add --no-cache \
     gcc \
     musl-dev \
     sqlite-dev \
     build-base
 
-ENV GO111MODULE=on \
-    CGO_ENABLED=1 \
-    GOOS=linux
-
 WORKDIR /build
 
 ADD go.mod go.sum ./
@@ -38,7 +58,10 @@ RUN go build -trimpath -ldflags "-s -w -X 'github.com/songquanpeng/one-api/commo
 
 FROM alpine:latest
 
-RUN apk add --no-cache ca-certificates tzdata
+# 使用阿里云 Alpine 源以加速 apk 包安装
+RUN sed -i 's/dl-cdn.alpinelinux.org/mirrors.aliyun.com/g' /etc/apk/repositories && \
+    apk update && \
+    apk add --no-cache ca-certificates tzdata
 
 COPY --from=builder2 /build/one-api /
 

README.md

@@ -72,7 +72,7 @@ _✨ 通过标准的 OpenAI API 格式访问所有的大模型，开箱即用
    + [x] [Anthropic Claude 系列模型](https://anthropic.com) (支持 AWS Claude)
    + [x] [Google PaLM2/Gemini 系列模型](https://developers.generativeai.google)
    + [x] [Mistral 系列模型](https://mistral.ai/)
-   + [x] [字节跳动豆包大模型](https://console.volcengine.com/ark/region:ark+cn-beijing/model)
+   + [x] [字节跳动豆包大模型（火山引擎）](https://www.volcengine.com/experience/ark?utm_term=202502dsinvite&ac=DSASUQY5&rc=2QXCA1VI)
    + [x] [百度文心一言系列模型](https://cloud.baidu.com/doc/WENXINWORKSHOP/index.html)
    + [x] [阿里通义千问系列模型](https://help.aliyun.com/document_detail/2400395.html)
    + [x] [讯飞星火认知大模型](https://www.xfyun.cn/doc/spark/Web.html)

controller/relay.go

@@ -36,6 +36,8 @@ func relayHelper(c *gin.Context, relayMode int) *model.ErrorWithStatusCode {
 		err = controller.RelayAudioHelper(c, relayMode)
 	case relaymode.Proxy:
 		err = controller.RelayProxyHelper(c, relayMode)
+	case relaymode.AnthropicMessages:
+		err = controller.RelayAnthropicHelper(c)
 	default:
 		err = controller.RelayTextHelper(c)
 	}

docker-compose.yml

@@ -1,9 +1,9 @@
-version: '3.4'
-
 services:
   one-api:
-    image: "${REGISTRY:-docker.io}/justsong/one-api:latest"
+    # image: "${REGISTRY:-docker.io}/justsong/one-api:latest"
+    build: .
     container_name: one-api
+    platform: linux/amd64
     restart: always
     command: --log-dir /app/logs
     ports:

relay/adaptor/ali/constants.go

@@ -14,10 +14,14 @@ var ModelList = []string{
 	"qwen2-72b-instruct", "qwen2-57b-a14b-instruct", "qwen2-7b-instruct", "qwen2-1.5b-instruct", "qwen2-0.5b-instruct",
 	"qwen1.5-110b-chat", "qwen1.5-72b-chat", "qwen1.5-32b-chat", "qwen1.5-14b-chat", "qwen1.5-7b-chat", "qwen1.5-1.8b-chat", "qwen1.5-0.5b-chat",
 	"qwen-72b-chat", "qwen-14b-chat", "qwen-7b-chat", "qwen-1.8b-chat", "qwen-1.8b-longcontext-chat",
+	"qvq-72b-preview",
+	"qwen2.5-vl-72b-instruct", "qwen2.5-vl-7b-instruct", "qwen2.5-vl-2b-instruct", "qwen2.5-vl-1b-instruct", "qwen2.5-vl-0.5b-instruct",
 	"qwen2-vl-7b-instruct", "qwen2-vl-2b-instruct", "qwen-vl-v1", "qwen-vl-chat-v1",
 	"qwen2-audio-instruct", "qwen-audio-chat",
 	"qwen2.5-math-72b-instruct", "qwen2.5-math-7b-instruct", "qwen2.5-math-1.5b-instruct", "qwen2-math-72b-instruct", "qwen2-math-7b-instruct", "qwen2-math-1.5b-instruct",
 	"qwen2.5-coder-32b-instruct", "qwen2.5-coder-14b-instruct", "qwen2.5-coder-7b-instruct", "qwen2.5-coder-3b-instruct", "qwen2.5-coder-1.5b-instruct", "qwen2.5-coder-0.5b-instruct",
 	"text-embedding-v1", "text-embedding-v3", "text-embedding-v2", "text-embedding-async-v2", "text-embedding-async-v1",
 	"ali-stable-diffusion-xl", "ali-stable-diffusion-v1.5", "wanx-v1",
+	"qwen-mt-plus", "qwen-mt-turbo",
+	"deepseek-r1", "deepseek-v3", "deepseek-r1-distill-qwen-1.5b", "deepseek-r1-distill-qwen-7b", "deepseek-r1-distill-qwen-14b", "deepseek-r1-distill-qwen-32b", "deepseek-r1-distill-llama-8b", "deepseek-r1-distill-llama-70b",
 }

relay/adaptor/anthropic/adaptor.go

@@ -11,6 +11,14 @@ import (
 	"github.com/songquanpeng/one-api/relay/adaptor"
 	"github.com/songquanpeng/one-api/relay/meta"
 	"github.com/songquanpeng/one-api/relay/model"
+	"github.com/songquanpeng/one-api/relay/relaymode"
+)
+
+const (
+	// NativeAnthropicEndpoint is the endpoint for native Anthropic API
+	NativeAnthropicEndpoint = "/v1/messages"
+	// ThirdPartyAnthropicEndpoint is the endpoint for third-party providers supporting Anthropic protocol
+	ThirdPartyAnthropicEndpoint = "/anthropic/v1/messages"
 )
 
 type Adaptor struct {
@@ -21,7 +29,15 @@ func (a *Adaptor) Init(meta *meta.Meta) {
 }
 
 func (a *Adaptor) GetRequestURL(meta *meta.Meta) (string, error) {
-	return fmt.Sprintf("%s/v1/messages", meta.BaseURL), nil
+	// For native Anthropic API
+	if strings.Contains(meta.BaseURL, "api.anthropic.com") {
+		return fmt.Sprintf("%s%s", meta.BaseURL, NativeAnthropicEndpoint), nil
+	}
+
+	// For third-party providers supporting Anthropic protocol (like DeepSeek)
+	// They typically expose the endpoint at /anthropic/v1/messages
+	baseURL := strings.TrimSuffix(meta.BaseURL, "/")
+	return fmt.Sprintf("%s%s", baseURL, ThirdPartyAnthropicEndpoint), nil
 }
 
 func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Request, meta *meta.Meta) error {
@@ -47,6 +63,15 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 	if request == nil {
 		return nil, errors.New("request is nil")
 	}
+
+	// For native Anthropic protocol requests, return the request as-is (no conversion needed)
+	if relayMode == relaymode.AnthropicMessages {
+		// The request should already be in Anthropic format, so we pass it through
+		// This will be handled by the caller which already has the anthropic request
+		return request, nil
+	}
+
+	// For OpenAI to Anthropic conversion (existing functionality)
 	return ConvertRequest(*request), nil
 }
 
@@ -62,6 +87,17 @@ func (a *Adaptor) DoRequest(c *gin.Context, meta *meta.Meta, requestBody io.Read
 }
 
 func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, meta *meta.Meta) (usage *model.Usage, err *model.ErrorWithStatusCode) {
+	// For native Anthropic protocol requests, handle response directly without conversion
+	if meta.Mode == relaymode.AnthropicMessages {
+		if meta.IsStream {
+			err, usage = DirectStreamHandler(c, resp)
+		} else {
+			err, usage = DirectHandler(c, resp, meta.PromptTokens, meta.ActualModelName)
+		}
+		return
+	}
+
+	// For OpenAI to Anthropic conversion (existing functionality)
 	if meta.IsStream {
 		err, usage = StreamHandler(c, resp)
 	} else {

relay/adaptor/anthropic/main.go

@@ -4,11 +4,12 @@ import (
 	"bufio"
 	"encoding/json"
 	"fmt"
-	"github.com/songquanpeng/one-api/common/render"
 	"io"
 	"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/helper"
@@ -89,8 +90,18 @@ func ConvertRequest(textRequest model.GeneralOpenAIRequest) *Request {
 		claudeRequest.Model = "claude-2.1"
 	}
 	for _, message := range textRequest.Messages {
-		if message.Role == "system" && claudeRequest.System == "" {
+		if message.Role == "system" && claudeRequest.System.IsEmpty() {
-			claudeRequest.System = message.StringContent()
+			// Create a SystemPrompt from the string content
+			systemPrompt := SystemPrompt{}
+			systemData, err := json.Marshal(message.StringContent()) // Safely escape string for JSON
+			if err != nil {
+				logger.SysError(fmt.Sprintf("Failed to marshal system prompt: %v", err))
+			} else {
+				if err := systemPrompt.UnmarshalJSON(systemData); err != nil {
+					logger.SysError(fmt.Sprintf("Failed to unmarshal system prompt: %v", err))
+				}
+				claudeRequest.System = systemPrompt
+			}
 			continue
 		}
 		claudeMessage := Message{
@@ -377,3 +388,128 @@ func Handler(c *gin.Context, resp *http.Response, promptTokens int, modelName st
 	_, err = c.Writer.Write(jsonResponse)
 	return nil, &usage
 }
+
+// DirectHandler handles native Anthropic API responses without conversion to OpenAI format
+func DirectHandler(c *gin.Context, resp *http.Response, promptTokens int, modelName string) (*model.ErrorWithStatusCode, *model.Usage) {
+	ctx := c.Request.Context()
+	logger.Debugf(ctx, "DirectHandler - Response status: %d", resp.StatusCode)
+
+	responseBody, err := io.ReadAll(resp.Body)
+	if err != nil {
+		logger.Errorf(ctx, "Failed to read response body: %s", err.Error())
+		return openai.ErrorWrapper(err, "read_response_body_failed", http.StatusInternalServerError), nil
+	}
+	err = resp.Body.Close()
+	if err != nil {
+		logger.Errorf(ctx, "Failed to close response body: %s", err.Error())
+		return openai.ErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError), nil
+	}
+
+	logger.Debugf(ctx, "Raw response body: %s", string(responseBody))
+
+	var claudeResponse Response
+	err = json.Unmarshal(responseBody, &claudeResponse)
+	if err != nil {
+		logger.Errorf(ctx, "Failed to unmarshal response: %s", err.Error())
+		// If we can't parse as Anthropic response, maybe it's an error response
+		// Let's try to write it directly and see what happens
+		c.Writer.Header().Set("Content-Type", "application/json")
+		c.Writer.WriteHeader(resp.StatusCode)
+		_, writeErr := c.Writer.Write(responseBody)
+		if writeErr != nil {
+			logger.Errorf(ctx, "Failed to write raw response: %s", writeErr.Error())
+			return openai.ErrorWrapper(writeErr, "write_response_failed", http.StatusInternalServerError), nil
+		}
+		// Return a minimal usage for tracking
+		usage := &model.Usage{PromptTokens: promptTokens, CompletionTokens: 0, TotalTokens: promptTokens}
+		return nil, usage
+	}
+
+	logger.Debugf(ctx, "Parsed response - ID: %s, Model: %s, Usage: %+v",
+		claudeResponse.Id, claudeResponse.Model, claudeResponse.Usage)
+
+	if claudeResponse.Error.Type != "" {
+		logger.Errorf(ctx, "Anthropic API error: %s - %s", claudeResponse.Error.Type, claudeResponse.Error.Message)
+		return &model.ErrorWithStatusCode{
+			Error: model.Error{
+				Message: claudeResponse.Error.Message,
+				Type:    claudeResponse.Error.Type,
+				Param:   "",
+				Code:    claudeResponse.Error.Type,
+			},
+			StatusCode: resp.StatusCode,
+		}, nil
+	}
+
+	// For direct mode, return the response as-is without conversion
+	usage := model.Usage{
+		PromptTokens:     claudeResponse.Usage.InputTokens,
+		CompletionTokens: claudeResponse.Usage.OutputTokens,
+		TotalTokens:      claudeResponse.Usage.InputTokens + claudeResponse.Usage.OutputTokens,
+	}
+
+	logger.Debugf(ctx, "Usage calculated: %+v", usage)
+
+	// Write the original Anthropic response directly
+	c.Writer.Header().Set("Content-Type", "application/json")
+	c.Writer.WriteHeader(resp.StatusCode)
+	_, err = c.Writer.Write(responseBody)
+	if err != nil {
+		logger.Errorf(ctx, "Failed to write response: %s", err.Error())
+		return openai.ErrorWrapper(err, "write_response_failed", http.StatusInternalServerError), nil
+	}
+
+	logger.Debugf(ctx, "Response written successfully")
+	return nil, &usage
+}
+
+// DirectStreamHandler handles native Anthropic API streaming responses without conversion
+func DirectStreamHandler(c *gin.Context, resp *http.Response) (*model.ErrorWithStatusCode, *model.Usage) {
+	defer resp.Body.Close()
+
+	// Set headers for streaming
+	c.Writer.Header().Set("Content-Type", "text/event-stream")
+	c.Writer.Header().Set("Cache-Control", "no-cache")
+	c.Writer.Header().Set("Connection", "keep-alive")
+	c.Writer.Header().Set("Access-Control-Allow-Origin", "*")
+	c.Writer.Header().Set("Access-Control-Allow-Headers", "Content-Type")
+	c.Writer.WriteHeader(resp.StatusCode)
+
+	// Stream the response directly without conversion
+	var usage model.Usage
+	scanner := bufio.NewScanner(resp.Body)
+	for scanner.Scan() {
+		data := scanner.Text()
+		if len(data) < 6 || !strings.HasPrefix(data, "data:") {
+			continue
+		}
+
+		// Parse usage information if available
+		if strings.Contains(data, "\"usage\":") {
+			var eventData map[string]interface{}
+			jsonData := strings.TrimPrefix(data, "data:")
+			jsonData = strings.TrimSpace(jsonData)
+			if err := json.Unmarshal([]byte(jsonData), &eventData); err == nil {
+				if usageData, ok := eventData["usage"].(map[string]interface{}); ok {
+					if inputTokens, ok := usageData["input_tokens"].(float64); ok {
+						usage.PromptTokens = int(inputTokens)
+					}
+					if outputTokens, ok := usageData["output_tokens"].(float64); ok {
+						usage.CompletionTokens = int(outputTokens)
+						usage.TotalTokens = usage.PromptTokens + usage.CompletionTokens
+					}
+				}
+			}
+		}
+
+		// Write data directly to the response
+		c.Writer.WriteString(data + "\n")
+		c.Writer.Flush()
+	}
+
+	if err := scanner.Err(); err != nil {
+		return openai.ErrorWrapper(err, "stream_read_failed", http.StatusInternalServerError), nil
+	}
+
+	return nil, &usage
+}

relay/adaptor/anthropic/model.go

@@ -1,5 +1,10 @@
 package anthropic
 
+import (
+	"encoding/json"
+	"fmt"
+)
+
 // https://docs.anthropic.com/claude/reference/messages_post
 
 type Metadata struct {
@@ -41,18 +46,92 @@ type InputSchema struct {
 	Required   any    `json:"required,omitempty"`
 }
 
+// SystemPrompt can handle both string and array formats for the system field
+type SystemPrompt struct {
+	value interface{}
+}
+
+// UnmarshalJSON implements json.Unmarshaler to handle both string and array formats
+func (s *SystemPrompt) UnmarshalJSON(data []byte) error {
+	// Try to unmarshal as string first
+	var str string
+	if err := json.Unmarshal(data, &str); err == nil {
+		s.value = str
+		return nil
+	}
+
+	// If that fails, try to unmarshal as array
+	var arr []interface{}
+	if err := json.Unmarshal(data, &arr); err == nil {
+		s.value = arr
+		return nil
+	}
+
+	return fmt.Errorf("system field must be either a string or an array")
+}
+
+// MarshalJSON implements json.Marshaler
+func (s SystemPrompt) MarshalJSON() ([]byte, error) {
+	return json.Marshal(s.value)
+}
+
+// String returns the system prompt as a string
+func (s SystemPrompt) String() string {
+	if s.value == nil {
+		return ""
+	}
+
+	switch v := s.value.(type) {
+	case string:
+		return v
+	case []interface{}:
+		// Convert array to string by concatenating text content
+		var result string
+		for _, item := range v {
+			if itemMap, ok := item.(map[string]interface{}); ok {
+				if text, exists := itemMap["text"]; exists {
+					if textStr, ok := text.(string); ok {
+						result += textStr + " "
+					}
+				}
+			} else if str, ok := item.(string); ok {
+				result += str + " "
+			}
+		}
+		return result
+	default:
+		return fmt.Sprintf("%v", v)
+	}
+}
+
+// IsEmpty returns true if the system prompt is empty
+func (s SystemPrompt) IsEmpty() bool {
+	if s.value == nil {
+		return true
+	}
+
+	switch v := s.value.(type) {
+	case string:
+		return v == ""
+	case []interface{}:
+		return len(v) == 0
+	default:
+		return false
+	}
+}
+
 type Request struct {
-	Model         string    `json:"model"`
+	Model         string       `json:"model"`
-	Messages      []Message `json:"messages"`
+	Messages      []Message    `json:"messages"`
-	System        string    `json:"system,omitempty"`
+	System        SystemPrompt `json:"system,omitempty"`
-	MaxTokens     int       `json:"max_tokens,omitempty"`
+	MaxTokens     int          `json:"max_tokens,omitempty"`
-	StopSequences []string  `json:"stop_sequences,omitempty"`
+	StopSequences []string     `json:"stop_sequences,omitempty"`
-	Stream        bool      `json:"stream,omitempty"`
+	Stream        bool         `json:"stream,omitempty"`
-	Temperature   *float64  `json:"temperature,omitempty"`
+	Temperature   *float64     `json:"temperature,omitempty"`
-	TopP          *float64  `json:"top_p,omitempty"`
+	TopP          *float64     `json:"top_p,omitempty"`
-	TopK          int       `json:"top_k,omitempty"`
+	TopK          int          `json:"top_k,omitempty"`
-	Tools         []Tool    `json:"tools,omitempty"`
+	Tools         []Tool       `json:"tools,omitempty"`
-	ToolChoice    any       `json:"tool_choice,omitempty"`
+	ToolChoice    any          `json:"tool_choice,omitempty"`
 	//Metadata    `json:"metadata,omitempty"`
 }
 

relay/adaptor/openrouter/constants.go

@@ -1,20 +1,235 @@
 package openrouter
 
 var ModelList = []string{
-	"openai/gpt-3.5-turbo",
+	"01-ai/yi-large",
-	"openai/chatgpt-4o-latest",
+	"aetherwiing/mn-starcannon-12b",
-	"openai/o1",
+	"ai21/jamba-1-5-large",
-	"openai/o1-preview",
+	"ai21/jamba-1-5-mini",
-	"openai/o1-mini",
+	"ai21/jamba-instruct",
-	"openai/o3-mini",
+	"aion-labs/aion-1.0",
-	"google/gemini-2.0-flash-001",
+	"aion-labs/aion-1.0-mini",
-	"google/gemini-2.0-flash-thinking-exp:free",
+	"aion-labs/aion-rp-llama-3.1-8b",
-	"google/gemini-2.0-flash-lite-preview-02-05:free",
+	"allenai/llama-3.1-tulu-3-405b",
-	"google/gemini-2.0-pro-exp-02-05:free",
+	"alpindale/goliath-120b",
-	"google/gemini-flash-1.5-8b",
+	"alpindale/magnum-72b",
-	"anthropic/claude-3.5-sonnet",
+	"amazon/nova-lite-v1",
+	"amazon/nova-micro-v1",
+	"amazon/nova-pro-v1",
+	"anthracite-org/magnum-v2-72b",
+	"anthracite-org/magnum-v4-72b",
+	"anthropic/claude-2",
+	"anthropic/claude-2.0",
+	"anthropic/claude-2.0:beta",
+	"anthropic/claude-2.1",
+	"anthropic/claude-2.1:beta",
+	"anthropic/claude-2:beta",
+	"anthropic/claude-3-haiku",
+	"anthropic/claude-3-haiku:beta",
+	"anthropic/claude-3-opus",
+	"anthropic/claude-3-opus:beta",
+	"anthropic/claude-3-sonnet",
+	"anthropic/claude-3-sonnet:beta",
 	"anthropic/claude-3.5-haiku",
-	"deepseek/deepseek-r1:free",
+	"anthropic/claude-3.5-haiku-20241022",
+	"anthropic/claude-3.5-haiku-20241022:beta",
+	"anthropic/claude-3.5-haiku:beta",
+	"anthropic/claude-3.5-sonnet",
+	"anthropic/claude-3.5-sonnet-20240620",
+	"anthropic/claude-3.5-sonnet-20240620:beta",
+	"anthropic/claude-3.5-sonnet:beta",
+	"cognitivecomputations/dolphin-mixtral-8x22b",
+	"cognitivecomputations/dolphin-mixtral-8x7b",
+	"cohere/command",
+	"cohere/command-r",
+	"cohere/command-r-03-2024",
+	"cohere/command-r-08-2024",
+	"cohere/command-r-plus",
+	"cohere/command-r-plus-04-2024",
+	"cohere/command-r-plus-08-2024",
+	"cohere/command-r7b-12-2024",
+	"databricks/dbrx-instruct",
+	"deepseek/deepseek-chat",
+	"deepseek/deepseek-chat-v2.5",
+	"deepseek/deepseek-chat:free",
 	"deepseek/deepseek-r1",
+	"deepseek/deepseek-r1-distill-llama-70b",
+	"deepseek/deepseek-r1-distill-llama-70b:free",
+	"deepseek/deepseek-r1-distill-llama-8b",
+	"deepseek/deepseek-r1-distill-qwen-1.5b",
+	"deepseek/deepseek-r1-distill-qwen-14b",
+	"deepseek/deepseek-r1-distill-qwen-32b",
+	"deepseek/deepseek-r1:free",
+	"eva-unit-01/eva-llama-3.33-70b",
+	"eva-unit-01/eva-qwen-2.5-32b",
+	"eva-unit-01/eva-qwen-2.5-72b",
+	"google/gemini-2.0-flash-001",
+	"google/gemini-2.0-flash-exp:free",
+	"google/gemini-2.0-flash-lite-preview-02-05:free",
+	"google/gemini-2.0-flash-thinking-exp-1219:free",
+	"google/gemini-2.0-flash-thinking-exp:free",
+	"google/gemini-2.0-pro-exp-02-05:free",
+	"google/gemini-exp-1206:free",
+	"google/gemini-flash-1.5",
+	"google/gemini-flash-1.5-8b",
+	"google/gemini-flash-1.5-8b-exp",
+	"google/gemini-pro",
+	"google/gemini-pro-1.5",
+	"google/gemini-pro-vision",
+	"google/gemma-2-27b-it",
+	"google/gemma-2-9b-it",
+	"google/gemma-2-9b-it:free",
+	"google/gemma-7b-it",
+	"google/learnlm-1.5-pro-experimental:free",
+	"google/palm-2-chat-bison",
+	"google/palm-2-chat-bison-32k",
+	"google/palm-2-codechat-bison",
+	"google/palm-2-codechat-bison-32k",
+	"gryphe/mythomax-l2-13b",
+	"gryphe/mythomax-l2-13b:free",
+	"huggingfaceh4/zephyr-7b-beta:free",
+	"infermatic/mn-inferor-12b",
+	"inflection/inflection-3-pi",
+	"inflection/inflection-3-productivity",
+	"jondurbin/airoboros-l2-70b",
+	"liquid/lfm-3b",
+	"liquid/lfm-40b",
+	"liquid/lfm-7b",
+	"mancer/weaver",
+	"meta-llama/llama-2-13b-chat",
+	"meta-llama/llama-2-70b-chat",
+	"meta-llama/llama-3-70b-instruct",
+	"meta-llama/llama-3-8b-instruct",
+	"meta-llama/llama-3-8b-instruct:free",
+	"meta-llama/llama-3.1-405b",
+	"meta-llama/llama-3.1-405b-instruct",
+	"meta-llama/llama-3.1-70b-instruct",
+	"meta-llama/llama-3.1-8b-instruct",
+	"meta-llama/llama-3.2-11b-vision-instruct",
+	"meta-llama/llama-3.2-11b-vision-instruct:free",
+	"meta-llama/llama-3.2-1b-instruct",
+	"meta-llama/llama-3.2-3b-instruct",
+	"meta-llama/llama-3.2-90b-vision-instruct",
+	"meta-llama/llama-3.3-70b-instruct",
+	"meta-llama/llama-3.3-70b-instruct:free",
+	"meta-llama/llama-guard-2-8b",
+	"microsoft/phi-3-medium-128k-instruct",
+	"microsoft/phi-3-medium-128k-instruct:free",
+	"microsoft/phi-3-mini-128k-instruct",
+	"microsoft/phi-3-mini-128k-instruct:free",
+	"microsoft/phi-3.5-mini-128k-instruct",
+	"microsoft/phi-4",
+	"microsoft/wizardlm-2-7b",
+	"microsoft/wizardlm-2-8x22b",
+	"minimax/minimax-01",
+	"mistralai/codestral-2501",
+	"mistralai/codestral-mamba",
+	"mistralai/ministral-3b",
+	"mistralai/ministral-8b",
+	"mistralai/mistral-7b-instruct",
+	"mistralai/mistral-7b-instruct-v0.1",
+	"mistralai/mistral-7b-instruct-v0.3",
+	"mistralai/mistral-7b-instruct:free",
+	"mistralai/mistral-large",
+	"mistralai/mistral-large-2407",
+	"mistralai/mistral-large-2411",
+	"mistralai/mistral-medium",
+	"mistralai/mistral-nemo",
+	"mistralai/mistral-nemo:free",
+	"mistralai/mistral-small",
+	"mistralai/mistral-small-24b-instruct-2501",
+	"mistralai/mistral-small-24b-instruct-2501:free",
+	"mistralai/mistral-tiny",
+	"mistralai/mixtral-8x22b-instruct",
+	"mistralai/mixtral-8x7b",
+	"mistralai/mixtral-8x7b-instruct",
+	"mistralai/pixtral-12b",
+	"mistralai/pixtral-large-2411",
+	"neversleep/llama-3-lumimaid-70b",
+	"neversleep/llama-3-lumimaid-8b",
+	"neversleep/llama-3-lumimaid-8b:extended",
+	"neversleep/llama-3.1-lumimaid-70b",
+	"neversleep/llama-3.1-lumimaid-8b",
+	"neversleep/noromaid-20b",
+	"nothingiisreal/mn-celeste-12b",
+	"nousresearch/hermes-2-pro-llama-3-8b",
+	"nousresearch/hermes-3-llama-3.1-405b",
+	"nousresearch/hermes-3-llama-3.1-70b",
+	"nousresearch/nous-hermes-2-mixtral-8x7b-dpo",
+	"nousresearch/nous-hermes-llama2-13b",
+	"nvidia/llama-3.1-nemotron-70b-instruct",
+	"nvidia/llama-3.1-nemotron-70b-instruct:free",
+	"openai/chatgpt-4o-latest",
+	"openai/gpt-3.5-turbo",
+	"openai/gpt-3.5-turbo-0125",
+	"openai/gpt-3.5-turbo-0613",
+	"openai/gpt-3.5-turbo-1106",
+	"openai/gpt-3.5-turbo-16k",
+	"openai/gpt-3.5-turbo-instruct",
+	"openai/gpt-4",
+	"openai/gpt-4-0314",
+	"openai/gpt-4-1106-preview",
+	"openai/gpt-4-32k",
+	"openai/gpt-4-32k-0314",
+	"openai/gpt-4-turbo",
+	"openai/gpt-4-turbo-preview",
+	"openai/gpt-4o",
+	"openai/gpt-4o-2024-05-13",
+	"openai/gpt-4o-2024-08-06",
+	"openai/gpt-4o-2024-11-20",
+	"openai/gpt-4o-mini",
+	"openai/gpt-4o-mini-2024-07-18",
+	"openai/gpt-4o:extended",
+	"openai/o1",
+	"openai/o1-mini",
+	"openai/o1-mini-2024-09-12",
+	"openai/o1-preview",
+	"openai/o1-preview-2024-09-12",
+	"openai/o3-mini",
+	"openai/o3-mini-high",
+	"openchat/openchat-7b",
+	"openchat/openchat-7b:free",
+	"openrouter/auto",
+	"perplexity/llama-3.1-sonar-huge-128k-online",
+	"perplexity/llama-3.1-sonar-large-128k-chat",
+	"perplexity/llama-3.1-sonar-large-128k-online",
+	"perplexity/llama-3.1-sonar-small-128k-chat",
+	"perplexity/llama-3.1-sonar-small-128k-online",
+	"perplexity/sonar",
+	"perplexity/sonar-reasoning",
+	"pygmalionai/mythalion-13b",
+	"qwen/qvq-72b-preview",
+	"qwen/qwen-2-72b-instruct",
+	"qwen/qwen-2-7b-instruct",
+	"qwen/qwen-2-7b-instruct:free",
+	"qwen/qwen-2-vl-72b-instruct",
+	"qwen/qwen-2-vl-7b-instruct",
+	"qwen/qwen-2.5-72b-instruct",
+	"qwen/qwen-2.5-7b-instruct",
+	"qwen/qwen-2.5-coder-32b-instruct",
+	"qwen/qwen-max",
+	"qwen/qwen-plus",
+	"qwen/qwen-turbo",
 	"qwen/qwen-vl-plus:free",
+	"qwen/qwen2.5-vl-72b-instruct:free",
+	"qwen/qwq-32b-preview",
+	"raifle/sorcererlm-8x22b",
+	"sao10k/fimbulvetr-11b-v2",
+	"sao10k/l3-euryale-70b",
+	"sao10k/l3-lunaris-8b",
+	"sao10k/l3.1-70b-hanami-x1",
+	"sao10k/l3.1-euryale-70b",
+	"sao10k/l3.3-euryale-70b",
+	"sophosympatheia/midnight-rose-70b",
+	"sophosympatheia/rogue-rose-103b-v0.2:free",
+	"teknium/openhermes-2.5-mistral-7b",
+	"thedrummer/rocinante-12b",
+	"thedrummer/unslopnemo-12b",
+	"undi95/remm-slerp-l2-13b",
+	"undi95/toppy-m-7b",
+	"undi95/toppy-m-7b:free",
+	"x-ai/grok-2-1212",
+	"x-ai/grok-2-vision-1212",
+	"x-ai/grok-beta",
+	"x-ai/grok-vision-beta",
+	"xwin-lm/xwin-lm-70b",
 }

relay/adaptor/vertexai/claude/adapter.go

@@ -36,7 +36,7 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 		AnthropicVersion: anthropicVersion,
 		// Model:            claudeReq.Model,
 		Messages:    claudeReq.Messages,
-		System:      claudeReq.System,
+		System:      claudeReq.System.String(), // Convert SystemPrompt to string
 		MaxTokens:   claudeReq.MaxTokens,
 		Temperature: claudeReq.Temperature,
 		TopP:        claudeReq.TopP,

relay/billing/ratio/model.go

@@ -59,6 +59,8 @@ var ModelRatio = map[string]float64{
 	"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,
@@ -159,91 +161,105 @@ var ModelRatio = map[string]float64{
 	"embedding-2":      0.0005 * RMB,
 	"embedding-3":      0.0005 * RMB,
 	// https://help.aliyun.com/zh/dashscope/developer-reference/tongyi-thousand-questions-metering-and-billing
-	"qwen-turbo":                  1.4286, // ￥0.02 / 1k tokens
+	"qwen-turbo":                    0.0003 * RMB,
-	"qwen-turbo-latest":           1.4286,
+	"qwen-turbo-latest":             0.0003 * RMB,
-	"qwen-plus":                   1.4286,
+	"qwen-plus":                     0.0008 * RMB,
-	"qwen-plus-latest":            1.4286,
+	"qwen-plus-latest":              0.0008 * RMB,
-	"qwen-max":                    1.4286,
+	"qwen-max":                      0.0024 * RMB,
-	"qwen-max-latest":             1.4286,
+	"qwen-max-latest":               0.0024 * RMB,
-	"qwen-max-longcontext":        1.4286,
+	"qwen-max-longcontext":          0.0005 * RMB,
-	"qwen-vl-max":                 1.4286,
+	"qwen-vl-max":                   0.003 * RMB,
-	"qwen-vl-max-latest":          1.4286,
+	"qwen-vl-max-latest":            0.003 * RMB,
-	"qwen-vl-plus":                1.4286,
+	"qwen-vl-plus":                  0.0015 * RMB,
-	"qwen-vl-plus-latest":         1.4286,
+	"qwen-vl-plus-latest":           0.0015 * RMB,
-	"qwen-vl-ocr":                 1.4286,
+	"qwen-vl-ocr":                   0.005 * RMB,
-	"qwen-vl-ocr-latest":          1.4286,
+	"qwen-vl-ocr-latest":            0.005 * RMB,
-	"qwen-audio-turbo":            1.4286,
+	"qwen-audio-turbo":              1.4286,
-	"qwen-math-plus":              1.4286,
+	"qwen-math-plus":                0.004 * RMB,
-	"qwen-math-plus-latest":       1.4286,
+	"qwen-math-plus-latest":         0.004 * RMB,
-	"qwen-math-turbo":             1.4286,
+	"qwen-math-turbo":               0.002 * RMB,
-	"qwen-math-turbo-latest":      1.4286,
+	"qwen-math-turbo-latest":        0.002 * RMB,
-	"qwen-coder-plus":             1.4286,
+	"qwen-coder-plus":               0.0035 * RMB,
-	"qwen-coder-plus-latest":      1.4286,
+	"qwen-coder-plus-latest":        0.0035 * RMB,
-	"qwen-coder-turbo":            1.4286,
+	"qwen-coder-turbo":              0.002 * RMB,
-	"qwen-coder-turbo-latest":     1.4286,
+	"qwen-coder-turbo-latest":       0.002 * RMB,
-	"qwq-32b-preview":             1.4286,
+	"qwen-mt-plus":                  0.015 * RMB,
-	"qwen2.5-72b-instruct":        1.4286,
+	"qwen-mt-turbo":                 0.001 * RMB,
-	"qwen2.5-32b-instruct":        1.4286,
+	"qwq-32b-preview":               0.002 * RMB,
-	"qwen2.5-14b-instruct":        1.4286,
+	"qwen2.5-72b-instruct":          0.004 * RMB,
-	"qwen2.5-7b-instruct":         1.4286,
+	"qwen2.5-32b-instruct":          0.03 * RMB,
-	"qwen2.5-3b-instruct":         1.4286,
+	"qwen2.5-14b-instruct":          0.001 * RMB,
-	"qwen2.5-1.5b-instruct":       1.4286,
+	"qwen2.5-7b-instruct":           0.0005 * RMB,
-	"qwen2.5-0.5b-instruct":       1.4286,
+	"qwen2.5-3b-instruct":           0.006 * RMB,
-	"qwen2-72b-instruct":          1.4286,
+	"qwen2.5-1.5b-instruct":         0.0003 * RMB,
-	"qwen2-57b-a14b-instruct":     1.4286,
+	"qwen2.5-0.5b-instruct":         0.0003 * RMB,
-	"qwen2-7b-instruct":           1.4286,
+	"qwen2-72b-instruct":            0.004 * RMB,
-	"qwen2-1.5b-instruct":         1.4286,
+	"qwen2-57b-a14b-instruct":       0.0035 * RMB,
-	"qwen2-0.5b-instruct":         1.4286,
+	"qwen2-7b-instruct":             0.001 * RMB,
-	"qwen1.5-110b-chat":           1.4286,
+	"qwen2-1.5b-instruct":           0.001 * RMB,
-	"qwen1.5-72b-chat":            1.4286,
+	"qwen2-0.5b-instruct":           0.001 * RMB,
-	"qwen1.5-32b-chat":            1.4286,
+	"qwen1.5-110b-chat":             0.007 * RMB,
-	"qwen1.5-14b-chat":            1.4286,
+	"qwen1.5-72b-chat":              0.005 * RMB,
-	"qwen1.5-7b-chat":             1.4286,
+	"qwen1.5-32b-chat":              0.0035 * RMB,
-	"qwen1.5-1.8b-chat":           1.4286,
+	"qwen1.5-14b-chat":              0.002 * RMB,
-	"qwen1.5-0.5b-chat":           1.4286,
+	"qwen1.5-7b-chat":               0.001 * RMB,
-	"qwen-72b-chat":               1.4286,
+	"qwen1.5-1.8b-chat":             0.001 * RMB,
-	"qwen-14b-chat":               1.4286,
+	"qwen1.5-0.5b-chat":             0.001 * RMB,
-	"qwen-7b-chat":                1.4286,
+	"qwen-72b-chat":                 0.02 * RMB,
-	"qwen-1.8b-chat":              1.4286,
+	"qwen-14b-chat":                 0.008 * RMB,
-	"qwen-1.8b-longcontext-chat":  1.4286,
+	"qwen-7b-chat":                  0.006 * RMB,
-	"qwen2-vl-7b-instruct":        1.4286,
+	"qwen-1.8b-chat":                0.006 * RMB,
-	"qwen2-vl-2b-instruct":        1.4286,
+	"qwen-1.8b-longcontext-chat":    0.006 * RMB,
-	"qwen-vl-v1":                  1.4286,
+	"qvq-72b-preview":               0.012 * RMB,
-	"qwen-vl-chat-v1":             1.4286,
+	"qwen2.5-vl-72b-instruct":       0.016 * RMB,
-	"qwen2-audio-instruct":        1.4286,
+	"qwen2.5-vl-7b-instruct":        0.002 * RMB,
-	"qwen-audio-chat":             1.4286,
+	"qwen2.5-vl-3b-instruct":        0.0012 * RMB,
-	"qwen2.5-math-72b-instruct":   1.4286,
+	"qwen2-vl-7b-instruct":          0.016 * RMB,
-	"qwen2.5-math-7b-instruct":    1.4286,
+	"qwen2-vl-2b-instruct":          0.002 * RMB,
-	"qwen2.5-math-1.5b-instruct":  1.4286,
+	"qwen-vl-v1":                    0.002 * RMB,
-	"qwen2-math-72b-instruct":     1.4286,
+	"qwen-vl-chat-v1":               0.002 * RMB,
-	"qwen2-math-7b-instruct":      1.4286,
+	"qwen2-audio-instruct":          0.002 * RMB,
-	"qwen2-math-1.5b-instruct":    1.4286,
+	"qwen-audio-chat":               0.002 * RMB,
-	"qwen2.5-coder-32b-instruct":  1.4286,
+	"qwen2.5-math-72b-instruct":     0.004 * RMB,
-	"qwen2.5-coder-14b-instruct":  1.4286,
+	"qwen2.5-math-7b-instruct":      0.001 * RMB,
-	"qwen2.5-coder-7b-instruct":   1.4286,
+	"qwen2.5-math-1.5b-instruct":    0.001 * RMB,
-	"qwen2.5-coder-3b-instruct":   1.4286,
+	"qwen2-math-72b-instruct":       0.004 * RMB,
-	"qwen2.5-coder-1.5b-instruct": 1.4286,
+	"qwen2-math-7b-instruct":        0.001 * RMB,
-	"qwen2.5-coder-0.5b-instruct": 1.4286,
+	"qwen2-math-1.5b-instruct":      0.001 * RMB,
-	"text-embedding-v1":           0.05, // ￥0.0007 / 1k tokens
+	"qwen2.5-coder-32b-instruct":    0.002 * RMB,
-	"text-embedding-v3":           0.05,
+	"qwen2.5-coder-14b-instruct":    0.002 * RMB,
-	"text-embedding-v2":           0.05,
+	"qwen2.5-coder-7b-instruct":     0.001 * RMB,
-	"text-embedding-async-v2":     0.05,
+	"qwen2.5-coder-3b-instruct":     0.001 * RMB,
-	"text-embedding-async-v1":     0.05,
+	"qwen2.5-coder-1.5b-instruct":   0.001 * RMB,
-	"ali-stable-diffusion-xl":     8.00,
+	"qwen2.5-coder-0.5b-instruct":   0.001 * RMB,
-	"ali-stable-diffusion-v1.5":   8.00,
+	"text-embedding-v1":             0.0007 * RMB, // ￥0.0007 / 1k tokens
-	"wanx-v1":                     8.00,
+	"text-embedding-v3":             0.0007 * RMB,
-	"SparkDesk":                   1.2858, // ￥0.018 / 1k tokens
+	"text-embedding-v2":             0.0007 * RMB,
-	"SparkDesk-v1.1":              1.2858, // ￥0.018 / 1k tokens
+	"text-embedding-async-v2":       0.0007 * RMB,
-	"SparkDesk-v2.1":              1.2858, // ￥0.018 / 1k tokens
+	"text-embedding-async-v1":       0.0007 * RMB,
-	"SparkDesk-v3.1":              1.2858, // ￥0.018 / 1k tokens
+	"ali-stable-diffusion-xl":       8.00,
-	"SparkDesk-v3.1-128K":         1.2858, // ￥0.018 / 1k tokens
+	"ali-stable-diffusion-v1.5":     8.00,
-	"SparkDesk-v3.5":              1.2858, // ￥0.018 / 1k tokens
+	"wanx-v1":                       8.00,
-	"SparkDesk-v3.5-32K":          1.2858, // ￥0.018 / 1k tokens
+	"deepseek-r1":                   0.002 * RMB,
-	"SparkDesk-v4.0":              1.2858, // ￥0.018 / 1k tokens
+	"deepseek-v3":                   0.001 * RMB,
-	"360GPT_S2_V9":                0.8572, // ¥0.012 / 1k tokens
+	"deepseek-r1-distill-qwen-1.5b": 0.001 * RMB,
-	"embedding-bert-512-v1":       0.0715, // ¥0.001 / 1k tokens
+	"deepseek-r1-distill-qwen-7b":   0.0005 * RMB,
-	"embedding_s1_v1":             0.0715, // ¥0.001 / 1k tokens
+	"deepseek-r1-distill-qwen-14b":  0.001 * RMB,
-	"semantic_similarity_s1_v1":   0.0715, // ¥0.001 / 1k tokens
+	"deepseek-r1-distill-qwen-32b":  0.002 * RMB,
+	"deepseek-r1-distill-llama-8b":  0.0005 * RMB,
+	"deepseek-r1-distill-llama-70b": 0.004 * RMB,
+	"SparkDesk":                     1.2858, // ￥0.018 / 1k tokens
+	"SparkDesk-v1.1":                1.2858, // ￥0.018 / 1k tokens
+	"SparkDesk-v2.1":                1.2858, // ￥0.018 / 1k tokens
+	"SparkDesk-v3.1":                1.2858, // ￥0.018 / 1k tokens
+	"SparkDesk-v3.1-128K":           1.2858, // ￥0.018 / 1k tokens
+	"SparkDesk-v3.5":                1.2858, // ￥0.018 / 1k tokens
+	"SparkDesk-v3.5-32K":            1.2858, // ￥0.018 / 1k tokens
+	"SparkDesk-v4.0":                1.2858, // ￥0.018 / 1k tokens
+	"360GPT_S2_V9":                  0.8572, // ¥0.012 / 1k tokens
+	"embedding-bert-512-v1":         0.0715, // ¥0.001 / 1k tokens
+	"embedding_s1_v1":               0.0715, // ¥0.001 / 1k tokens
+	"semantic_similarity_s1_v1":     0.0715, // ¥0.001 / 1k tokens
 	// https://cloud.tencent.com/document/product/1729/97731#e0e6be58-60c8-469f-bdeb-6c264ce3b4d0
 	"hunyuan-turbo":             0.015 * RMB,
 	"hunyuan-large":             0.004 * RMB,
@@ -371,6 +387,238 @@ var ModelRatio = map[string]float64{
 	"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,
+	//https://openrouter.ai/models
+	"01-ai/yi-large":                                  1.5,
+	"aetherwiing/mn-starcannon-12b":                   0.6,
+	"ai21/jamba-1-5-large":                            4.0,
+	"ai21/jamba-1-5-mini":                             0.2,
+	"ai21/jamba-instruct":                             0.35,
+	"aion-labs/aion-1.0":                              6.0,
+	"aion-labs/aion-1.0-mini":                         1.2,
+	"aion-labs/aion-rp-llama-3.1-8b":                  0.1,
+	"allenai/llama-3.1-tulu-3-405b":                   5.0,
+	"alpindale/goliath-120b":                          4.6875,
+	"alpindale/magnum-72b":                            1.125,
+	"amazon/nova-lite-v1":                             0.12,
+	"amazon/nova-micro-v1":                            0.07,
+	"amazon/nova-pro-v1":                              1.6,
+	"anthracite-org/magnum-v2-72b":                    1.5,
+	"anthracite-org/magnum-v4-72b":                    1.125,
+	"anthropic/claude-2":                              12.0,
+	"anthropic/claude-2.0":                            12.0,
+	"anthropic/claude-2.0:beta":                       12.0,
+	"anthropic/claude-2.1":                            12.0,
+	"anthropic/claude-2.1:beta":                       12.0,
+	"anthropic/claude-2:beta":                         12.0,
+	"anthropic/claude-3-haiku":                        0.625,
+	"anthropic/claude-3-haiku:beta":                   0.625,
+	"anthropic/claude-3-opus":                         37.5,
+	"anthropic/claude-3-opus:beta":                    37.5,
+	"anthropic/claude-3-sonnet":                       7.5,
+	"anthropic/claude-3-sonnet:beta":                  7.5,
+	"anthropic/claude-3.5-haiku":                      2.0,
+	"anthropic/claude-3.5-haiku-20241022":             2.0,
+	"anthropic/claude-3.5-haiku-20241022:beta":        2.0,
+	"anthropic/claude-3.5-haiku:beta":                 2.0,
+	"anthropic/claude-3.5-sonnet":                     7.5,
+	"anthropic/claude-3.5-sonnet-20240620":            7.5,
+	"anthropic/claude-3.5-sonnet-20240620:beta":       7.5,
+	"anthropic/claude-3.5-sonnet:beta":                7.5,
+	"cognitivecomputations/dolphin-mixtral-8x22b":     0.45,
+	"cognitivecomputations/dolphin-mixtral-8x7b":      0.25,
+	"cohere/command":                                  0.95,
+	"cohere/command-r":                                0.7125,
+	"cohere/command-r-03-2024":                        0.7125,
+	"cohere/command-r-08-2024":                        0.285,
+	"cohere/command-r-plus":                           7.125,
+	"cohere/command-r-plus-04-2024":                   7.125,
+	"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-v2.5":                     1.0,
+	"deepseek/deepseek-chat:free":                     0.0,
+	"deepseek/deepseek-r1":                            1.2,
+	"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,
+	"deepseek/deepseek-r1-distill-qwen-1.5b":          0.09,
+	"deepseek/deepseek-r1-distill-qwen-14b":           0.075,
+	"deepseek/deepseek-r1-distill-qwen-32b":           0.09,
+	"deepseek/deepseek-r1:free":                       0.0,
+	"eva-unit-01/eva-llama-3.33-70b":                  3.0,
+	"eva-unit-01/eva-qwen-2.5-32b":                    1.7,
+	"eva-unit-01/eva-qwen-2.5-72b":                    3.0,
+	"google/gemini-2.0-flash-001":                     0.2,
+	"google/gemini-2.0-flash-exp:free":                0.0,
+	"google/gemini-2.0-flash-lite-preview-02-05:free": 0.0,
+	"google/gemini-2.0-flash-thinking-exp-1219:free":  0.0,
+	"google/gemini-2.0-flash-thinking-exp:free":       0.0,
+	"google/gemini-2.0-pro-exp-02-05:free":            0.0,
+	"google/gemini-exp-1206:free":                     0.0,
+	"google/gemini-flash-1.5":                         0.15,
+	"google/gemini-flash-1.5-8b":                      0.075,
+	"google/gemini-flash-1.5-8b-exp":                  0.0,
+	"google/gemini-pro":                               0.75,
+	"google/gemini-pro-1.5":                           2.5,
+	"google/gemini-pro-vision":                        0.75,
+	"google/gemma-2-27b-it":                           0.135,
+	"google/gemma-2-9b-it":                            0.03,
+	"google/gemma-2-9b-it:free":                       0.0,
+	"google/gemma-7b-it":                              0.075,
+	"google/learnlm-1.5-pro-experimental:free":        0.0,
+	"google/palm-2-chat-bison":                        1.0,
+	"google/palm-2-chat-bison-32k":                    1.0,
+	"google/palm-2-codechat-bison":                    1.0,
+	"google/palm-2-codechat-bison-32k":                1.0,
+	"gryphe/mythomax-l2-13b":                          0.0325,
+	"gryphe/mythomax-l2-13b:free":                     0.0,
+	"huggingfaceh4/zephyr-7b-beta:free":               0.0,
+	"infermatic/mn-inferor-12b":                       0.6,
+	"inflection/inflection-3-pi":                      5.0,
+	"inflection/inflection-3-productivity":            5.0,
+	"jondurbin/airoboros-l2-70b":                      0.25,
+	"liquid/lfm-3b":                                   0.01,
+	"liquid/lfm-40b":                                  0.075,
+	"liquid/lfm-7b":                                   0.005,
+	"mancer/weaver":                                   1.125,
+	"meta-llama/llama-2-13b-chat":                     0.11,
+	"meta-llama/llama-2-70b-chat":                     0.45,
+	"meta-llama/llama-3-70b-instruct":                 0.2,
+	"meta-llama/llama-3-8b-instruct":                  0.03,
+	"meta-llama/llama-3-8b-instruct:free":             0.0,
+	"meta-llama/llama-3.1-405b":                       1.0,
+	"meta-llama/llama-3.1-405b-instruct":              0.4,
+	"meta-llama/llama-3.1-70b-instruct":               0.15,
+	"meta-llama/llama-3.1-8b-instruct":                0.025,
+	"meta-llama/llama-3.2-11b-vision-instruct":        0.0275,
+	"meta-llama/llama-3.2-11b-vision-instruct:free":   0.0,
+	"meta-llama/llama-3.2-1b-instruct":                0.005,
+	"meta-llama/llama-3.2-3b-instruct":                0.0125,
+	"meta-llama/llama-3.2-90b-vision-instruct":        0.8,
+	"meta-llama/llama-3.3-70b-instruct":               0.15,
+	"meta-llama/llama-3.3-70b-instruct:free":          0.0,
+	"meta-llama/llama-guard-2-8b":                     0.1,
+	"microsoft/phi-3-medium-128k-instruct":            0.5,
+	"microsoft/phi-3-medium-128k-instruct:free":       0.0,
+	"microsoft/phi-3-mini-128k-instruct":              0.05,
+	"microsoft/phi-3-mini-128k-instruct:free":         0.0,
+	"microsoft/phi-3.5-mini-128k-instruct":            0.05,
+	"microsoft/phi-4":                                 0.07,
+	"microsoft/wizardlm-2-7b":                         0.035,
+	"microsoft/wizardlm-2-8x22b":                      0.25,
+	"minimax/minimax-01":                              0.55,
+	"mistralai/codestral-2501":                        0.45,
+	"mistralai/codestral-mamba":                       0.125,
+	"mistralai/ministral-3b":                          0.02,
+	"mistralai/ministral-8b":                          0.05,
+	"mistralai/mistral-7b-instruct":                   0.0275,
+	"mistralai/mistral-7b-instruct-v0.1":              0.1,
+	"mistralai/mistral-7b-instruct-v0.3":              0.0275,
+	"mistralai/mistral-7b-instruct:free":              0.0,
+	"mistralai/mistral-large":                         3.0,
+	"mistralai/mistral-large-2407":                    3.0,
+	"mistralai/mistral-large-2411":                    3.0,
+	"mistralai/mistral-medium":                        4.05,
+	"mistralai/mistral-nemo":                          0.04,
+	"mistralai/mistral-nemo:free":                     0.0,
+	"mistralai/mistral-small":                         0.3,
+	"mistralai/mistral-small-24b-instruct-2501":       0.07,
+	"mistralai/mistral-small-24b-instruct-2501:free":  0.0,
+	"mistralai/mistral-tiny":                          0.125,
+	"mistralai/mixtral-8x22b-instruct":                0.45,
+	"mistralai/mixtral-8x7b":                          0.3,
+	"mistralai/mixtral-8x7b-instruct":                 0.12,
+	"mistralai/pixtral-12b":                           0.05,
+	"mistralai/pixtral-large-2411":                    3.0,
+	"neversleep/llama-3-lumimaid-70b":                 2.25,
+	"neversleep/llama-3-lumimaid-8b":                  0.5625,
+	"neversleep/llama-3-lumimaid-8b:extended":         0.5625,
+	"neversleep/llama-3.1-lumimaid-70b":               2.25,
+	"neversleep/llama-3.1-lumimaid-8b":                0.5625,
+	"neversleep/noromaid-20b":                         1.125,
+	"nothingiisreal/mn-celeste-12b":                   0.6,
+	"nousresearch/hermes-2-pro-llama-3-8b":            0.02,
+	"nousresearch/hermes-3-llama-3.1-405b":            0.4,
+	"nousresearch/hermes-3-llama-3.1-70b":             0.15,
+	"nousresearch/nous-hermes-2-mixtral-8x7b-dpo":     0.3,
+	"nousresearch/nous-hermes-llama2-13b":             0.085,
+	"nvidia/llama-3.1-nemotron-70b-instruct":          0.15,
+	"nvidia/llama-3.1-nemotron-70b-instruct:free":     0.0,
+	"openai/chatgpt-4o-latest":                        7.5,
+	"openai/gpt-3.5-turbo":                            0.75,
+	"openai/gpt-3.5-turbo-0125":                       0.75,
+	"openai/gpt-3.5-turbo-0613":                       1.0,
+	"openai/gpt-3.5-turbo-1106":                       1.0,
+	"openai/gpt-3.5-turbo-16k":                        2.0,
+	"openai/gpt-3.5-turbo-instruct":                   1.0,
+	"openai/gpt-4":                                    30.0,
+	"openai/gpt-4-0314":                               30.0,
+	"openai/gpt-4-1106-preview":                       15.0,
+	"openai/gpt-4-32k":                                60.0,
+	"openai/gpt-4-32k-0314":                           60.0,
+	"openai/gpt-4-turbo":                              15.0,
+	"openai/gpt-4-turbo-preview":                      15.0,
+	"openai/gpt-4o":                                   5.0,
+	"openai/gpt-4o-2024-05-13":                        7.5,
+	"openai/gpt-4o-2024-08-06":                        5.0,
+	"openai/gpt-4o-2024-11-20":                        5.0,
+	"openai/gpt-4o-mini":                              0.3,
+	"openai/gpt-4o-mini-2024-07-18":                   0.3,
+	"openai/gpt-4o:extended":                          9.0,
+	"openai/o1":                                       30.0,
+	"openai/o1-mini":                                  2.2,
+	"openai/o1-mini-2024-09-12":                       2.2,
+	"openai/o1-preview":                               30.0,
+	"openai/o1-preview-2024-09-12":                    30.0,
+	"openai/o3-mini":                                  2.2,
+	"openai/o3-mini-high":                             2.2,
+	"openchat/openchat-7b":                            0.0275,
+	"openchat/openchat-7b:free":                       0.0,
+	"openrouter/auto":                                 -500000.0,
+	"perplexity/llama-3.1-sonar-huge-128k-online":     2.5,
+	"perplexity/llama-3.1-sonar-large-128k-chat":      0.5,
+	"perplexity/llama-3.1-sonar-large-128k-online":    0.5,
+	"perplexity/llama-3.1-sonar-small-128k-chat":      0.1,
+	"perplexity/llama-3.1-sonar-small-128k-online":    0.1,
+	"perplexity/sonar":                                0.5,
+	"perplexity/sonar-reasoning":                      2.5,
+	"pygmalionai/mythalion-13b":                       0.6,
+	"qwen/qvq-72b-preview":                            0.25,
+	"qwen/qwen-2-72b-instruct":                        0.45,
+	"qwen/qwen-2-7b-instruct":                         0.027,
+	"qwen/qwen-2-7b-instruct:free":                    0.0,
+	"qwen/qwen-2-vl-72b-instruct":                     0.2,
+	"qwen/qwen-2-vl-7b-instruct":                      0.05,
+	"qwen/qwen-2.5-72b-instruct":                      0.2,
+	"qwen/qwen-2.5-7b-instruct":                       0.025,
+	"qwen/qwen-2.5-coder-32b-instruct":                0.08,
+	"qwen/qwen-max":                                   3.2,
+	"qwen/qwen-plus":                                  0.6,
+	"qwen/qwen-turbo":                                 0.1,
+	"qwen/qwen-vl-plus:free":                          0.0,
+	"qwen/qwen2.5-vl-72b-instruct:free":               0.0,
+	"qwen/qwq-32b-preview":                            0.09,
+	"raifle/sorcererlm-8x22b":                         2.25,
+	"sao10k/fimbulvetr-11b-v2":                        0.6,
+	"sao10k/l3-euryale-70b":                           0.4,
+	"sao10k/l3-lunaris-8b":                            0.03,
+	"sao10k/l3.1-70b-hanami-x1":                       1.5,
+	"sao10k/l3.1-euryale-70b":                         0.4,
+	"sao10k/l3.3-euryale-70b":                         0.4,
+	"sophosympatheia/midnight-rose-70b":               0.4,
+	"sophosympatheia/rogue-rose-103b-v0.2:free":       0.0,
+	"teknium/openhermes-2.5-mistral-7b":               0.085,
+	"thedrummer/rocinante-12b":                        0.25,
+	"thedrummer/unslopnemo-12b":                       0.25,
+	"undi95/remm-slerp-l2-13b":                        0.6,
+	"undi95/toppy-m-7b":                               0.035,
+	"undi95/toppy-m-7b:free":                          0.0,
+	"x-ai/grok-2-1212":                                5.0,
+	"x-ai/grok-2-vision-1212":                         5.0,
+	"x-ai/grok-beta":                                  7.5,
+	"x-ai/grok-vision-beta":                           7.5,
+	"xwin-lm/xwin-lm-70b":                             1.875,
 }
 
 var CompletionRatio = map[string]float64{

relay/controller/anthropic.go

@@ -0,0 +1,224 @@
+package controller
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+
+	"github.com/gin-gonic/gin"
+
+	"github.com/songquanpeng/one-api/common"
+	"github.com/songquanpeng/one-api/common/config"
+	"github.com/songquanpeng/one-api/common/logger"
+	dbmodel "github.com/songquanpeng/one-api/model"
+	"github.com/songquanpeng/one-api/relay"
+	"github.com/songquanpeng/one-api/relay/adaptor/anthropic"
+	"github.com/songquanpeng/one-api/relay/adaptor/openai"
+	"github.com/songquanpeng/one-api/relay/billing"
+	billingratio "github.com/songquanpeng/one-api/relay/billing/ratio"
+	"github.com/songquanpeng/one-api/relay/meta"
+	"github.com/songquanpeng/one-api/relay/model"
+)
+
+// RelayAnthropicHelper handles native Anthropic API requests (anthropic -> anthropic passthrough)
+func RelayAnthropicHelper(c *gin.Context) *model.ErrorWithStatusCode {
+	ctx := c.Request.Context()
+	meta := meta.GetByContext(c)
+
+	logger.Infof(ctx, "Anthropic request received - URL: %s", c.Request.URL.String())
+
+	// get & validate anthropic request
+	anthropicRequest, err := getAndValidateAnthropicRequest(c)
+	if err != nil {
+		logger.Errorf(ctx, "getAndValidateAnthropicRequest failed: %s", err.Error())
+		return openai.ErrorWrapper(err, "invalid_anthropic_request", http.StatusBadRequest)
+	}
+	logger.Debugf(ctx, "Parsed anthropic request - Model: %s, Stream: %v, Messages: %d",
+		anthropicRequest.Model, anthropicRequest.Stream, len(anthropicRequest.Messages))
+	meta.IsStream = anthropicRequest.Stream
+
+	// map model name
+	meta.OriginModelName = anthropicRequest.Model
+	mappedModel, _ := getMappedModelName(anthropicRequest.Model, meta.ModelMapping)
+	anthropicRequest.Model = mappedModel
+	meta.ActualModelName = anthropicRequest.Model
+
+	// estimate token usage for anthropic request
+	promptTokens := estimateAnthropicTokens(anthropicRequest)
+	meta.PromptTokens = promptTokens
+
+	// get model ratio & group ratio
+	modelRatio := billingratio.GetModelRatio(anthropicRequest.Model, meta.ChannelType)
+	groupRatio := billingratio.GetGroupRatio(meta.Group)
+	ratio := modelRatio * groupRatio
+
+	// pre-consume quota
+	preConsumedQuota, bizErr := preConsumeQuotaForAnthropic(ctx, anthropicRequest, promptTokens, ratio, meta)
+	if bizErr != nil {
+		logger.Warnf(ctx, "preConsumeQuota failed: %+v", *bizErr)
+		return bizErr
+	}
+
+	logger.Debugf(ctx, "Meta info - APIType: %d, ChannelType: %d, BaseURL: %s", meta.APIType, meta.ChannelType, meta.BaseURL)
+
+	adaptor := relay.GetAdaptor(meta.APIType)
+	if adaptor == nil {
+		logger.Errorf(ctx, "Failed to get adaptor for API type: %d", meta.APIType)
+		return openai.ErrorWrapper(fmt.Errorf("invalid api type: %d", meta.APIType), "invalid_api_type", http.StatusBadRequest)
+	}
+	logger.Debugf(ctx, "Using adaptor: %s", adaptor.GetChannelName())
+	adaptor.Init(meta)
+
+	// get request body - for anthropic passthrough, we directly use the request body
+	requestBody, err := getAnthropicRequestBody(c, anthropicRequest)
+	if err != nil {
+		return openai.ErrorWrapper(err, "convert_anthropic_request_failed", http.StatusInternalServerError)
+	}
+
+	// do request
+	resp, err := adaptor.DoRequest(c, meta, requestBody)
+	if err != nil {
+		logger.Errorf(ctx, "DoRequest failed: %s", err.Error())
+		return openai.ErrorWrapper(err, "do_request_failed", http.StatusInternalServerError)
+	}
+	logger.Debugf(ctx, "Received response - Status: %d", resp.StatusCode)
+
+	if isErrorHappened(meta, resp) {
+		logger.Errorf(ctx, "Error detected in response")
+		billing.ReturnPreConsumedQuota(ctx, preConsumedQuota, meta.TokenId)
+		return RelayErrorHandler(resp)
+	}
+
+	// do response - for anthropic native requests, we need to handle the response directly
+	usage, respErr := handleAnthropicResponse(c, resp, meta)
+	if respErr != nil {
+		logger.Errorf(ctx, "respErr is not nil: %+v", respErr)
+		billing.ReturnPreConsumedQuota(ctx, preConsumedQuota, meta.TokenId)
+		return respErr
+	}
+
+	logger.Infof(ctx, "Anthropic request completed - Usage: %+v", usage)
+
+	// post-consume quota - for anthropic, we create a placeholder GeneralOpenAIRequest
+	placeholderRequest := &model.GeneralOpenAIRequest{
+		Model: anthropicRequest.Model,
+	}
+	go postConsumeQuota(ctx, usage, meta, placeholderRequest, ratio, preConsumedQuota, modelRatio, groupRatio, false)
+	return nil
+}
+
+func getAndValidateAnthropicRequest(c *gin.Context) (*anthropic.Request, error) {
+	anthropicRequest := &anthropic.Request{}
+	err := common.UnmarshalBodyReusable(c, anthropicRequest)
+	if err != nil {
+		return nil, err
+	}
+
+	// Basic validation
+	if anthropicRequest.Model == "" {
+		return nil, fmt.Errorf("model is required")
+	}
+	if len(anthropicRequest.Messages) == 0 {
+		return nil, fmt.Errorf("messages are required")
+	}
+	if anthropicRequest.MaxTokens == 0 {
+		anthropicRequest.MaxTokens = 4096 // default max tokens
+	}
+
+	return anthropicRequest, nil
+}
+
+func getAnthropicRequestBody(c *gin.Context, anthropicRequest *anthropic.Request) (io.Reader, error) {
+	// For anthropic native requests, we marshal the request back to JSON
+	jsonData, err := json.Marshal(anthropicRequest)
+	if err != nil {
+		logger.Debugf(c.Request.Context(), "anthropic request json_marshal_failed: %s\n", err.Error())
+		return nil, err
+	}
+	logger.Debugf(c.Request.Context(), "anthropic request: \n%s", string(jsonData))
+	return bytes.NewBuffer(jsonData), nil
+}
+
+const (
+	// CHARS_PER_TOKEN represents the rough character-to-token ratio for Anthropic models
+	// This is a conservative estimate: approximately 1 token per 4 characters
+	CHARS_PER_TOKEN = 4
+)
+
+func estimateAnthropicTokens(request *anthropic.Request) int {
+	// Simple token estimation for Anthropic requests
+	// This is a rough estimation, actual implementation might need more sophisticated logic
+	totalTokens := 0
+
+	// Count tokens in system prompt
+	if !request.System.IsEmpty() {
+		systemText := request.System.String()
+		totalTokens += len(systemText) / CHARS_PER_TOKEN // rough estimate: 1 token per 4 characters
+	}
+
+	// Count tokens in messages
+	for _, message := range request.Messages {
+		for _, content := range message.Content {
+			if content.Type == "text" {
+				totalTokens += len(content.Text) / CHARS_PER_TOKEN
+			}
+		}
+	}
+
+	return totalTokens
+}
+
+func handleAnthropicResponse(c *gin.Context, resp *http.Response, meta *meta.Meta) (*model.Usage, *model.ErrorWithStatusCode) {
+	// For anthropic native requests, use direct handlers to maintain Anthropic format
+	if meta.IsStream {
+		// Handle streaming response - note: DirectStreamHandler returns (error, usage)
+		err, usage := anthropic.DirectStreamHandler(c, resp)
+		return usage, err
+	} else {
+		// Handle non-streaming response - note: DirectHandler returns (error, usage)
+		err, usage := anthropic.DirectHandler(c, resp, meta.PromptTokens, meta.ActualModelName)
+		return usage, err
+	}
+}
+
+func preConsumeQuotaForAnthropic(ctx context.Context, request *anthropic.Request, promptTokens int, ratio float64, meta *meta.Meta) (int64, *model.ErrorWithStatusCode) {
+	// Use the same quota logic as text requests but adapted for Anthropic
+	preConsumedTokens := config.PreConsumedQuota + int64(promptTokens)
+	if request.MaxTokens != 0 {
+		preConsumedTokens += int64(request.MaxTokens)
+	}
+	preConsumedQuota := int64(float64(preConsumedTokens) * ratio)
+
+	userQuota, err := dbmodel.CacheGetUserQuota(ctx, meta.UserId)
+	if err != nil {
+		return preConsumedQuota, openai.ErrorWrapper(err, "get_user_quota_failed", http.StatusInternalServerError)
+	}
+
+	if userQuota-preConsumedQuota < 0 {
+		return preConsumedQuota, openai.ErrorWrapper(fmt.Errorf("user quota is not enough"), "insufficient_user_quota", http.StatusForbidden)
+	}
+
+	err = dbmodel.CacheDecreaseUserQuota(meta.UserId, preConsumedQuota)
+	if err != nil {
+		return preConsumedQuota, openai.ErrorWrapper(err, "decrease_user_quota_failed", http.StatusInternalServerError)
+	}
+
+	if userQuota > 100*preConsumedQuota {
+		// in this case, we do not pre-consume quota
+		// because the user has enough quota
+		preConsumedQuota = 0
+		logger.Info(ctx, fmt.Sprintf("user %d has enough quota %d, trusted and no need to pre-consume", meta.UserId, userQuota))
+	}
+
+	if preConsumedQuota > 0 {
+		err := dbmodel.PreConsumeTokenQuota(meta.TokenId, preConsumedQuota)
+		if err != nil {
+			return preConsumedQuota, openai.ErrorWrapper(err, "pre_consume_token_quota_failed", http.StatusForbidden)
+		}
+	}
+
+	return preConsumedQuota, nil
+}

relay/meta/relay_meta.go

@@ -8,6 +8,7 @@ import (
 
 	"github.com/songquanpeng/one-api/common/ctxkey"
 	"github.com/songquanpeng/one-api/model"
+	"github.com/songquanpeng/one-api/relay/apitype"
 	"github.com/songquanpeng/one-api/relay/channeltype"
 	"github.com/songquanpeng/one-api/relay/relaymode"
 )
@@ -62,5 +63,11 @@ func GetByContext(c *gin.Context) *Meta {
 		meta.BaseURL = channeltype.ChannelBaseURLs[meta.ChannelType]
 	}
 	meta.APIType = channeltype.ToAPIType(meta.ChannelType)
+
+	// Force Anthropic API type for native Anthropic protocol requests
+	if meta.Mode == relaymode.AnthropicMessages {
+		meta.APIType = apitype.Anthropic
+	}
+
 	return &meta
 }

relay/relaymode/define.go

@@ -13,4 +13,6 @@ const (
 	AudioTranslation
 	// Proxy is a special relay mode for proxying requests to custom upstream
 	Proxy
+	// AnthropicMessages is for native Anthropic API messages endpoint
+	AnthropicMessages
 )

relay/relaymode/helper.go

@@ -26,6 +26,8 @@ func GetByPath(path string) int {
 		relayMode = AudioTranslation
 	} else if strings.HasPrefix(path, "/v1/oneapi/proxy") {
 		relayMode = Proxy
+	} else if strings.HasPrefix(path, "/anthropic/v1/messages") {
+		relayMode = AnthropicMessages
 	}
 	return relayMode
 }

router/relay.go

@@ -71,4 +71,16 @@ func SetRelayRouter(router *gin.Engine) {
 		relayV1Router.GET("/threads/:id/runs/:runsId/steps/:stepId", controller.RelayNotImplemented)
 		relayV1Router.GET("/threads/:id/runs/:runsId/steps", controller.RelayNotImplemented)
 	}
+
+	// Anthropic API compatibility - https://docs.anthropic.com/claude/reference/
+	anthropicRouter := router.Group("/anthropic")
+	anthropicRouter.Use(middleware.RelayPanicRecover(), middleware.TokenAuth(), middleware.Distribute())
+	{
+		// Models API
+		anthropicRouter.GET("/v1/models", controller.ListModels)
+		anthropicRouter.GET("/v1/models/:model", controller.RetrieveModel)
+
+		// Messages API - main endpoint for chat completions
+		anthropicRouter.POST("/v1/messages", controller.Relay)
+	}
 }