Merge branch 'upstream-main' into fix/pr-2540

# Conflicts:
#	relay/channel/gemini/relay-gemini.go

relay/audio_handler.go

@@ -70,7 +70,7 @@ func AudioHelper(c *gin.Context, info *relaycommon.RelayInfo) (newAPIError *type
 	if usage.(*dto.Usage).CompletionTokenDetails.AudioTokens > 0 || usage.(*dto.Usage).PromptTokensDetails.AudioTokens > 0 {
 		service.PostAudioConsumeQuota(c, info, usage.(*dto.Usage), "")
 	} else {
-		postConsumeQuota(c, info, usage.(*dto.Usage), "")
+		postConsumeQuota(c, info, usage.(*dto.Usage))
 	}
 
 	return nil

relay/channel/ali/adaptor.go

@@ -13,12 +13,37 @@ import (
 	"github.com/QuantumNous/new-api/relay/channel/openai"
 	relaycommon "github.com/QuantumNous/new-api/relay/common"
 	"github.com/QuantumNous/new-api/relay/constant"
+	"github.com/QuantumNous/new-api/setting/model_setting"
+	"github.com/QuantumNous/new-api/service"
 	"github.com/QuantumNous/new-api/types"
 
 	"github.com/gin-gonic/gin"
 )
 
 type Adaptor struct {
+	IsSyncImageModel bool
+}
+
+/*
+	var syncModels = []string{
+		"z-image",
+		"qwen-image",
+		"wan2.6",
+	}
+*/
+func supportsAliAnthropicMessages(modelName string) bool {
+	// Only models with the "qwen" designation can use the Claude-compatible interface; others require conversion.
+	return strings.Contains(strings.ToLower(modelName), "qwen")
+}
+
+var syncModels = []string{
+	"z-image",
+	"qwen-image",
+	"wan2.6",
+}
+
+func isSyncImageModel(modelName string) bool {
+	return model_setting.IsSyncImageModel(modelName)
 }
 
 func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dto.GeminiChatRequest) (any, error) {
@@ -27,7 +52,18 @@ func (a *Adaptor) ConvertGeminiRequest(*gin.Context, *relaycommon.RelayInfo, *dt
 }
 
 func (a *Adaptor) ConvertClaudeRequest(c *gin.Context, info *relaycommon.RelayInfo, req *dto.ClaudeRequest) (any, error) {
-	return req, nil
+	if supportsAliAnthropicMessages(info.UpstreamModelName) {
+		return req, nil
+	}
+
+	oaiReq, err := service.ClaudeToOpenAIRequest(*req, info)
+	if err != nil {
+		return nil, err
+	}
+	if info.SupportStreamOptions && info.IsStream {
+		oaiReq.StreamOptions = &dto.StreamOptions{IncludeUsage: true}
+	}
+	return a.ConvertOpenAIRequest(c, info, oaiReq)
 }
 
 func (a *Adaptor) Init(info *relaycommon.RelayInfo) {
@@ -37,7 +73,11 @@ func (a *Adaptor) GetRequestURL(info *relaycommon.RelayInfo) (string, error) {
 	var fullRequestURL string
 	switch info.RelayFormat {
 	case types.RelayFormatClaude:
-		fullRequestURL = fmt.Sprintf("%s/api/v2/apps/claude-code-proxy/v1/messages", info.ChannelBaseUrl)
+		if supportsAliAnthropicMessages(info.UpstreamModelName) {
+			fullRequestURL = fmt.Sprintf("%s/apps/anthropic/v1/messages", info.ChannelBaseUrl)
+		} else {
+			fullRequestURL = fmt.Sprintf("%s/compatible-mode/v1/chat/completions", info.ChannelBaseUrl)
+		}
 	default:
 		switch info.RelayMode {
 		case constant.RelayModeEmbeddings:
@@ -45,10 +85,16 @@ func (a *Adaptor) GetRequestURL(info *relaycommon.RelayInfo) (string, error) {
 		case constant.RelayModeRerank:
 			fullRequestURL = fmt.Sprintf("%s/api/v1/services/rerank/text-rerank/text-rerank", info.ChannelBaseUrl)
 		case constant.RelayModeImagesGenerations:
-			fullRequestURL = fmt.Sprintf("%s/api/v1/services/aigc/text2image/image-synthesis", info.ChannelBaseUrl)
+			if isSyncImageModel(info.OriginModelName) {
+				fullRequestURL = fmt.Sprintf("%s/api/v1/services/aigc/multimodal-generation/generation", info.ChannelBaseUrl)
+			} else {
+				fullRequestURL = fmt.Sprintf("%s/api/v1/services/aigc/text2image/image-synthesis", info.ChannelBaseUrl)
+			}
 		case constant.RelayModeImagesEdits:
-			if isWanModel(info.OriginModelName) {
+			if isOldWanModel(info.OriginModelName) {
 				fullRequestURL = fmt.Sprintf("%s/api/v1/services/aigc/image2image/image-synthesis", info.ChannelBaseUrl)
+			} else if isWanModel(info.OriginModelName) {
+				fullRequestURL = fmt.Sprintf("%s/api/v1/services/aigc/image-generation/generation", info.ChannelBaseUrl)
 			} else {
 				fullRequestURL = fmt.Sprintf("%s/api/v1/services/aigc/multimodal-generation/generation", info.ChannelBaseUrl)
 			}
@@ -72,7 +118,11 @@ func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Header, info *rel
 		req.Set("X-DashScope-Plugin", c.GetString("plugin"))
 	}
 	if info.RelayMode == constant.RelayModeImagesGenerations {
-		req.Set("X-DashScope-Async", "enable")
+		if isSyncImageModel(info.OriginModelName) {
+
+		} else {
+			req.Set("X-DashScope-Async", "enable")
+		}
 	}
 	if info.RelayMode == constant.RelayModeImagesEdits {
 		if isWanModel(info.OriginModelName) {
@@ -108,15 +158,25 @@ func (a *Adaptor) ConvertOpenAIRequest(c *gin.Context, info *relaycommon.RelayIn
 
 func (a *Adaptor) ConvertImageRequest(c *gin.Context, info *relaycommon.RelayInfo, request dto.ImageRequest) (any, error) {
 	if info.RelayMode == constant.RelayModeImagesGenerations {
-		aliRequest, err := oaiImage2Ali(request)
+		if isSyncImageModel(info.OriginModelName) {
+			a.IsSyncImageModel = true
+		}
+		aliRequest, err := oaiImage2AliImageRequest(info, request, a.IsSyncImageModel)
 		if err != nil {
-			return nil, fmt.Errorf("convert image request failed: %w", err)
+			return nil, fmt.Errorf("convert image request to async ali image request failed: %w", err)
 		}
 		return aliRequest, nil
 	} else if info.RelayMode == constant.RelayModeImagesEdits {
-		if isWanModel(info.OriginModelName) {
+		if isOldWanModel(info.OriginModelName) {
 			return oaiFormEdit2WanxImageEdit(c, info, request)
 		}
+		if isSyncImageModel(info.OriginModelName) {
+			if isWanModel(info.OriginModelName) {
+				a.IsSyncImageModel = false
+			} else {
+				a.IsSyncImageModel = true
+			}
+		}
 		// ali image edit https://bailian.console.aliyun.com/?tab=api#/api/?type=model&url=2976416
 		// 如果用户使用表单，则需要解析表单数据
 		if strings.Contains(c.Request.Header.Get("Content-Type"), "multipart/form-data") {
@@ -126,9 +186,9 @@ func (a *Adaptor) ConvertImageRequest(c *gin.Context, info *relaycommon.RelayInf
 			}
 			return aliRequest, nil
 		} else {
-			aliRequest, err := oaiImage2Ali(request)
+			aliRequest, err := oaiImage2AliImageRequest(info, request, a.IsSyncImageModel)
 			if err != nil {
-				return nil, fmt.Errorf("convert image request failed: %w", err)
+				return nil, fmt.Errorf("convert image request to async ali image request failed: %w", err)
 			}
 			return aliRequest, nil
 		}
@@ -150,7 +210,7 @@ func (a *Adaptor) ConvertAudioRequest(c *gin.Context, info *relaycommon.RelayInf
 }
 
 func (a *Adaptor) ConvertOpenAIResponsesRequest(c *gin.Context, info *relaycommon.RelayInfo, request dto.OpenAIResponsesRequest) (any, error) {
-	// TODO implement me
+	//TODO implement me
 	return nil, errors.New("not implemented")
 }
 
@@ -161,21 +221,22 @@ func (a *Adaptor) DoRequest(c *gin.Context, info *relaycommon.RelayInfo, request
 func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (usage any, err *types.NewAPIError) {
 	switch info.RelayFormat {
 	case types.RelayFormatClaude:
-		if info.IsStream {
-			return claude.ClaudeStreamHandler(c, resp, info, claude.RequestModeMessage)
-		} else {
+		if supportsAliAnthropicMessages(info.UpstreamModelName) {
+			if info.IsStream {
+				return claude.ClaudeStreamHandler(c, resp, info, claude.RequestModeMessage)
+			}
+
 			return claude.ClaudeHandler(c, resp, info, claude.RequestModeMessage)
 		}
+
+		adaptor := openai.Adaptor{}
+		return adaptor.DoResponse(c, resp, info)
 	default:
 		switch info.RelayMode {
 		case constant.RelayModeImagesGenerations:
-			err, usage = aliImageHandler(c, resp, info)
+			err, usage = aliImageHandler(a, c, resp, info)
 		case constant.RelayModeImagesEdits:
-			if isWanModel(info.OriginModelName) {
-				err, usage = aliImageHandler(c, resp, info)
-			} else {
-				err, usage = aliImageEditHandler(c, resp, info)
-			}
+			err, usage = aliImageHandler(a, c, resp, info)
 		case constant.RelayModeRerank:
 			err, usage = RerankHandler(c, resp, info)
 		default:

relay/channel/ali/dto.go

@@ -1,6 +1,13 @@
 package ali
 
-import "github.com/QuantumNous/new-api/dto"
+import (
+	"strings"
+
+	"github.com/QuantumNous/new-api/dto"
+	"github.com/QuantumNous/new-api/logger"
+	"github.com/QuantumNous/new-api/service"
+	"github.com/gin-gonic/gin"
+)
 
 type AliMessage struct {
 	Content any    `json:"content"`
@@ -65,6 +72,7 @@ type AliUsage struct {
 	InputTokens  int `json:"input_tokens"`
 	OutputTokens int `json:"output_tokens"`
 	TotalTokens  int `json:"total_tokens"`
+	ImageCount   int `json:"image_count,omitempty"`
 }
 
 type TaskResult struct {
@@ -75,14 +83,78 @@ type TaskResult struct {
 }
 
 type AliOutput struct {
-	TaskId       string           `json:"task_id,omitempty"`
-	TaskStatus   string           `json:"task_status,omitempty"`
-	Text         string           `json:"text"`
-	FinishReason string           `json:"finish_reason"`
-	Message      string           `json:"message,omitempty"`
-	Code         string           `json:"code,omitempty"`
-	Results      []TaskResult     `json:"results,omitempty"`
-	Choices      []map[string]any `json:"choices,omitempty"`
+	TaskId       string       `json:"task_id,omitempty"`
+	TaskStatus   string       `json:"task_status,omitempty"`
+	Text         string       `json:"text"`
+	FinishReason string       `json:"finish_reason"`
+	Message      string       `json:"message,omitempty"`
+	Code         string       `json:"code,omitempty"`
+	Results      []TaskResult `json:"results,omitempty"`
+	Choices      []struct {
+		FinishReason string `json:"finish_reason,omitempty"`
+		Message      struct {
+			Role             string            `json:"role,omitempty"`
+			Content          []AliMediaContent `json:"content,omitempty"`
+			ReasoningContent string            `json:"reasoning_content,omitempty"`
+		} `json:"message,omitempty"`
+	} `json:"choices,omitempty"`
+}
+
+func (o *AliOutput) ChoicesToOpenAIImageDate(c *gin.Context, responseFormat string) []dto.ImageData {
+	var imageData []dto.ImageData
+	if len(o.Choices) > 0 {
+		for _, choice := range o.Choices {
+			var data dto.ImageData
+			for _, content := range choice.Message.Content {
+				if content.Image != "" {
+					if strings.HasPrefix(content.Image, "http") {
+						var b64Json string
+						if responseFormat == "b64_json" {
+							_, b64, err := service.GetImageFromUrl(content.Image)
+							if err != nil {
+								logger.LogError(c, "get_image_data_failed: "+err.Error())
+								continue
+							}
+							b64Json = b64
+						}
+						data.Url = content.Image
+						data.B64Json = b64Json
+					} else {
+						data.B64Json = content.Image
+					}
+				} else if content.Text != "" {
+					data.RevisedPrompt = content.Text
+				}
+			}
+			imageData = append(imageData, data)
+		}
+	}
+
+	return imageData
+}
+
+func (o *AliOutput) ResultToOpenAIImageDate(c *gin.Context, responseFormat string) []dto.ImageData {
+	var imageData []dto.ImageData
+	for _, data := range o.Results {
+		var b64Json string
+		if responseFormat == "b64_json" {
+			_, b64, err := service.GetImageFromUrl(data.Url)
+			if err != nil {
+				logger.LogError(c, "get_image_data_failed: "+err.Error())
+				continue
+			}
+			b64Json = b64
+		} else {
+			b64Json = data.B64Image
+		}
+
+		imageData = append(imageData, dto.ImageData{
+			Url:           data.Url,
+			B64Json:       b64Json,
+			RevisedPrompt: "",
+		})
+	}
+	return imageData
 }
 
 type AliResponse struct {
@@ -92,18 +164,26 @@ type AliResponse struct {
 }
 
 type AliImageRequest struct {
-	Model          string `json:"model"`
-	Input          any    `json:"input"`
-	Parameters     any    `json:"parameters,omitempty"`
-	ResponseFormat string `json:"response_format,omitempty"`
+	Model          string             `json:"model"`
+	Input          any                `json:"input"`
+	Parameters     AliImageParameters `json:"parameters,omitempty"`
+	ResponseFormat string             `json:"response_format,omitempty"`
 }
 
 type AliImageParameters struct {
-	Size      string `json:"size,omitempty"`
-	N         int    `json:"n,omitempty"`
-	Steps     string `json:"steps,omitempty"`
-	Scale     string `json:"scale,omitempty"`
-	Watermark *bool  `json:"watermark,omitempty"`
+	Size         string `json:"size,omitempty"`
+	N            int    `json:"n,omitempty"`
+	Steps        string `json:"steps,omitempty"`
+	Scale        string `json:"scale,omitempty"`
+	Watermark    *bool  `json:"watermark,omitempty"`
+	PromptExtend *bool  `json:"prompt_extend,omitempty"`
+}
+
+func (p *AliImageParameters) PromptExtendValue() bool {
+	if p != nil && p.PromptExtend != nil {
+		return *p.PromptExtend
+	}
+	return false
 }
 
 type AliImageInput struct {

relay/channel/ali/image.go

@@ -1,7 +1,6 @@
 package ali
 
 import (
-	"context"
 	"encoding/base64"
 	"errors"
 	"fmt"
@@ -21,17 +20,23 @@ import (
 	"github.com/gin-gonic/gin"
 )
 
-func oaiImage2Ali(request dto.ImageRequest) (*AliImageRequest, error) {
+func oaiImage2AliImageRequest(info *relaycommon.RelayInfo, request dto.ImageRequest, isSync bool) (*AliImageRequest, error) {
 	var imageRequest AliImageRequest
 	imageRequest.Model = request.Model
 	imageRequest.ResponseFormat = request.ResponseFormat
-	logger.LogJson(context.Background(), "oaiImage2Ali request extra", request.Extra)
 	if request.Extra != nil {
 		if val, ok := request.Extra["parameters"]; ok {
 			err := common.Unmarshal(val, &imageRequest.Parameters)
 			if err != nil {
 				return nil, fmt.Errorf("invalid parameters field: %w", err)
 			}
+		} else {
+			// 兼容没有parameters字段的情况，从openai标准字段中提取参数
+			imageRequest.Parameters = AliImageParameters{
+				Size:      strings.Replace(request.Size, "x", "*", -1),
+				N:         int(request.N),
+				Watermark: request.Watermark,
+			}
 		}
 		if val, ok := request.Extra["input"]; ok {
 			err := common.Unmarshal(val, &imageRequest.Input)
@@ -41,23 +46,44 @@ func oaiImage2Ali(request dto.ImageRequest) (*AliImageRequest, error) {
 		}
 	}
 
-	if imageRequest.Parameters == nil {
-		imageRequest.Parameters = AliImageParameters{
-			Size:      strings.Replace(request.Size, "x", "*", -1),
-			N:         int(request.N),
-			Watermark: request.Watermark,
+	if strings.Contains(request.Model, "z-image") {
+		// z-image 开启prompt_extend后，按2倍计费
+		if imageRequest.Parameters.PromptExtendValue() {
+			info.PriceData.AddOtherRatio("prompt_extend", 2)
 		}
 	}
 
-	if imageRequest.Input == nil {
-		imageRequest.Input = AliImageInput{
-			Prompt: request.Prompt,
+	// 检查n参数
+	if imageRequest.Parameters.N != 0 {
+		info.PriceData.AddOtherRatio("n", float64(imageRequest.Parameters.N))
+	}
+
+	// 同步图片模型和异步图片模型请求格式不一样
+	if isSync {
+		if imageRequest.Input == nil {
+			imageRequest.Input = AliImageInput{
+				Messages: []AliMessage{
+					{
+						Role: "user",
+						Content: []AliMediaContent{
+							{
+								Text: request.Prompt,
+							},
+						},
+					},
+				},
+			}
+		}
+	} else {
+		if imageRequest.Input == nil {
+			imageRequest.Input = AliImageInput{
+				Prompt: request.Prompt,
+			}
 		}
 	}
 
 	return &imageRequest, nil
 }
-
 func getImageBase64sFromForm(c *gin.Context, fieldName string) ([]string, error) {
 	mf := c.Request.MultipartForm
 	if mf == nil {
@@ -199,6 +225,8 @@ func asyncTaskWait(c *gin.Context, info *relaycommon.RelayInfo, taskID string) (
 	var taskResponse AliResponse
 	var responseBody []byte
 
+	time.Sleep(time.Duration(5) * time.Second)
+
 	for {
 		logger.LogDebug(c, fmt.Sprintf("asyncTaskWait step %d/%d, wait %d seconds", step, maxStep, waitSeconds))
 		step++
@@ -238,32 +266,17 @@ func responseAli2OpenAIImage(c *gin.Context, response *AliResponse, originBody [
 		Created: info.StartTime.Unix(),
 	}
 
-	for _, data := range response.Output.Results {
-		var b64Json string
-		if responseFormat == "b64_json" {
-			_, b64, err := service.GetImageFromUrl(data.Url)
-			if err != nil {
-				logger.LogError(c, "get_image_data_failed: "+err.Error())
-				continue
-			}
-			b64Json = b64
-		} else {
-			b64Json = data.B64Image
-		}
-
-		imageResponse.Data = append(imageResponse.Data, dto.ImageData{
-			Url:           data.Url,
-			B64Json:       b64Json,
-			RevisedPrompt: "",
-		})
+	if len(response.Output.Results) > 0 {
+		imageResponse.Data = response.Output.ResultToOpenAIImageDate(c, responseFormat)
+	} else if len(response.Output.Choices) > 0 {
+		imageResponse.Data = response.Output.ChoicesToOpenAIImageDate(c, responseFormat)
 	}
-	var mapResponse map[string]any
-	_ = common.Unmarshal(originBody, &mapResponse)
-	imageResponse.Extra = mapResponse
+
+	imageResponse.Metadata = originBody
 	return &imageResponse
 }
 
-func aliImageHandler(c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (*types.NewAPIError, *dto.Usage) {
+func aliImageHandler(a *Adaptor, c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (*types.NewAPIError, *dto.Usage) {
 	responseFormat := c.GetString("response_format")
 
 	var aliTaskResponse AliResponse
@@ -282,66 +295,49 @@ func aliImageHandler(c *gin.Context, resp *http.Response, info *relaycommon.Rela
 		return types.NewError(errors.New(aliTaskResponse.Message), types.ErrorCodeBadResponse), nil
 	}
 
-	aliResponse, originRespBody, err := asyncTaskWait(c, info, aliTaskResponse.Output.TaskId)
-	if err != nil {
-		return types.NewError(err, types.ErrorCodeBadResponse), nil
-	}
+	var (
+		aliResponse    *AliResponse
+		originRespBody []byte
+	)
 
-	if aliResponse.Output.TaskStatus != "SUCCEEDED" {
-		return types.WithOpenAIError(types.OpenAIError{
-			Message: aliResponse.Output.Message,
-			Type:    "ali_error",
-			Param:   "",
-			Code:    aliResponse.Output.Code,
-		}, resp.StatusCode), nil
-	}
-
-	fullTextResponse := responseAli2OpenAIImage(c, aliResponse, originRespBody, info, responseFormat)
-	jsonResponse, err := common.Marshal(fullTextResponse)
-	if err != nil {
-		return types.NewError(err, types.ErrorCodeBadResponseBody), nil
-	}
-	service.IOCopyBytesGracefully(c, resp, jsonResponse)
-	return nil, &dto.Usage{}
-}
-
-func aliImageEditHandler(c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (*types.NewAPIError, *dto.Usage) {
-	var aliResponse AliResponse
-	responseBody, err := io.ReadAll(resp.Body)
-	if err != nil {
-		return types.NewOpenAIError(err, types.ErrorCodeReadResponseBodyFailed, http.StatusInternalServerError), nil
-	}
-
-	service.CloseResponseBodyGracefully(resp)
-	err = common.Unmarshal(responseBody, &aliResponse)
-	if err != nil {
-		return types.NewOpenAIError(err, types.ErrorCodeBadResponseBody, http.StatusInternalServerError), nil
-	}
-
-	if aliResponse.Message != "" {
-		logger.LogError(c, "ali_task_failed: "+aliResponse.Message)
-		return types.NewError(errors.New(aliResponse.Message), types.ErrorCodeBadResponse), nil
-	}
-	var fullTextResponse dto.ImageResponse
-	if len(aliResponse.Output.Choices) > 0 {
-		fullTextResponse = dto.ImageResponse{
-			Created: info.StartTime.Unix(),
-			Data: []dto.ImageData{
-				{
-					Url:     aliResponse.Output.Choices[0]["message"].(map[string]any)["content"].([]any)[0].(map[string]any)["image"].(string),
-					B64Json: "",
-				},
-			},
+	if a.IsSyncImageModel {
+		aliResponse = &aliTaskResponse
+		originRespBody = responseBody
+	} else {
+		// 异步图片模型需要轮询任务结果
+		aliResponse, originRespBody, err = asyncTaskWait(c, info, aliTaskResponse.Output.TaskId)
+		if err != nil {
+			return types.NewError(err, types.ErrorCodeBadResponse), nil
+		}
+		if aliResponse.Output.TaskStatus != "SUCCEEDED" {
+			return types.WithOpenAIError(types.OpenAIError{
+				Message: aliResponse.Output.Message,
+				Type:    "ali_error",
+				Param:   "",
+				Code:    aliResponse.Output.Code,
+			}, resp.StatusCode), nil
 		}
 	}
 
-	var mapResponse map[string]any
-	_ = common.Unmarshal(responseBody, &mapResponse)
-	fullTextResponse.Extra = mapResponse
-	jsonResponse, err := common.Marshal(fullTextResponse)
+	//logger.LogDebug(c, "ali_async_task_result: "+string(originRespBody))
+	if a.IsSyncImageModel {
+		logger.LogDebug(c, "ali_sync_image_result: "+string(originRespBody))
+	} else {
+		logger.LogDebug(c, "ali_async_image_result: "+string(originRespBody))
+	}
+
+	imageResponses := responseAli2OpenAIImage(c, aliResponse, originRespBody, info, responseFormat)
+	// 可能生成多张图片，修正计费数量n
+	if aliResponse.Usage.ImageCount != 0 {
+		info.PriceData.AddOtherRatio("n", float64(aliResponse.Usage.ImageCount))
+	} else if len(imageResponses.Data) != 0 {
+		info.PriceData.AddOtherRatio("n", float64(len(imageResponses.Data)))
+	}
+	jsonResponse, err := common.Marshal(imageResponses)
 	if err != nil {
 		return types.NewError(err, types.ErrorCodeBadResponseBody), nil
 	}
 	service.IOCopyBytesGracefully(c, resp, jsonResponse)
+
 	return nil, &dto.Usage{}
 }

relay/channel/ali/image_wan.go

@@ -26,14 +26,22 @@ func oaiFormEdit2WanxImageEdit(c *gin.Context, info *relaycommon.RelayInfo, requ
 	if wanInput.Images, err = getImageBase64sFromForm(c, "image"); err != nil {
 		return nil, fmt.Errorf("get image base64s from form failed: %w", err)
 	}
-	wanParams := WanImageParameters{
+	//wanParams := WanImageParameters{
+	//	N: int(request.N),
+	//}
+	imageRequest.Input = wanInput
+	imageRequest.Parameters = AliImageParameters{
 		N: int(request.N),
 	}
-	imageRequest.Input = wanInput
-	imageRequest.Parameters = wanParams
+	info.PriceData.AddOtherRatio("n", float64(imageRequest.Parameters.N))
+
 	return &imageRequest, nil
 }
 
+func isOldWanModel(modelName string) bool {
+	return strings.Contains(modelName, "wan") && !strings.Contains(modelName, "wan2.6")
+}
+
 func isWanModel(modelName string) bool {
 	return strings.Contains(modelName, "wan")
 }

relay/channel/aws/relay-aws.go

@@ -1,11 +1,13 @@
 package aws
 
 import (
+	"context"
 	"encoding/json"
 	"fmt"
 	"io"
 	"net/http"
 	"strings"
+	"time"
 
 	"github.com/QuantumNous/new-api/common"
 	"github.com/QuantumNous/new-api/dto"
@@ -37,6 +39,13 @@ func getAwsErrorStatusCode(err error) int {
 	return http.StatusInternalServerError
 }
 
+func newAwsInvokeContext() (context.Context, context.CancelFunc) {
+	if common.RelayTimeout <= 0 {
+		return context.Background(), func() {}
+	}
+	return context.WithTimeout(context.Background(), time.Duration(common.RelayTimeout)*time.Second)
+}
+
 func newAwsClient(c *gin.Context, info *relaycommon.RelayInfo) (*bedrockruntime.Client, error) {
 	var (
 		httpClient *http.Client
@@ -117,6 +126,7 @@ func doAwsClientRequest(c *gin.Context, info *relaycommon.RelayInfo, a *Adaptor,
 			return nil, types.NewError(errors.Wrap(err, "marshal nova request"), types.ErrorCodeBadResponseBody)
 		}
 		awsReq.Body = reqBody
+		a.AwsReq = awsReq
 		return nil, nil
 	} else {
 		awsClaudeReq, err := formatRequest(requestBody, requestHeader)
@@ -201,7 +211,10 @@ func getAwsModelID(requestModel string) string {
 
 func awsHandler(c *gin.Context, info *relaycommon.RelayInfo, a *Adaptor) (*types.NewAPIError, *dto.Usage) {
 
-	awsResp, err := a.AwsClient.InvokeModel(c.Request.Context(), a.AwsReq.(*bedrockruntime.InvokeModelInput))
+	ctx, cancel := newAwsInvokeContext()
+	defer cancel()
+
+	awsResp, err := a.AwsClient.InvokeModel(ctx, a.AwsReq.(*bedrockruntime.InvokeModelInput))
 	if err != nil {
 		statusCode := getAwsErrorStatusCode(err)
 		return types.NewOpenAIError(errors.Wrap(err, "InvokeModel"), types.ErrorCodeAwsInvokeError, statusCode), nil
@@ -228,7 +241,10 @@ func awsHandler(c *gin.Context, info *relaycommon.RelayInfo, a *Adaptor) (*types
 }
 
 func awsStreamHandler(c *gin.Context, info *relaycommon.RelayInfo, a *Adaptor) (*types.NewAPIError, *dto.Usage) {
-	awsResp, err := a.AwsClient.InvokeModelWithResponseStream(c.Request.Context(), a.AwsReq.(*bedrockruntime.InvokeModelWithResponseStreamInput))
+	ctx, cancel := newAwsInvokeContext()
+	defer cancel()
+
+	awsResp, err := a.AwsClient.InvokeModelWithResponseStream(ctx, a.AwsReq.(*bedrockruntime.InvokeModelWithResponseStreamInput))
 	if err != nil {
 		statusCode := getAwsErrorStatusCode(err)
 		return types.NewOpenAIError(errors.Wrap(err, "InvokeModelWithResponseStream"), types.ErrorCodeAwsInvokeError, statusCode), nil
@@ -268,7 +284,10 @@ func awsStreamHandler(c *gin.Context, info *relaycommon.RelayInfo, a *Adaptor) (
 // Nova模型处理函数
 func handleNovaRequest(c *gin.Context, info *relaycommon.RelayInfo, a *Adaptor) (*types.NewAPIError, *dto.Usage) {
 
-	awsResp, err := a.AwsClient.InvokeModel(c.Request.Context(), a.AwsReq.(*bedrockruntime.InvokeModelInput))
+	ctx, cancel := newAwsInvokeContext()
+	defer cancel()
+
+	awsResp, err := a.AwsClient.InvokeModel(ctx, a.AwsReq.(*bedrockruntime.InvokeModelInput))
 	if err != nil {
 		statusCode := getAwsErrorStatusCode(err)
 		return types.NewOpenAIError(errors.Wrap(err, "InvokeModel"), types.ErrorCodeAwsInvokeError, statusCode), nil

relay/channel/codex/adaptor.go

@@ -0,0 +1,164 @@
+package codex
+
+import (
+	"encoding/json"
+	"errors"
+	"io"
+	"net/http"
+	"strings"
+
+	"github.com/QuantumNous/new-api/common"
+	"github.com/QuantumNous/new-api/dto"
+	"github.com/QuantumNous/new-api/relay/channel"
+	"github.com/QuantumNous/new-api/relay/channel/openai"
+	relaycommon "github.com/QuantumNous/new-api/relay/common"
+	relayconstant "github.com/QuantumNous/new-api/relay/constant"
+	"github.com/QuantumNous/new-api/types"
+
+	"github.com/gin-gonic/gin"
+)
+
+type Adaptor struct {
+}
+
+func (a *Adaptor) ConvertGeminiRequest(c *gin.Context, info *relaycommon.RelayInfo, request *dto.GeminiChatRequest) (any, error) {
+	return nil, errors.New("codex channel: endpoint not supported")
+}
+
+func (a *Adaptor) ConvertClaudeRequest(*gin.Context, *relaycommon.RelayInfo, *dto.ClaudeRequest) (any, error) {
+	return nil, errors.New("codex channel: endpoint not supported")
+}
+
+func (a *Adaptor) ConvertAudioRequest(c *gin.Context, info *relaycommon.RelayInfo, request dto.AudioRequest) (io.Reader, error) {
+	return nil, errors.New("codex channel: endpoint not supported")
+}
+
+func (a *Adaptor) ConvertImageRequest(c *gin.Context, info *relaycommon.RelayInfo, request dto.ImageRequest) (any, error) {
+	return nil, errors.New("codex channel: endpoint not supported")
+}
+
+func (a *Adaptor) Init(info *relaycommon.RelayInfo) {
+}
+
+func (a *Adaptor) ConvertOpenAIRequest(c *gin.Context, info *relaycommon.RelayInfo, request *dto.GeneralOpenAIRequest) (any, error) {
+	return nil, errors.New("codex channel: endpoint not supported")
+}
+
+func (a *Adaptor) ConvertRerankRequest(c *gin.Context, relayMode int, request dto.RerankRequest) (any, error) {
+	return nil, errors.New("codex channel: endpoint not supported")
+}
+
+func (a *Adaptor) ConvertEmbeddingRequest(c *gin.Context, info *relaycommon.RelayInfo, request dto.EmbeddingRequest) (any, error) {
+	return nil, errors.New("codex channel: endpoint not supported")
+}
+
+func (a *Adaptor) ConvertOpenAIResponsesRequest(c *gin.Context, info *relaycommon.RelayInfo, request dto.OpenAIResponsesRequest) (any, error) {
+	if info != nil && info.ChannelSetting.SystemPrompt != "" {
+		systemPrompt := info.ChannelSetting.SystemPrompt
+
+		if len(request.Instructions) == 0 {
+			if b, err := common.Marshal(systemPrompt); err == nil {
+				request.Instructions = b
+			} else {
+				return nil, err
+			}
+		} else if info.ChannelSetting.SystemPromptOverride {
+			var existing string
+			if err := common.Unmarshal(request.Instructions, &existing); err == nil {
+				existing = strings.TrimSpace(existing)
+				if existing == "" {
+					if b, err := common.Marshal(systemPrompt); err == nil {
+						request.Instructions = b
+					} else {
+						return nil, err
+					}
+				} else {
+					if b, err := common.Marshal(systemPrompt + "\n" + existing); err == nil {
+						request.Instructions = b
+					} else {
+						return nil, err
+					}
+				}
+			} else {
+				if b, err := common.Marshal(systemPrompt); err == nil {
+					request.Instructions = b
+				} else {
+					return nil, err
+				}
+			}
+		}
+	}
+
+	// codex: store must be false
+	request.Store = json.RawMessage("false")
+	// rm max_output_tokens
+	request.MaxOutputTokens = 0
+	request.Temperature = nil
+	return request, nil
+}
+
+func (a *Adaptor) DoRequest(c *gin.Context, info *relaycommon.RelayInfo, requestBody io.Reader) (any, error) {
+	return channel.DoApiRequest(a, c, info, requestBody)
+}
+
+func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (usage any, err *types.NewAPIError) {
+	if info.RelayMode != relayconstant.RelayModeResponses {
+		return nil, types.NewError(errors.New("codex channel: endpoint not supported"), types.ErrorCodeInvalidRequest)
+	}
+
+	if info.IsStream {
+		return openai.OaiResponsesStreamHandler(c, info, resp)
+	}
+	return openai.OaiResponsesHandler(c, info, resp)
+}
+
+func (a *Adaptor) GetModelList() []string {
+	return ModelList
+}
+
+func (a *Adaptor) GetChannelName() string {
+	return ChannelName
+}
+
+func (a *Adaptor) GetRequestURL(info *relaycommon.RelayInfo) (string, error) {
+	if info.RelayMode != relayconstant.RelayModeResponses {
+		return "", errors.New("codex channel: only /v1/responses is supported")
+	}
+	return relaycommon.GetFullRequestURL(info.ChannelBaseUrl, "/backend-api/codex/responses", info.ChannelType), nil
+}
+
+func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Header, info *relaycommon.RelayInfo) error {
+	channel.SetupApiRequestHeader(info, c, req)
+
+	key := strings.TrimSpace(info.ApiKey)
+	if !strings.HasPrefix(key, "{") {
+		return errors.New("codex channel: key must be a JSON object")
+	}
+
+	oauthKey, err := ParseOAuthKey(key)
+	if err != nil {
+		return err
+	}
+
+	accessToken := strings.TrimSpace(oauthKey.AccessToken)
+	accountID := strings.TrimSpace(oauthKey.AccountID)
+
+	if accessToken == "" {
+		return errors.New("codex channel: access_token is required")
+	}
+	if accountID == "" {
+		return errors.New("codex channel: account_id is required")
+	}
+
+	req.Set("Authorization", "Bearer "+accessToken)
+	req.Set("chatgpt-account-id", accountID)
+
+	if req.Get("OpenAI-Beta") == "" {
+		req.Set("OpenAI-Beta", "responses=experimental")
+	}
+	if req.Get("originator") == "" {
+		req.Set("originator", "codex_cli_rs")
+	}
+
+	return nil
+}

relay/channel/codex/constants.go

@@ -0,0 +1,9 @@
+package codex
+
+var ModelList = []string{
+	"gpt-5", "gpt-5-codex", "gpt-5-codex-mini",
+	"gpt-5.1", "gpt-5.1-codex", "gpt-5.1-codex-max", "gpt-5.1-codex-mini",
+	"gpt-5.2", "gpt-5.2-codex",
+}
+
+const ChannelName = "codex"

relay/channel/codex/oauth_key.go

@@ -0,0 +1,30 @@
+package codex
+
+import (
+	"errors"
+
+	"github.com/QuantumNous/new-api/common"
+)
+
+type OAuthKey struct {
+	IDToken      string `json:"id_token,omitempty"`
+	AccessToken  string `json:"access_token,omitempty"`
+	RefreshToken string `json:"refresh_token,omitempty"`
+
+	AccountID   string `json:"account_id,omitempty"`
+	LastRefresh string `json:"last_refresh,omitempty"`
+	Email       string `json:"email,omitempty"`
+	Type        string `json:"type,omitempty"`
+	Expired     string `json:"expired,omitempty"`
+}
+
+func ParseOAuthKey(raw string) (*OAuthKey, error) {
+	if raw == "" {
+		return nil, errors.New("codex channel: empty oauth key")
+	}
+	var key OAuthKey
+	if err := common.Unmarshal([]byte(raw), &key); err != nil {
+		return nil, errors.New("codex channel: invalid oauth key json")
+	}
+	return &key, nil
+}

relay/channel/gemini/relay-gemini.go

@@ -1,6 +1,7 @@
 package gemini
 
 import (
+	"context"
 	"encoding/json"
 	"errors"
 	"fmt"
@@ -8,6 +9,7 @@ import (
 	"net/http"
 	"strconv"
 	"strings"
+	"time"
 	"unicode/utf8"
 
 	"github.com/QuantumNous/new-api/common"
@@ -32,6 +34,7 @@ var geminiSupportedMimeTypes = map[string]bool{
 	"audio/wav":       true,
 	"image/png":       true,
 	"image/jpeg":      true,
+	"image/jpg":       true, // support old image/jpeg
 	"image/webp":      true,
 	"text/plain":      true,
 	"video/mov":       true,
@@ -381,7 +384,7 @@ func CovertOpenAI2Gemini(c *gin.Context, textRequest dto.GeneralOpenAIRequest, i
 	var system_content []string
 	//shouldAddDummyModelMessage := false
 	for _, message := range textRequest.Messages {
-		if message.Role == "system" {
+		if message.Role == "system" || message.Role == "developer" {
 			system_content = append(system_content, message.StringContent())
 			continue
 		} else if message.Role == "tool" || message.Role == "function" {
@@ -659,101 +662,84 @@ func getSupportedMimeTypesList() []string {
 	return keys
 }
 
+var geminiOpenAPISchemaAllowedFields = map[string]struct{}{
+	"anyOf":            {},
+	"default":          {},
+	"description":      {},
+	"enum":             {},
+	"example":          {},
+	"format":           {},
+	"items":            {},
+	"maxItems":         {},
+	"maxLength":        {},
+	"maxProperties":    {},
+	"maximum":          {},
+	"minItems":         {},
+	"minLength":        {},
+	"minProperties":    {},
+	"minimum":          {},
+	"nullable":         {},
+	"pattern":          {},
+	"properties":       {},
+	"propertyOrdering": {},
+	"required":         {},
+	"title":            {},
+	"type":             {},
+}
+
+const geminiFunctionSchemaMaxDepth = 64
+
 // cleanFunctionParameters recursively removes unsupported fields from Gemini function parameters.
 func cleanFunctionParameters(params interface{}) interface{} {
+	return cleanFunctionParametersWithDepth(params, 0)
+}
+
+func cleanFunctionParametersWithDepth(params interface{}, depth int) interface{} {
 	if params == nil {
 		return nil
 	}
 
+	if depth >= geminiFunctionSchemaMaxDepth {
+		return cleanFunctionParametersShallow(params)
+	}
+
 	switch v := params.(type) {
 	case map[string]interface{}:
-		// Create a copy to avoid modifying the original
-		cleanedMap := make(map[string]interface{})
+		// Keep only Gemini-supported OpenAPI schema subset fields (per official SDK Schema).
+		cleanedMap := make(map[string]interface{}, len(v))
 		for k, val := range v {
-			cleanedMap[k] = val
-		}
-
-		// Remove unsupported root-level fields
-		delete(cleanedMap, "default")
-		delete(cleanedMap, "exclusiveMaximum")
-		delete(cleanedMap, "exclusiveMinimum")
-		delete(cleanedMap, "$schema")
-		delete(cleanedMap, "additionalProperties")
-
-		// Check and clean 'format' for string types
-		if propType, typeExists := cleanedMap["type"].(string); typeExists && propType == "string" {
-			if formatValue, formatExists := cleanedMap["format"].(string); formatExists {
-				if formatValue != "enum" && formatValue != "date-time" {
-					delete(cleanedMap, "format")
-				}
+			if _, ok := geminiOpenAPISchemaAllowedFields[k]; ok {
+				cleanedMap[k] = val
 			}
 		}
 
+		normalizeGeminiSchemaTypeAndNullable(cleanedMap)
+
 		// Clean properties
 		if props, ok := cleanedMap["properties"].(map[string]interface{}); ok && props != nil {
 			cleanedProps := make(map[string]interface{})
 			for propName, propValue := range props {
-				cleanedProps[propName] = cleanFunctionParameters(propValue)
+				cleanedProps[propName] = cleanFunctionParametersWithDepth(propValue, depth+1)
 			}
 			cleanedMap["properties"] = cleanedProps
 		}
 
 		// Recursively clean items in arrays
 		if items, ok := cleanedMap["items"].(map[string]interface{}); ok && items != nil {
-			cleanedMap["items"] = cleanFunctionParameters(items)
+			cleanedMap["items"] = cleanFunctionParametersWithDepth(items, depth+1)
 		}
-		// Also handle items if it's an array of schemas
-		if itemsArray, ok := cleanedMap["items"].([]interface{}); ok {
-			cleanedItemsArray := make([]interface{}, len(itemsArray))
-			for i, item := range itemsArray {
-				cleanedItemsArray[i] = cleanFunctionParameters(item)
-			}
-			cleanedMap["items"] = cleanedItemsArray
+		// OpenAPI tuple-style items is not supported by Gemini SDK Schema; keep first to avoid API rejection.
+		if itemsArray, ok := cleanedMap["items"].([]interface{}); ok && len(itemsArray) > 0 {
+			cleanedMap["items"] = cleanFunctionParametersWithDepth(itemsArray[0], depth+1)
 		}
 
-		// Recursively clean other schema composition keywords
-		for _, field := range []string{"allOf", "anyOf", "oneOf"} {
-			if nested, ok := cleanedMap[field].([]interface{}); ok {
-				cleanedNested := make([]interface{}, len(nested))
-				for i, item := range nested {
-					cleanedNested[i] = cleanFunctionParameters(item)
-				}
-				cleanedMap[field] = cleanedNested
-			}
-		}
-
-		// Recursively clean patternProperties
-		if patternProps, ok := cleanedMap["patternProperties"].(map[string]interface{}); ok {
-			cleanedPatternProps := make(map[string]interface{})
-			for pattern, schema := range patternProps {
-				cleanedPatternProps[pattern] = cleanFunctionParameters(schema)
-			}
-			cleanedMap["patternProperties"] = cleanedPatternProps
-		}
-
-		// Recursively clean definitions
-		if definitions, ok := cleanedMap["definitions"].(map[string]interface{}); ok {
-			cleanedDefinitions := make(map[string]interface{})
-			for defName, defSchema := range definitions {
-				cleanedDefinitions[defName] = cleanFunctionParameters(defSchema)
-			}
-			cleanedMap["definitions"] = cleanedDefinitions
-		}
-
-		// Recursively clean $defs (newer JSON Schema draft)
-		if defs, ok := cleanedMap["$defs"].(map[string]interface{}); ok {
-			cleanedDefs := make(map[string]interface{})
-			for defName, defSchema := range defs {
-				cleanedDefs[defName] = cleanFunctionParameters(defSchema)
-			}
-			cleanedMap["$defs"] = cleanedDefs
-		}
-
-		// Clean conditional keywords
-		for _, field := range []string{"if", "then", "else", "not"} {
-			if nested, ok := cleanedMap[field]; ok {
-				cleanedMap[field] = cleanFunctionParameters(nested)
+		// Recursively clean anyOf
+		if nested, ok := cleanedMap["anyOf"].([]interface{}); ok && nested != nil {
+			cleanedNested := make([]interface{}, len(nested))
+			for i, item := range nested {
+				cleanedNested[i] = cleanFunctionParametersWithDepth(item, depth+1)
 			}
+			cleanedMap["anyOf"] = cleanedNested
 		}
 
 		return cleanedMap
@@ -762,7 +748,7 @@ func cleanFunctionParameters(params interface{}) interface{} {
 		// Handle arrays of schemas
 		cleanedArray := make([]interface{}, len(v))
 		for i, item := range v {
-			cleanedArray[i] = cleanFunctionParameters(item)
+			cleanedArray[i] = cleanFunctionParametersWithDepth(item, depth+1)
 		}
 		return cleanedArray
 
@@ -772,6 +758,91 @@ func cleanFunctionParameters(params interface{}) interface{} {
 	}
 }
 
+func cleanFunctionParametersShallow(params interface{}) interface{} {
+	switch v := params.(type) {
+	case map[string]interface{}:
+		cleanedMap := make(map[string]interface{}, len(v))
+		for k, val := range v {
+			if _, ok := geminiOpenAPISchemaAllowedFields[k]; ok {
+				cleanedMap[k] = val
+			}
+		}
+		normalizeGeminiSchemaTypeAndNullable(cleanedMap)
+		// Stop recursion and avoid retaining huge nested structures.
+		delete(cleanedMap, "properties")
+		delete(cleanedMap, "items")
+		delete(cleanedMap, "anyOf")
+		return cleanedMap
+	case []interface{}:
+		// Prefer an empty list over deep recursion on attacker-controlled inputs.
+		return []interface{}{}
+	default:
+		return params
+	}
+}
+
+func normalizeGeminiSchemaTypeAndNullable(schema map[string]interface{}) {
+	rawType, ok := schema["type"]
+	if !ok || rawType == nil {
+		return
+	}
+
+	normalize := func(t string) (string, bool) {
+		switch strings.ToLower(strings.TrimSpace(t)) {
+		case "object":
+			return "OBJECT", false
+		case "array":
+			return "ARRAY", false
+		case "string":
+			return "STRING", false
+		case "integer":
+			return "INTEGER", false
+		case "number":
+			return "NUMBER", false
+		case "boolean":
+			return "BOOLEAN", false
+		case "null":
+			return "", true
+		default:
+			return t, false
+		}
+	}
+
+	switch t := rawType.(type) {
+	case string:
+		normalized, isNull := normalize(t)
+		if isNull {
+			schema["nullable"] = true
+			delete(schema, "type")
+			return
+		}
+		schema["type"] = normalized
+	case []interface{}:
+		nullable := false
+		var chosen string
+		for _, item := range t {
+			if s, ok := item.(string); ok {
+				normalized, isNull := normalize(s)
+				if isNull {
+					nullable = true
+					continue
+				}
+				if chosen == "" {
+					chosen = normalized
+				}
+			}
+		}
+		if nullable {
+			schema["nullable"] = true
+		}
+		if chosen != "" {
+			schema["type"] = chosen
+		} else {
+			delete(schema, "type")
+		}
+	}
+}
+
 func removeAdditionalPropertiesWithDepth(schema interface{}, depth int) interface{} {
 	if depth >= 5 {
 		return schema
@@ -1183,6 +1254,8 @@ func GeminiChatStreamHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *
 	id := helper.GetResponseID(c)
 	createAt := common.GetTimestamp()
 	finishReason := constant.FinishReasonStop
+	toolCallIndexByChoice := make(map[int]map[string]int)
+	nextToolCallIndexByChoice := make(map[int]int)
 
 	usage, err := geminiStreamHandler(c, info, resp, func(data string, geminiResponse *dto.GeminiChatResponse) bool {
 		response, isStop := streamResponseGeminiChat2OpenAI(geminiResponse)
@@ -1190,6 +1263,28 @@ func GeminiChatStreamHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *
 		response.Id = id
 		response.Created = createAt
 		response.Model = info.UpstreamModelName
+		for choiceIdx := range response.Choices {
+			choiceKey := response.Choices[choiceIdx].Index
+			for toolIdx := range response.Choices[choiceIdx].Delta.ToolCalls {
+				tool := &response.Choices[choiceIdx].Delta.ToolCalls[toolIdx]
+				if tool.ID == "" {
+					continue
+				}
+				m := toolCallIndexByChoice[choiceKey]
+				if m == nil {
+					m = make(map[string]int)
+					toolCallIndexByChoice[choiceKey] = m
+				}
+				if idx, ok := m[tool.ID]; ok {
+					tool.SetIndex(idx)
+					continue
+				}
+				idx := nextToolCallIndexByChoice[choiceKey]
+				nextToolCallIndexByChoice[choiceKey] = idx + 1
+				m[tool.ID] = idx
+				tool.SetIndex(idx)
+			}
+		}
 
 		logger.LogDebug(c, fmt.Sprintf("info.SendResponseCount = %d", info.SendResponseCount))
 		if info.SendResponseCount == 0 {
@@ -1417,6 +1512,79 @@ func GeminiImageHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http.
 	return usage, nil
 }
 
+type GeminiModelsResponse struct {
+	Models        []dto.GeminiModel `json:"models"`
+	NextPageToken string            `json:"nextPageToken"`
+}
+
+func FetchGeminiModels(baseURL, apiKey, proxyURL string) ([]string, error) {
+	client, err := service.GetHttpClientWithProxy(proxyURL)
+	if err != nil {
+		return nil, fmt.Errorf("创建HTTP客户端失败: %v", err)
+	}
+
+	allModels := make([]string, 0)
+	nextPageToken := ""
+	maxPages := 100 // Safety limit to prevent infinite loops
+
+	for page := 0; page < maxPages; page++ {
+		url := fmt.Sprintf("%s/v1beta/models", baseURL)
+		if nextPageToken != "" {
+			url = fmt.Sprintf("%s?pageToken=%s", url, nextPageToken)
+		}
+
+		ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
+		request, err := http.NewRequestWithContext(ctx, "GET", url, nil)
+		if err != nil {
+			cancel()
+			return nil, fmt.Errorf("创建请求失败: %v", err)
+		}
+
+		request.Header.Set("x-goog-api-key", apiKey)
+
+		response, err := client.Do(request)
+		if err != nil {
+			cancel()
+			return nil, fmt.Errorf("请求失败: %v", err)
+		}
+
+		if response.StatusCode != http.StatusOK {
+			body, _ := io.ReadAll(response.Body)
+			response.Body.Close()
+			cancel()
+			return nil, fmt.Errorf("服务器返回错误 %d: %s", response.StatusCode, string(body))
+		}
+
+		body, err := io.ReadAll(response.Body)
+		response.Body.Close()
+		cancel()
+		if err != nil {
+			return nil, fmt.Errorf("读取响应失败: %v", err)
+		}
+
+		var modelsResponse GeminiModelsResponse
+		if err = common.Unmarshal(body, &modelsResponse); err != nil {
+			return nil, fmt.Errorf("解析响应失败: %v", err)
+		}
+
+		for _, model := range modelsResponse.Models {
+			modelNameValue, ok := model.Name.(string)
+			if !ok {
+				continue
+			}
+			modelName := strings.TrimPrefix(modelNameValue, "models/")
+			allModels = append(allModels, modelName)
+		}
+
+		nextPageToken = modelsResponse.NextPageToken
+		if nextPageToken == "" {
+			break
+		}
+	}
+
+	return allModels, nil
+}
+
 // convertToolChoiceToGeminiConfig converts OpenAI tool_choice to Gemini toolConfig
 // OpenAI tool_choice values:
 //   - "auto": Let the model decide (default)

relay/channel/minimax/constants.go

@@ -14,6 +14,9 @@ var ModelList = []string{
 	"speech-02-turbo",
 	"speech-01-hd",
 	"speech-01-turbo",
+	"MiniMax-M2.1",
+	"MiniMax-M2.1-lightning",
+	"MiniMax-M2",
 }
 
 var ChannelName = "minimax"

relay/channel/ollama/dto.go

@@ -67,3 +67,40 @@ type OllamaEmbeddingResponse struct {
 	Embeddings      [][]float64 `json:"embeddings"`
 	PromptEvalCount int         `json:"prompt_eval_count,omitempty"`
 }
+
+type OllamaTagsResponse struct {
+	Models []OllamaModel `json:"models"`
+}
+
+type OllamaModel struct {
+	Name       string            `json:"name"`
+	Size       int64             `json:"size"`
+	Digest     string            `json:"digest,omitempty"`
+	ModifiedAt string            `json:"modified_at"`
+	Details    OllamaModelDetail `json:"details,omitempty"`
+}
+
+type OllamaModelDetail struct {
+	ParentModel       string   `json:"parent_model,omitempty"`
+	Format            string   `json:"format,omitempty"`
+	Family            string   `json:"family,omitempty"`
+	Families          []string `json:"families,omitempty"`
+	ParameterSize     string   `json:"parameter_size,omitempty"`
+	QuantizationLevel string   `json:"quantization_level,omitempty"`
+}
+
+type OllamaPullRequest struct {
+	Name   string `json:"name"`
+	Stream bool   `json:"stream,omitempty"`
+}
+
+type OllamaPullResponse struct {
+	Status    string `json:"status"`
+	Digest    string `json:"digest,omitempty"`
+	Total     int64  `json:"total,omitempty"`
+	Completed int64  `json:"completed,omitempty"`
+}
+
+type OllamaDeleteRequest struct {
+	Name string `json:"name"`
+}

relay/channel/ollama/relay-ollama.go

@@ -1,11 +1,13 @@
 package ollama
 
 import (
+	"bufio"
 	"encoding/json"
 	"fmt"
 	"io"
 	"net/http"
 	"strings"
+	"time"
 
 	"github.com/QuantumNous/new-api/common"
 	"github.com/QuantumNous/new-api/dto"
@@ -283,3 +285,246 @@ func ollamaEmbeddingHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *h
 	service.IOCopyBytesGracefully(c, resp, out)
 	return usage, nil
 }
+
+func FetchOllamaModels(baseURL, apiKey string) ([]OllamaModel, error) {
+	url := fmt.Sprintf("%s/api/tags", baseURL)
+
+	client := &http.Client{}
+	request, err := http.NewRequest("GET", url, nil)
+	if err != nil {
+		return nil, fmt.Errorf("创建请求失败: %v", err)
+	}
+
+	// Ollama 通常不需要 Bearer token，但为了兼容性保留
+	if apiKey != "" {
+		request.Header.Set("Authorization", "Bearer "+apiKey)
+	}
+
+	response, err := client.Do(request)
+	if err != nil {
+		return nil, fmt.Errorf("请求失败: %v", err)
+	}
+	defer response.Body.Close()
+
+	if response.StatusCode != http.StatusOK {
+		body, _ := io.ReadAll(response.Body)
+		return nil, fmt.Errorf("服务器返回错误 %d: %s", response.StatusCode, string(body))
+	}
+
+	var tagsResponse OllamaTagsResponse
+	body, err := io.ReadAll(response.Body)
+	if err != nil {
+		return nil, fmt.Errorf("读取响应失败: %v", err)
+	}
+
+	err = common.Unmarshal(body, &tagsResponse)
+	if err != nil {
+		return nil, fmt.Errorf("解析响应失败: %v", err)
+	}
+
+	return tagsResponse.Models, nil
+}
+
+// 拉取 Ollama 模型 (非流式)
+func PullOllamaModel(baseURL, apiKey, modelName string) error {
+	url := fmt.Sprintf("%s/api/pull", baseURL)
+
+	pullRequest := OllamaPullRequest{
+		Name:   modelName,
+		Stream: false, // 非流式，简化处理
+	}
+
+	requestBody, err := common.Marshal(pullRequest)
+	if err != nil {
+		return fmt.Errorf("序列化请求失败: %v", err)
+	}
+
+	client := &http.Client{
+		Timeout: 30 * 60 * 1000 * time.Millisecond, // 30分钟超时，支持大模型
+	}
+	request, err := http.NewRequest("POST", url, strings.NewReader(string(requestBody)))
+	if err != nil {
+		return fmt.Errorf("创建请求失败: %v", err)
+	}
+
+	request.Header.Set("Content-Type", "application/json")
+	if apiKey != "" {
+		request.Header.Set("Authorization", "Bearer "+apiKey)
+	}
+
+	response, err := client.Do(request)
+	if err != nil {
+		return fmt.Errorf("请求失败: %v", err)
+	}
+	defer response.Body.Close()
+
+	if response.StatusCode != http.StatusOK {
+		body, _ := io.ReadAll(response.Body)
+		return fmt.Errorf("拉取模型失败 %d: %s", response.StatusCode, string(body))
+	}
+
+	return nil
+}
+
+// 流式拉取 Ollama 模型 (支持进度回调)
+func PullOllamaModelStream(baseURL, apiKey, modelName string, progressCallback func(OllamaPullResponse)) error {
+	url := fmt.Sprintf("%s/api/pull", baseURL)
+
+	pullRequest := OllamaPullRequest{
+		Name:   modelName,
+		Stream: true, // 启用流式
+	}
+
+	requestBody, err := common.Marshal(pullRequest)
+	if err != nil {
+		return fmt.Errorf("序列化请求失败: %v", err)
+	}
+
+	client := &http.Client{
+		Timeout: 60 * 60 * 1000 * time.Millisecond, // 1小时超时，支持超大模型
+	}
+	request, err := http.NewRequest("POST", url, strings.NewReader(string(requestBody)))
+	if err != nil {
+		return fmt.Errorf("创建请求失败: %v", err)
+	}
+
+	request.Header.Set("Content-Type", "application/json")
+	if apiKey != "" {
+		request.Header.Set("Authorization", "Bearer "+apiKey)
+	}
+
+	response, err := client.Do(request)
+	if err != nil {
+		return fmt.Errorf("请求失败: %v", err)
+	}
+	defer response.Body.Close()
+
+	if response.StatusCode != http.StatusOK {
+		body, _ := io.ReadAll(response.Body)
+		return fmt.Errorf("拉取模型失败 %d: %s", response.StatusCode, string(body))
+	}
+
+	// 读取流式响应
+	scanner := bufio.NewScanner(response.Body)
+	successful := false
+	for scanner.Scan() {
+		line := scanner.Text()
+		if strings.TrimSpace(line) == "" {
+			continue
+		}
+
+		var pullResponse OllamaPullResponse
+		if err := common.Unmarshal([]byte(line), &pullResponse); err != nil {
+			continue // 忽略解析失败的行
+		}
+
+		if progressCallback != nil {
+			progressCallback(pullResponse)
+		}
+
+		// 检查是否出现错误或完成
+		if strings.EqualFold(pullResponse.Status, "error") {
+			return fmt.Errorf("拉取模型失败: %s", strings.TrimSpace(line))
+		}
+		if strings.EqualFold(pullResponse.Status, "success") {
+			successful = true
+			break
+		}
+	}
+
+	if err := scanner.Err(); err != nil {
+		return fmt.Errorf("读取流式响应失败: %v", err)
+	}
+
+	if !successful {
+		return fmt.Errorf("拉取模型未完成: 未收到成功状态")
+	}
+
+	return nil
+}
+
+// 删除 Ollama 模型
+func DeleteOllamaModel(baseURL, apiKey, modelName string) error {
+	url := fmt.Sprintf("%s/api/delete", baseURL)
+
+	deleteRequest := OllamaDeleteRequest{
+		Name: modelName,
+	}
+
+	requestBody, err := common.Marshal(deleteRequest)
+	if err != nil {
+		return fmt.Errorf("序列化请求失败: %v", err)
+	}
+
+	client := &http.Client{}
+	request, err := http.NewRequest("DELETE", url, strings.NewReader(string(requestBody)))
+	if err != nil {
+		return fmt.Errorf("创建请求失败: %v", err)
+	}
+
+	request.Header.Set("Content-Type", "application/json")
+	if apiKey != "" {
+		request.Header.Set("Authorization", "Bearer "+apiKey)
+	}
+
+	response, err := client.Do(request)
+	if err != nil {
+		return fmt.Errorf("请求失败: %v", err)
+	}
+	defer response.Body.Close()
+
+	if response.StatusCode != http.StatusOK {
+		body, _ := io.ReadAll(response.Body)
+		return fmt.Errorf("删除模型失败 %d: %s", response.StatusCode, string(body))
+	}
+
+	return nil
+}
+
+func FetchOllamaVersion(baseURL, apiKey string) (string, error) {
+	trimmedBase := strings.TrimRight(baseURL, "/")
+	if trimmedBase == "" {
+		return "", fmt.Errorf("baseURL 为空")
+	}
+
+	url := fmt.Sprintf("%s/api/version", trimmedBase)
+
+	client := &http.Client{Timeout: 10 * time.Second}
+	request, err := http.NewRequest("GET", url, nil)
+	if err != nil {
+		return "", fmt.Errorf("创建请求失败: %v", err)
+	}
+
+	if apiKey != "" {
+		request.Header.Set("Authorization", "Bearer "+apiKey)
+	}
+
+	response, err := client.Do(request)
+	if err != nil {
+		return "", fmt.Errorf("请求失败: %v", err)
+	}
+	defer response.Body.Close()
+
+	body, err := io.ReadAll(response.Body)
+	if err != nil {
+		return "", fmt.Errorf("读取响应失败: %v", err)
+	}
+
+	if response.StatusCode != http.StatusOK {
+		return "", fmt.Errorf("查询版本失败 %d: %s", response.StatusCode, string(body))
+	}
+
+	var versionResp struct {
+		Version string `json:"version"`
+	}
+
+	if err := json.Unmarshal(body, &versionResp); err != nil {
+		return "", fmt.Errorf("解析响应失败: %v", err)
+	}
+
+	if versionResp.Version == "" {
+		return "", fmt.Errorf("未返回版本信息")
+	}
+
+	return versionResp.Version, nil
+}

relay/channel/openai/chat_via_responses.go

@@ -0,0 +1,369 @@
+package openai
+
+import (
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+	"time"
+
+	"github.com/QuantumNous/new-api/common"
+	"github.com/QuantumNous/new-api/dto"
+	"github.com/QuantumNous/new-api/logger"
+	relaycommon "github.com/QuantumNous/new-api/relay/common"
+	"github.com/QuantumNous/new-api/relay/helper"
+	"github.com/QuantumNous/new-api/service"
+	"github.com/QuantumNous/new-api/types"
+
+	"github.com/gin-gonic/gin"
+)
+
+func OaiResponsesToChatHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http.Response) (*dto.Usage, *types.NewAPIError) {
+	if resp == nil || resp.Body == nil {
+		return nil, types.NewOpenAIError(fmt.Errorf("invalid response"), types.ErrorCodeBadResponse, http.StatusInternalServerError)
+	}
+
+	defer service.CloseResponseBodyGracefully(resp)
+
+	var responsesResp dto.OpenAIResponsesResponse
+	body, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, types.NewOpenAIError(err, types.ErrorCodeReadResponseBodyFailed, http.StatusInternalServerError)
+	}
+
+	if err := common.Unmarshal(body, &responsesResp); err != nil {
+		return nil, types.NewOpenAIError(err, types.ErrorCodeBadResponseBody, http.StatusInternalServerError)
+	}
+
+	if oaiError := responsesResp.GetOpenAIError(); oaiError != nil && oaiError.Type != "" {
+		return nil, types.WithOpenAIError(*oaiError, resp.StatusCode)
+	}
+
+	chatId := helper.GetResponseID(c)
+	chatResp, usage, err := service.ResponsesResponseToChatCompletionsResponse(&responsesResp, chatId)
+	if err != nil {
+		return nil, types.NewOpenAIError(err, types.ErrorCodeBadResponseBody, http.StatusInternalServerError)
+	}
+
+	if usage == nil || usage.TotalTokens == 0 {
+		text := service.ExtractOutputTextFromResponses(&responsesResp)
+		usage = service.ResponseText2Usage(c, text, info.UpstreamModelName, info.GetEstimatePromptTokens())
+		chatResp.Usage = *usage
+	}
+
+	chatBody, err := common.Marshal(chatResp)
+	if err != nil {
+		return nil, types.NewOpenAIError(err, types.ErrorCodeJsonMarshalFailed, http.StatusInternalServerError)
+	}
+
+	service.IOCopyBytesGracefully(c, resp, chatBody)
+	return usage, nil
+}
+
+func OaiResponsesToChatStreamHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http.Response) (*dto.Usage, *types.NewAPIError) {
+	if resp == nil || resp.Body == nil {
+		return nil, types.NewOpenAIError(fmt.Errorf("invalid response"), types.ErrorCodeBadResponse, http.StatusInternalServerError)
+	}
+
+	defer service.CloseResponseBodyGracefully(resp)
+
+	responseId := helper.GetResponseID(c)
+	createAt := time.Now().Unix()
+	model := info.UpstreamModelName
+
+	var (
+		usage       = &dto.Usage{}
+		outputText  strings.Builder
+		usageText   strings.Builder
+		sentStart   bool
+		sentStop    bool
+		sawToolCall bool
+		streamErr   *types.NewAPIError
+	)
+
+	toolCallIndexByID := make(map[string]int)
+	toolCallNameByID := make(map[string]string)
+	toolCallArgsByID := make(map[string]string)
+	toolCallNameSent := make(map[string]bool)
+	toolCallCanonicalIDByItemID := make(map[string]string)
+
+	sendStartIfNeeded := func() bool {
+		if sentStart {
+			return true
+		}
+		if err := helper.ObjectData(c, helper.GenerateStartEmptyResponse(responseId, createAt, model, nil)); err != nil {
+			streamErr = types.NewOpenAIError(err, types.ErrorCodeBadResponse, http.StatusInternalServerError)
+			return false
+		}
+		sentStart = true
+		return true
+	}
+
+	sendToolCallDelta := func(callID string, name string, argsDelta string) bool {
+		if callID == "" {
+			return true
+		}
+		if outputText.Len() > 0 {
+			// Prefer streaming assistant text over tool calls to match non-stream behavior.
+			return true
+		}
+		if !sendStartIfNeeded() {
+			return false
+		}
+
+		idx, ok := toolCallIndexByID[callID]
+		if !ok {
+			idx = len(toolCallIndexByID)
+			toolCallIndexByID[callID] = idx
+		}
+		if name != "" {
+			toolCallNameByID[callID] = name
+		}
+		if toolCallNameByID[callID] != "" {
+			name = toolCallNameByID[callID]
+		}
+
+		tool := dto.ToolCallResponse{
+			ID:   callID,
+			Type: "function",
+			Function: dto.FunctionResponse{
+				Arguments: argsDelta,
+			},
+		}
+		tool.SetIndex(idx)
+		if name != "" && !toolCallNameSent[callID] {
+			tool.Function.Name = name
+			toolCallNameSent[callID] = true
+		}
+
+		chunk := &dto.ChatCompletionsStreamResponse{
+			Id:      responseId,
+			Object:  "chat.completion.chunk",
+			Created: createAt,
+			Model:   model,
+			Choices: []dto.ChatCompletionsStreamResponseChoice{
+				{
+					Index: 0,
+					Delta: dto.ChatCompletionsStreamResponseChoiceDelta{
+						ToolCalls: []dto.ToolCallResponse{tool},
+					},
+				},
+			},
+		}
+		if err := helper.ObjectData(c, chunk); err != nil {
+			streamErr = types.NewOpenAIError(err, types.ErrorCodeBadResponse, http.StatusInternalServerError)
+			return false
+		}
+		sawToolCall = true
+
+		// Include tool call data in the local builder for fallback token estimation.
+		if tool.Function.Name != "" {
+			usageText.WriteString(tool.Function.Name)
+		}
+		if argsDelta != "" {
+			usageText.WriteString(argsDelta)
+		}
+		return true
+	}
+
+	helper.StreamScannerHandler(c, resp, info, func(data string) bool {
+		if streamErr != nil {
+			return false
+		}
+
+		var streamResp dto.ResponsesStreamResponse
+		if err := common.UnmarshalJsonStr(data, &streamResp); err != nil {
+			logger.LogError(c, "failed to unmarshal responses stream event: "+err.Error())
+			return true
+		}
+
+		switch streamResp.Type {
+		case "response.created":
+			if streamResp.Response != nil {
+				if streamResp.Response.Model != "" {
+					model = streamResp.Response.Model
+				}
+				if streamResp.Response.CreatedAt != 0 {
+					createAt = int64(streamResp.Response.CreatedAt)
+				}
+			}
+
+		case "response.output_text.delta":
+			if !sendStartIfNeeded() {
+				return false
+			}
+
+			if streamResp.Delta != "" {
+				outputText.WriteString(streamResp.Delta)
+				usageText.WriteString(streamResp.Delta)
+				delta := streamResp.Delta
+				chunk := &dto.ChatCompletionsStreamResponse{
+					Id:      responseId,
+					Object:  "chat.completion.chunk",
+					Created: createAt,
+					Model:   model,
+					Choices: []dto.ChatCompletionsStreamResponseChoice{
+						{
+							Index: 0,
+							Delta: dto.ChatCompletionsStreamResponseChoiceDelta{
+								Content: &delta,
+							},
+						},
+					},
+				}
+				if err := helper.ObjectData(c, chunk); err != nil {
+					streamErr = types.NewOpenAIError(err, types.ErrorCodeBadResponse, http.StatusInternalServerError)
+					return false
+				}
+			}
+
+		case "response.output_item.added", "response.output_item.done":
+			if streamResp.Item == nil {
+				break
+			}
+			if streamResp.Item.Type != "function_call" {
+				break
+			}
+
+			itemID := strings.TrimSpace(streamResp.Item.ID)
+			callID := strings.TrimSpace(streamResp.Item.CallId)
+			if callID == "" {
+				callID = itemID
+			}
+			if itemID != "" && callID != "" {
+				toolCallCanonicalIDByItemID[itemID] = callID
+			}
+			name := strings.TrimSpace(streamResp.Item.Name)
+			if name != "" {
+				toolCallNameByID[callID] = name
+			}
+
+			newArgs := streamResp.Item.Arguments
+			prevArgs := toolCallArgsByID[callID]
+			argsDelta := ""
+			if newArgs != "" {
+				if strings.HasPrefix(newArgs, prevArgs) {
+					argsDelta = newArgs[len(prevArgs):]
+				} else {
+					argsDelta = newArgs
+				}
+				toolCallArgsByID[callID] = newArgs
+			}
+
+			if !sendToolCallDelta(callID, name, argsDelta) {
+				return false
+			}
+
+		case "response.function_call_arguments.delta":
+			itemID := strings.TrimSpace(streamResp.ItemID)
+			callID := toolCallCanonicalIDByItemID[itemID]
+			if callID == "" {
+				callID = itemID
+			}
+			if callID == "" {
+				break
+			}
+			toolCallArgsByID[callID] += streamResp.Delta
+			if !sendToolCallDelta(callID, "", streamResp.Delta) {
+				return false
+			}
+
+		case "response.function_call_arguments.done":
+
+		case "response.completed":
+			if streamResp.Response != nil {
+				if streamResp.Response.Model != "" {
+					model = streamResp.Response.Model
+				}
+				if streamResp.Response.CreatedAt != 0 {
+					createAt = int64(streamResp.Response.CreatedAt)
+				}
+				if streamResp.Response.Usage != nil {
+					if streamResp.Response.Usage.InputTokens != 0 {
+						usage.PromptTokens = streamResp.Response.Usage.InputTokens
+						usage.InputTokens = streamResp.Response.Usage.InputTokens
+					}
+					if streamResp.Response.Usage.OutputTokens != 0 {
+						usage.CompletionTokens = streamResp.Response.Usage.OutputTokens
+						usage.OutputTokens = streamResp.Response.Usage.OutputTokens
+					}
+					if streamResp.Response.Usage.TotalTokens != 0 {
+						usage.TotalTokens = streamResp.Response.Usage.TotalTokens
+					} else {
+						usage.TotalTokens = usage.PromptTokens + usage.CompletionTokens
+					}
+					if streamResp.Response.Usage.InputTokensDetails != nil {
+						usage.PromptTokensDetails.CachedTokens = streamResp.Response.Usage.InputTokensDetails.CachedTokens
+						usage.PromptTokensDetails.ImageTokens = streamResp.Response.Usage.InputTokensDetails.ImageTokens
+						usage.PromptTokensDetails.AudioTokens = streamResp.Response.Usage.InputTokensDetails.AudioTokens
+					}
+					if streamResp.Response.Usage.CompletionTokenDetails.ReasoningTokens != 0 {
+						usage.CompletionTokenDetails.ReasoningTokens = streamResp.Response.Usage.CompletionTokenDetails.ReasoningTokens
+					}
+				}
+			}
+
+			if !sendStartIfNeeded() {
+				return false
+			}
+			if !sentStop {
+				finishReason := "stop"
+				if sawToolCall && outputText.Len() == 0 {
+					finishReason = "tool_calls"
+				}
+				stop := helper.GenerateStopResponse(responseId, createAt, model, finishReason)
+				if err := helper.ObjectData(c, stop); err != nil {
+					streamErr = types.NewOpenAIError(err, types.ErrorCodeBadResponse, http.StatusInternalServerError)
+					return false
+				}
+				sentStop = true
+			}
+
+		case "response.error", "response.failed":
+			if streamResp.Response != nil {
+				if oaiErr := streamResp.Response.GetOpenAIError(); oaiErr != nil && oaiErr.Type != "" {
+					streamErr = types.WithOpenAIError(*oaiErr, http.StatusInternalServerError)
+					return false
+				}
+			}
+			streamErr = types.NewOpenAIError(fmt.Errorf("responses stream error: %s", streamResp.Type), types.ErrorCodeBadResponse, http.StatusInternalServerError)
+			return false
+
+		default:
+		}
+
+		return true
+	})
+
+	if streamErr != nil {
+		return nil, streamErr
+	}
+
+	if usage.TotalTokens == 0 {
+		usage = service.ResponseText2Usage(c, usageText.String(), info.UpstreamModelName, info.GetEstimatePromptTokens())
+	}
+
+	if !sentStart {
+		if err := helper.ObjectData(c, helper.GenerateStartEmptyResponse(responseId, createAt, model, nil)); err != nil {
+			return nil, types.NewOpenAIError(err, types.ErrorCodeBadResponse, http.StatusInternalServerError)
+		}
+	}
+	if !sentStop {
+		finishReason := "stop"
+		if sawToolCall && outputText.Len() == 0 {
+			finishReason = "tool_calls"
+		}
+		stop := helper.GenerateStopResponse(responseId, createAt, model, finishReason)
+		if err := helper.ObjectData(c, stop); err != nil {
+			return nil, types.NewOpenAIError(err, types.ErrorCodeBadResponse, http.StatusInternalServerError)
+		}
+	}
+	if info.ShouldIncludeUsage && usage != nil {
+		if err := helper.ObjectData(c, helper.GenerateFinalUsageResponse(responseId, createAt, model, *usage)); err != nil {
+			return nil, types.NewOpenAIError(err, types.ErrorCodeBadResponse, http.StatusInternalServerError)
+		}
+	}
+
+	helper.Done(c)
+	return usage, nil
+}

relay/channel/openai/helper.go

@@ -208,7 +208,6 @@ func HandleFinalResponse(c *gin.Context, info *relaycommon.RelayInfo, lastStream
 		helper.Done(c)
 
 	case types.RelayFormatClaude:
-		info.ClaudeConvertInfo.Done = true
 		var streamResponse dto.ChatCompletionsStreamResponse
 		if err := common.Unmarshal(common.StringToByteSlice(lastStreamData), &streamResponse); err != nil {
 			common.SysLog("error unmarshalling stream response: " + err.Error())
@@ -221,6 +220,7 @@ func HandleFinalResponse(c *gin.Context, info *relaycommon.RelayInfo, lastStream
 		for _, resp := range claudeResponses {
 			_ = helper.ClaudeData(c, *resp)
 		}
+		info.ClaudeConvertInfo.Done = true
 
 	case types.RelayFormatGemini:
 		var streamResponse dto.ChatCompletionsStreamResponse

relay/channel/openai/relay-openai.go

@@ -186,7 +186,7 @@ func OaiStreamHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http.Re
 		usage.CompletionTokens += toolCount * 7
 	}
 
-	applyUsagePostProcessing(info, usage, nil)
+	applyUsagePostProcessing(info, usage, common.StringToByteSlice(lastStreamData))
 
 	HandleFinalResponse(c, info, lastStreamData, responseId, createAt, model, systemFingerprint, usage, containStreamUsage)
 
@@ -596,7 +596,8 @@ func applyUsagePostProcessing(info *relaycommon.RelayInfo, usage *dto.Usage, res
 		if usage.PromptTokensDetails.CachedTokens == 0 && usage.PromptCacheHitTokens != 0 {
 			usage.PromptTokensDetails.CachedTokens = usage.PromptCacheHitTokens
 		}
-	case constant.ChannelTypeZhipu_v4, constant.ChannelTypeMoonshot:
+	case constant.ChannelTypeZhipu_v4:
+		// 智普的cached_tokens在标准位置: usage.prompt_tokens_details.cached_tokens
 		if usage.PromptTokensDetails.CachedTokens == 0 {
 			if usage.InputTokensDetails != nil && usage.InputTokensDetails.CachedTokens > 0 {
 				usage.PromptTokensDetails.CachedTokens = usage.InputTokensDetails.CachedTokens
@@ -606,6 +607,19 @@ func applyUsagePostProcessing(info *relaycommon.RelayInfo, usage *dto.Usage, res
 				usage.PromptTokensDetails.CachedTokens = usage.PromptCacheHitTokens
 			}
 		}
+	case constant.ChannelTypeMoonshot:
+		// Moonshot的cached_tokens在非标准位置: choices[].usage.cached_tokens
+		if usage.PromptTokensDetails.CachedTokens == 0 {
+			if usage.InputTokensDetails != nil && usage.InputTokensDetails.CachedTokens > 0 {
+				usage.PromptTokensDetails.CachedTokens = usage.InputTokensDetails.CachedTokens
+			} else if cachedTokens, ok := extractMoonshotCachedTokensFromBody(responseBody); ok {
+				usage.PromptTokensDetails.CachedTokens = cachedTokens
+			} else if cachedTokens, ok := extractCachedTokensFromBody(responseBody); ok {
+				usage.PromptTokensDetails.CachedTokens = cachedTokens
+			} else if usage.PromptCacheHitTokens > 0 {
+				usage.PromptTokensDetails.CachedTokens = usage.PromptCacheHitTokens
+			}
+		}
 	}
 }
 
@@ -639,3 +653,32 @@ func extractCachedTokensFromBody(body []byte) (int, bool) {
 	}
 	return 0, false
 }
+
+// extractMoonshotCachedTokensFromBody 从Moonshot的非标准位置提取cached_tokens
+// Moonshot的流式响应格式: {"choices":[{"usage":{"cached_tokens":111}}]}
+func extractMoonshotCachedTokensFromBody(body []byte) (int, bool) {
+	if len(body) == 0 {
+		return 0, false
+	}
+
+	var payload struct {
+		Choices []struct {
+			Usage struct {
+				CachedTokens *int `json:"cached_tokens"`
+			} `json:"usage"`
+		} `json:"choices"`
+	}
+
+	if err := common.Unmarshal(body, &payload); err != nil {
+		return 0, false
+	}
+
+	// 遍历choices查找cached_tokens
+	for _, choice := range payload.Choices {
+		if choice.Usage.CachedTokens != nil && *choice.Usage.CachedTokens > 0 {
+			return *choice.Usage.CachedTokens, true
+		}
+	}
+
+	return 0, false
+}

relay/channel/task/ali/adaptor.go

@@ -192,6 +192,10 @@ func sizeToResolution(size string) (string, error) {
 func ProcessAliOtherRatios(aliReq *AliVideoRequest) (map[string]float64, error) {
 	otherRatios := make(map[string]float64)
 	aliRatios := map[string]map[string]float64{
+		"wan2.6-i2v": {
+			"720P":  1,
+			"1080P": 1 / 0.6,
+		},
 		"wan2.5-t2v-preview": {
 			"480P":  1,
 			"720P":  2,
@@ -287,7 +291,9 @@ func (a *TaskAdaptor) convertToAliRequest(info *relaycommon.RelayInfo, req relay
 				aliReq.Parameters.Size = "1280*720"
 			}
 		} else {
-			if strings.HasPrefix(req.Model, "wan2.5") {
+			if strings.HasPrefix(req.Model, "wan2.6") {
+				aliReq.Parameters.Resolution = "1080P"
+			} else if strings.HasPrefix(req.Model, "wan2.5") {
 				aliReq.Parameters.Resolution = "1080P"
 			} else if strings.HasPrefix(req.Model, "wan2.2-i2v-flash") {
 				aliReq.Parameters.Resolution = "720P"

relay/channel/task/doubao/adaptor.go

@@ -6,6 +6,9 @@ import (
 	"fmt"
 	"io"
 	"net/http"
+	"time"
+
+	"github.com/QuantumNous/new-api/common"
 
 	"github.com/QuantumNous/new-api/constant"
 	"github.com/QuantumNous/new-api/dto"
@@ -23,18 +26,36 @@ import (
 // ============================
 
 type ContentItem struct {
-	Type     string    `json:"type"`                // "text" or "image_url"
-	Text     string    `json:"text,omitempty"`      // for text type
-	ImageURL *ImageURL `json:"image_url,omitempty"` // for image_url type
+	Type     string          `json:"type"`                // "text", "image_url" or "video"
+	Text     string          `json:"text,omitempty"`      // for text type
+	ImageURL *ImageURL       `json:"image_url,omitempty"` // for image_url type
+	Video    *VideoReference `json:"video,omitempty"`     // for video (sample) type
 }
 
 type ImageURL struct {
 	URL string `json:"url"`
 }
 
+type VideoReference struct {
+	URL string `json:"url"` // Draft video URL
+}
+
 type requestPayload struct {
-	Model   string        `json:"model"`
-	Content []ContentItem `json:"content"`
+	Model                 string         `json:"model"`
+	Content               []ContentItem  `json:"content"`
+	CallbackURL           string         `json:"callback_url,omitempty"`
+	ReturnLastFrame       *dto.BoolValue `json:"return_last_frame,omitempty"`
+	ServiceTier           string         `json:"service_tier,omitempty"`
+	ExecutionExpiresAfter dto.IntValue   `json:"execution_expires_after,omitempty"`
+	GenerateAudio         *dto.BoolValue `json:"generate_audio,omitempty"`
+	Draft                 *dto.BoolValue `json:"draft,omitempty"`
+	Resolution            string         `json:"resolution,omitempty"`
+	Ratio                 string         `json:"ratio,omitempty"`
+	Duration              dto.IntValue   `json:"duration,omitempty"`
+	Frames                dto.IntValue   `json:"frames,omitempty"`
+	Seed                  dto.IntValue   `json:"seed,omitempty"`
+	CameraFixed           *dto.BoolValue `json:"camera_fixed,omitempty"`
+	Watermark             *dto.BoolValue `json:"watermark,omitempty"`
 }
 
 type responsePayload struct {
@@ -53,6 +74,7 @@ type responseTask struct {
 	Duration        int    `json:"duration"`
 	Ratio           string `json:"ratio"`
 	FramesPerSecond int    `json:"framespersecond"`
+	ServiceTier     string `json:"service_tier"`
 	Usage           struct {
 		CompletionTokens int `json:"completion_tokens"`
 		TotalTokens      int `json:"total_tokens"`
@@ -98,16 +120,16 @@ func (a *TaskAdaptor) BuildRequestHeader(c *gin.Context, req *http.Request, info
 
 // BuildRequestBody converts request into Doubao specific format.
 func (a *TaskAdaptor) BuildRequestBody(c *gin.Context, info *relaycommon.RelayInfo) (io.Reader, error) {
-	v, exists := c.Get("task_request")
-	if !exists {
-		return nil, fmt.Errorf("request not found in context")
+	req, err := relaycommon.GetTaskRequest(c)
+	if err != nil {
+		return nil, err
 	}
-	req := v.(relaycommon.TaskSubmitReq)
 
 	body, err := a.convertToRequestPayload(&req)
 	if err != nil {
 		return nil, errors.Wrap(err, "convert request payload failed")
 	}
+	info.UpstreamModelName = body.Model
 	data, err := json.Marshal(body)
 	if err != nil {
 		return nil, err
@@ -141,7 +163,13 @@ func (a *TaskAdaptor) DoResponse(c *gin.Context, resp *http.Response, info *rela
 		return
 	}
 
-	c.JSON(http.StatusOK, gin.H{"task_id": dResp.ID})
+	ov := dto.NewOpenAIVideo()
+	ov.ID = dResp.ID
+	ov.TaskID = dResp.ID
+	ov.CreatedAt = time.Now().Unix()
+	ov.Model = info.OriginModelName
+
+	c.JSON(http.StatusOK, ov)
 	return dResp.ID, responseBody, nil
 }
 
@@ -204,12 +232,15 @@ func (a *TaskAdaptor) convertToRequestPayload(req *relaycommon.TaskSubmitReq) (*
 		}
 	}
 
-	// TODO: Add support for additional parameters from metadata
-	// such as ratio, duration, seed, etc.
-	// metadata := req.Metadata
-	// if metadata != nil {
-	//     // Parse and apply metadata parameters
-	// }
+	metadata := req.Metadata
+	medaBytes, err := json.Marshal(metadata)
+	if err != nil {
+		return nil, errors.Wrap(err, "metadata marshal metadata failed")
+	}
+	err = json.Unmarshal(medaBytes, &r)
+	if err != nil {
+		return nil, errors.Wrap(err, "unmarshal metadata failed")
+	}
 
 	return &r, nil
 }
@@ -229,7 +260,7 @@ func (a *TaskAdaptor) ParseTaskResult(respBody []byte) (*relaycommon.TaskInfo, e
 	case "pending", "queued":
 		taskResult.Status = model.TaskStatusQueued
 		taskResult.Progress = "10%"
-	case "processing":
+	case "processing", "running":
 		taskResult.Status = model.TaskStatusInProgress
 		taskResult.Progress = "50%"
 	case "succeeded":
@@ -251,3 +282,30 @@ func (a *TaskAdaptor) ParseTaskResult(respBody []byte) (*relaycommon.TaskInfo, e
 
 	return &taskResult, nil
 }
+
+func (a *TaskAdaptor) ConvertToOpenAIVideo(originTask *model.Task) ([]byte, error) {
+	var dResp responseTask
+	if err := json.Unmarshal(originTask.Data, &dResp); err != nil {
+		return nil, errors.Wrap(err, "unmarshal doubao task data failed")
+	}
+
+	openAIVideo := dto.NewOpenAIVideo()
+	openAIVideo.ID = originTask.TaskID
+	openAIVideo.TaskID = originTask.TaskID
+	openAIVideo.Status = originTask.Status.ToVideoStatus()
+	openAIVideo.SetProgressStr(originTask.Progress)
+	openAIVideo.SetMetadata("url", dResp.Content.VideoURL)
+	openAIVideo.CreatedAt = originTask.CreatedAt
+	openAIVideo.CompletedAt = originTask.UpdatedAt
+	openAIVideo.Model = originTask.Properties.OriginModelName
+
+	if dResp.Status == "failed" {
+		openAIVideo.Error = &dto.OpenAIVideoError{
+			Message: "task failed",
+			Code:    "failed",
+		}
+	}
+
+	jsonData, _ := common.Marshal(openAIVideo)
+	return jsonData, nil
+}

relay/channel/task/doubao/constants.go

@@ -4,6 +4,7 @@ var ModelList = []string{
 	"doubao-seedance-1-0-pro-250528",
 	"doubao-seedance-1-0-lite-t2v",
 	"doubao-seedance-1-0-lite-i2v",
+	"doubao-seedance-1-5-pro-251215",
 }
 
 var ChannelName = "doubao-video"

relay/channel/task/jimeng/adaptor.go

@@ -17,6 +17,7 @@ import (
 
 	"github.com/QuantumNous/new-api/common"
 	"github.com/QuantumNous/new-api/model"
+	"github.com/samber/lo"
 
 	"github.com/gin-gonic/gin"
 	"github.com/pkg/errors"
@@ -409,14 +410,15 @@ func (a *TaskAdaptor) convertToRequestPayload(req *relaycommon.TaskSubmitReq) (*
 
 	// 即梦视频3.0 ReqKey转换
 	// https://www.volcengine.com/docs/85621/1792707
+	imageLen := lo.Max([]int{len(req.Images), len(r.BinaryDataBase64), len(r.ImageUrls)})
 	if strings.Contains(r.ReqKey, "jimeng_v30") {
 		if r.ReqKey == "jimeng_v30_pro" {
 			// 3.0 pro只有固定的jimeng_ti2v_v30_pro
 			r.ReqKey = "jimeng_ti2v_v30_pro"
-		} else if len(req.Images) > 1 {
+		} else if imageLen > 1 {
 			// 多张图片：首尾帧生成
 			r.ReqKey = strings.TrimSuffix(strings.Replace(r.ReqKey, "jimeng_v30", "jimeng_i2v_first_tail_v30", 1), "p")
-		} else if len(req.Images) == 1 {
+		} else if imageLen == 1 {
 			// 单张图片：图生视频
 			r.ReqKey = strings.TrimSuffix(strings.Replace(r.ReqKey, "jimeng_v30", "jimeng_i2v_first_v30", 1), "p")
 		} else {

relay/channel/task/kling/adaptor.go

@@ -346,7 +346,7 @@ func (a *TaskAdaptor) ParseTaskResult(respBody []byte) (*relaycommon.TaskInfo, e
 	}
 	taskInfo.Code = resPayload.Code
 	taskInfo.TaskID = resPayload.Data.TaskId
-	taskInfo.Reason = resPayload.Message
+	taskInfo.Reason = resPayload.Data.TaskStatusMsg
 	//任务状态，枚举值：submitted（已提交）、processing（处理中）、succeed（成功）、failed（失败）
 	status := resPayload.Data.TaskStatus
 	switch status {

relay/channel/vertex/adaptor.go

@@ -40,6 +40,7 @@ var claudeModelMap = map[string]string{
 	"claude-opus-4-20250514":     "claude-opus-4@20250514",
 	"claude-opus-4-1-20250805":   "claude-opus-4-1@20250805",
 	"claude-sonnet-4-5-20250929": "claude-sonnet-4-5@20250929",
+	"claude-haiku-4-5-20251001":  "claude-haiku-4-5@20251001",
 	"claude-opus-4-5-20251101":   "claude-opus-4-5@20251101",
 }
 

relay/channel/volcengine/adaptor.go

@@ -270,6 +270,8 @@ func (a *Adaptor) GetRequestURL(info *relaycommon.RelayInfo) (string, error) {
 		//	return fmt.Sprintf("%s/api/v3/images/edits", baseUrl), nil
 		case constant.RelayModeRerank:
 			return fmt.Sprintf("%s/api/v3/rerank", baseUrl), nil
+		case constant.RelayModeResponses:
+			return fmt.Sprintf("%s/api/v3/responses", baseUrl), nil
 		case constant.RelayModeAudioSpeech:
 			if baseUrl == channelconstant.ChannelBaseURLs[channelconstant.ChannelTypeVolcEngine] {
 				return "wss://openspeech.bytedance.com/api/v1/tts/ws_binary", nil
@@ -323,7 +325,7 @@ func (a *Adaptor) ConvertEmbeddingRequest(c *gin.Context, info *relaycommon.Rela
 }
 
 func (a *Adaptor) ConvertOpenAIResponsesRequest(c *gin.Context, info *relaycommon.RelayInfo, request dto.OpenAIResponsesRequest) (any, error) {
-	return nil, errors.New("not implemented")
+	return request, nil
 }
 
 func (a *Adaptor) DoRequest(c *gin.Context, info *relaycommon.RelayInfo, requestBody io.Reader) (any, error) {

relay/chat_completions_via_responses.go

@@ -0,0 +1,162 @@
+package relay
+
+import (
+	"bytes"
+	"net/http"
+	"strings"
+
+	"github.com/QuantumNous/new-api/common"
+	"github.com/QuantumNous/new-api/constant"
+	"github.com/QuantumNous/new-api/dto"
+	"github.com/QuantumNous/new-api/relay/channel"
+	openaichannel "github.com/QuantumNous/new-api/relay/channel/openai"
+	relaycommon "github.com/QuantumNous/new-api/relay/common"
+	relayconstant "github.com/QuantumNous/new-api/relay/constant"
+	"github.com/QuantumNous/new-api/service"
+	"github.com/QuantumNous/new-api/types"
+
+	"github.com/gin-gonic/gin"
+)
+
+func applySystemPromptIfNeeded(c *gin.Context, info *relaycommon.RelayInfo, request *dto.GeneralOpenAIRequest) {
+	if info == nil || request == nil {
+		return
+	}
+	if info.ChannelSetting.SystemPrompt == "" {
+		return
+	}
+
+	systemRole := request.GetSystemRoleName()
+
+	containSystemPrompt := false
+	for _, message := range request.Messages {
+		if message.Role == systemRole {
+			containSystemPrompt = true
+			break
+		}
+	}
+	if !containSystemPrompt {
+		systemMessage := dto.Message{
+			Role:    systemRole,
+			Content: info.ChannelSetting.SystemPrompt,
+		}
+		request.Messages = append([]dto.Message{systemMessage}, request.Messages...)
+		return
+	}
+
+	if !info.ChannelSetting.SystemPromptOverride {
+		return
+	}
+
+	common.SetContextKey(c, constant.ContextKeySystemPromptOverride, true)
+	for i, message := range request.Messages {
+		if message.Role != systemRole {
+			continue
+		}
+		if message.IsStringContent() {
+			request.Messages[i].SetStringContent(info.ChannelSetting.SystemPrompt + "\n" + message.StringContent())
+			return
+		}
+		contents := message.ParseContent()
+		contents = append([]dto.MediaContent{
+			{
+				Type: dto.ContentTypeText,
+				Text: info.ChannelSetting.SystemPrompt,
+			},
+		}, contents...)
+		request.Messages[i].Content = contents
+		return
+	}
+}
+
+func chatCompletionsViaResponses(c *gin.Context, info *relaycommon.RelayInfo, adaptor channel.Adaptor, request *dto.GeneralOpenAIRequest) (*dto.Usage, *types.NewAPIError) {
+	overrideCtx := relaycommon.BuildParamOverrideContext(info)
+	chatJSON, err := common.Marshal(request)
+	if err != nil {
+		return nil, types.NewError(err, types.ErrorCodeConvertRequestFailed, types.ErrOptionWithSkipRetry())
+	}
+
+	chatJSON, err = relaycommon.RemoveDisabledFields(chatJSON, info.ChannelOtherSettings)
+	if err != nil {
+		return nil, types.NewError(err, types.ErrorCodeConvertRequestFailed, types.ErrOptionWithSkipRetry())
+	}
+
+	if len(info.ParamOverride) > 0 {
+		chatJSON, err = relaycommon.ApplyParamOverride(chatJSON, info.ParamOverride, overrideCtx)
+		if err != nil {
+			return nil, types.NewError(err, types.ErrorCodeChannelParamOverrideInvalid, types.ErrOptionWithSkipRetry())
+		}
+	}
+
+	var overriddenChatReq dto.GeneralOpenAIRequest
+	if err := common.Unmarshal(chatJSON, &overriddenChatReq); err != nil {
+		return nil, types.NewError(err, types.ErrorCodeChannelParamOverrideInvalid, types.ErrOptionWithSkipRetry())
+	}
+
+	responsesReq, err := service.ChatCompletionsRequestToResponsesRequest(&overriddenChatReq)
+	if err != nil {
+		return nil, types.NewErrorWithStatusCode(err, types.ErrorCodeInvalidRequest, http.StatusBadRequest, types.ErrOptionWithSkipRetry())
+	}
+	info.AppendRequestConversion(types.RelayFormatOpenAIResponses)
+
+	savedRelayMode := info.RelayMode
+	savedRequestURLPath := info.RequestURLPath
+	defer func() {
+		info.RelayMode = savedRelayMode
+		info.RequestURLPath = savedRequestURLPath
+	}()
+
+	info.RelayMode = relayconstant.RelayModeResponses
+	info.RequestURLPath = "/v1/responses"
+
+	convertedRequest, err := adaptor.ConvertOpenAIResponsesRequest(c, info, *responsesReq)
+	if err != nil {
+		return nil, types.NewError(err, types.ErrorCodeConvertRequestFailed, types.ErrOptionWithSkipRetry())
+	}
+	relaycommon.AppendRequestConversionFromRequest(info, convertedRequest)
+
+	jsonData, err := common.Marshal(convertedRequest)
+	if err != nil {
+		return nil, types.NewError(err, types.ErrorCodeConvertRequestFailed, types.ErrOptionWithSkipRetry())
+	}
+
+	jsonData, err = relaycommon.RemoveDisabledFields(jsonData, info.ChannelOtherSettings)
+	if err != nil {
+		return nil, types.NewError(err, types.ErrorCodeConvertRequestFailed, types.ErrOptionWithSkipRetry())
+	}
+
+	var httpResp *http.Response
+	resp, err := adaptor.DoRequest(c, info, bytes.NewBuffer(jsonData))
+	if err != nil {
+		return nil, types.NewOpenAIError(err, types.ErrorCodeDoRequestFailed, http.StatusInternalServerError)
+	}
+	if resp == nil {
+		return nil, types.NewOpenAIError(nil, types.ErrorCodeBadResponse, http.StatusInternalServerError)
+	}
+
+	statusCodeMappingStr := c.GetString("status_code_mapping")
+
+	httpResp = resp.(*http.Response)
+	info.IsStream = info.IsStream || strings.HasPrefix(httpResp.Header.Get("Content-Type"), "text/event-stream")
+	if httpResp.StatusCode != http.StatusOK {
+		newApiErr := service.RelayErrorHandler(c.Request.Context(), httpResp, false)
+		service.ResetStatusCode(newApiErr, statusCodeMappingStr)
+		return nil, newApiErr
+	}
+
+	if info.IsStream {
+		usage, newApiErr := openaichannel.OaiResponsesToChatStreamHandler(c, info, httpResp)
+		if newApiErr != nil {
+			service.ResetStatusCode(newApiErr, statusCodeMappingStr)
+			return nil, newApiErr
+		}
+		return usage, nil
+	}
+
+	usage, newApiErr := openaichannel.OaiResponsesToChatHandler(c, info, httpResp)
+	if newApiErr != nil {
+		service.ResetStatusCode(newApiErr, statusCodeMappingStr)
+		return nil, newApiErr
+	}
+	return usage, nil
+}

relay/claude_handler.go

@@ -110,6 +110,7 @@ func ClaudeHelper(c *gin.Context, info *relaycommon.RelayInfo) (newAPIError *typ
 		if err != nil {
 			return types.NewError(err, types.ErrorCodeConvertRequestFailed, types.ErrOptionWithSkipRetry())
 		}
+		relaycommon.AppendRequestConversionFromRequest(info, convertedRequest)
 		jsonData, err := common.Marshal(convertedRequest)
 		if err != nil {
 			return types.NewError(err, types.ErrorCodeConvertRequestFailed, types.ErrOptionWithSkipRetry())

relay/common/override.go

@@ -23,7 +23,7 @@ type ConditionOperation struct {
 
 type ParamOperation struct {
 	Path       string               `json:"path"`
-	Mode       string               `json:"mode"` // delete, set, move, prepend, append
+	Mode       string               `json:"mode"` // delete, set, move, copy, prepend, append, trim_prefix, trim_suffix, ensure_prefix, ensure_suffix, trim_space, to_lower, to_upper, replace, regex_replace
 	Value      interface{}          `json:"value"`
 	KeepOrigin bool                 `json:"keep_origin"`
 	From       string               `json:"from,omitempty"`
@@ -330,8 +330,6 @@ func applyOperations(jsonStr string, operations []ParamOperation, conditionConte
 		}
 		// 处理路径中的负数索引
 		opPath := processNegativeIndex(result, op.Path)
-		opFrom := processNegativeIndex(result, op.From)
-		opTo := processNegativeIndex(result, op.To)
 
 		switch op.Mode {
 		case "delete":
@@ -342,11 +340,38 @@ func applyOperations(jsonStr string, operations []ParamOperation, conditionConte
 			}
 			result, err = sjson.Set(result, opPath, op.Value)
 		case "move":
+			opFrom := processNegativeIndex(result, op.From)
+			opTo := processNegativeIndex(result, op.To)
 			result, err = moveValue(result, opFrom, opTo)
+		case "copy":
+			if op.From == "" || op.To == "" {
+				return "", fmt.Errorf("copy from/to is required")
+			}
+			opFrom := processNegativeIndex(result, op.From)
+			opTo := processNegativeIndex(result, op.To)
+			result, err = copyValue(result, opFrom, opTo)
 		case "prepend":
 			result, err = modifyValue(result, opPath, op.Value, op.KeepOrigin, true)
 		case "append":
 			result, err = modifyValue(result, opPath, op.Value, op.KeepOrigin, false)
+		case "trim_prefix":
+			result, err = trimStringValue(result, opPath, op.Value, true)
+		case "trim_suffix":
+			result, err = trimStringValue(result, opPath, op.Value, false)
+		case "ensure_prefix":
+			result, err = ensureStringAffix(result, opPath, op.Value, true)
+		case "ensure_suffix":
+			result, err = ensureStringAffix(result, opPath, op.Value, false)
+		case "trim_space":
+			result, err = transformStringValue(result, opPath, strings.TrimSpace)
+		case "to_lower":
+			result, err = transformStringValue(result, opPath, strings.ToLower)
+		case "to_upper":
+			result, err = transformStringValue(result, opPath, strings.ToUpper)
+		case "replace":
+			result, err = replaceStringValue(result, opPath, op.From, op.To)
+		case "regex_replace":
+			result, err = regexReplaceStringValue(result, opPath, op.From, op.To)
 		default:
 			return "", fmt.Errorf("unknown operation: %s", op.Mode)
 		}
@@ -369,6 +394,14 @@ func moveValue(jsonStr, fromPath, toPath string) (string, error) {
 	return sjson.Delete(result, fromPath)
 }
 
+func copyValue(jsonStr, fromPath, toPath string) (string, error) {
+	sourceValue := gjson.Get(jsonStr, fromPath)
+	if !sourceValue.Exists() {
+		return jsonStr, fmt.Errorf("source path does not exist: %s", fromPath)
+	}
+	return sjson.Set(jsonStr, toPath, sourceValue.Value())
+}
+
 func modifyValue(jsonStr, path string, value interface{}, keepOrigin, isPrepend bool) (string, error) {
 	current := gjson.Get(jsonStr, path)
 	switch {
@@ -422,6 +455,88 @@ func modifyString(jsonStr, path string, value interface{}, isPrepend bool) (stri
 	return sjson.Set(jsonStr, path, newStr)
 }
 
+func trimStringValue(jsonStr, path string, value interface{}, isPrefix bool) (string, error) {
+	current := gjson.Get(jsonStr, path)
+	if current.Type != gjson.String {
+		return jsonStr, fmt.Errorf("operation not supported for type: %v", current.Type)
+	}
+
+	if value == nil {
+		return jsonStr, fmt.Errorf("trim value is required")
+	}
+	valueStr := fmt.Sprintf("%v", value)
+
+	var newStr string
+	if isPrefix {
+		newStr = strings.TrimPrefix(current.String(), valueStr)
+	} else {
+		newStr = strings.TrimSuffix(current.String(), valueStr)
+	}
+	return sjson.Set(jsonStr, path, newStr)
+}
+
+func ensureStringAffix(jsonStr, path string, value interface{}, isPrefix bool) (string, error) {
+	current := gjson.Get(jsonStr, path)
+	if current.Type != gjson.String {
+		return jsonStr, fmt.Errorf("operation not supported for type: %v", current.Type)
+	}
+
+	if value == nil {
+		return jsonStr, fmt.Errorf("ensure value is required")
+	}
+	valueStr := fmt.Sprintf("%v", value)
+	if valueStr == "" {
+		return jsonStr, fmt.Errorf("ensure value is required")
+	}
+
+	currentStr := current.String()
+	if isPrefix {
+		if strings.HasPrefix(currentStr, valueStr) {
+			return jsonStr, nil
+		}
+		return sjson.Set(jsonStr, path, valueStr+currentStr)
+	}
+
+	if strings.HasSuffix(currentStr, valueStr) {
+		return jsonStr, nil
+	}
+	return sjson.Set(jsonStr, path, currentStr+valueStr)
+}
+
+func transformStringValue(jsonStr, path string, transform func(string) string) (string, error) {
+	current := gjson.Get(jsonStr, path)
+	if current.Type != gjson.String {
+		return jsonStr, fmt.Errorf("operation not supported for type: %v", current.Type)
+	}
+	return sjson.Set(jsonStr, path, transform(current.String()))
+}
+
+func replaceStringValue(jsonStr, path, from, to string) (string, error) {
+	current := gjson.Get(jsonStr, path)
+	if current.Type != gjson.String {
+		return jsonStr, fmt.Errorf("operation not supported for type: %v", current.Type)
+	}
+	if from == "" {
+		return jsonStr, fmt.Errorf("replace from is required")
+	}
+	return sjson.Set(jsonStr, path, strings.ReplaceAll(current.String(), from, to))
+}
+
+func regexReplaceStringValue(jsonStr, path, pattern, replacement string) (string, error) {
+	current := gjson.Get(jsonStr, path)
+	if current.Type != gjson.String {
+		return jsonStr, fmt.Errorf("operation not supported for type: %v", current.Type)
+	}
+	if pattern == "" {
+		return jsonStr, fmt.Errorf("regex pattern is required")
+	}
+	re, err := regexp.Compile(pattern)
+	if err != nil {
+		return jsonStr, err
+	}
+	return sjson.Set(jsonStr, path, re.ReplaceAllString(current.String(), replacement))
+}
+
 func mergeObjects(jsonStr, path string, value interface{}, keepOrigin bool) (string, error) {
 	current := gjson.Get(jsonStr, path)
 	var currentMap, newMap map[string]interface{}
@@ -455,18 +570,19 @@ func mergeObjects(jsonStr, path string, value interface{}, keepOrigin bool) (str
 
 // BuildParamOverrideContext 提供 ApplyParamOverride 可用的上下文信息。
 // 目前内置以下字段：
-//   - model：优先使用上游模型名（UpstreamModelName），若不存在则回落到原始模型名（OriginModelName）。
-//   - upstream_model：始终为通道映射后的上游模型名。
+//   - upstream_model/model：始终为通道映射后的上游模型名。
 //   - original_model：请求最初指定的模型名。
+//   - request_path：请求路径
+//   - is_channel_test：是否为渠道测试请求（同 is_test）。
 func BuildParamOverrideContext(info *RelayInfo) map[string]interface{} {
-	if info == nil || info.ChannelMeta == nil {
+	if info == nil {
 		return nil
 	}
 
 	ctx := make(map[string]interface{})
-	if info.UpstreamModelName != "" {
-		ctx["model"] = info.UpstreamModelName
-		ctx["upstream_model"] = info.UpstreamModelName
+	if info.ChannelMeta != nil && info.ChannelMeta.UpstreamModelName != "" {
+		ctx["model"] = info.ChannelMeta.UpstreamModelName
+		ctx["upstream_model"] = info.ChannelMeta.UpstreamModelName
 	}
 	if info.OriginModelName != "" {
 		ctx["original_model"] = info.OriginModelName
@@ -475,8 +591,13 @@ func BuildParamOverrideContext(info *RelayInfo) map[string]interface{} {
 		}
 	}
 
-	if len(ctx) == 0 {
-		return nil
+	if info.RequestURLPath != "" {
+		requestPath := info.RequestURLPath
+		if requestPath != "" {
+			ctx["request_path"] = requestPath
+		}
 	}
+
+	ctx["is_channel_test"] = info.IsChannelTest
 	return ctx
 }

relay/common/override_test.go

@@ -0,0 +1,791 @@
+package common
+
+import (
+	"encoding/json"
+	"reflect"
+	"testing"
+)
+
+func TestApplyParamOverrideTrimPrefix(t *testing.T) {
+	// trim_prefix example:
+	// {"operations":[{"path":"model","mode":"trim_prefix","value":"openai/"}]}
+	input := []byte(`{"model":"openai/gpt-4","temperature":0.7}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path":  "model",
+				"mode":  "trim_prefix",
+				"value": "openai/",
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"model":"gpt-4","temperature":0.7}`, string(out))
+}
+
+func TestApplyParamOverrideTrimSuffix(t *testing.T) {
+	// trim_suffix example:
+	// {"operations":[{"path":"model","mode":"trim_suffix","value":"-latest"}]}
+	input := []byte(`{"model":"gpt-4-latest","temperature":0.7}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path":  "model",
+				"mode":  "trim_suffix",
+				"value": "-latest",
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"model":"gpt-4","temperature":0.7}`, string(out))
+}
+
+func TestApplyParamOverrideTrimNoop(t *testing.T) {
+	// trim_prefix no-op example:
+	// {"operations":[{"path":"model","mode":"trim_prefix","value":"openai/"}]}
+	input := []byte(`{"model":"gpt-4","temperature":0.7}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path":  "model",
+				"mode":  "trim_prefix",
+				"value": "openai/",
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"model":"gpt-4","temperature":0.7}`, string(out))
+}
+
+func TestApplyParamOverrideTrimRequiresValue(t *testing.T) {
+	// trim_prefix requires value example:
+	// {"operations":[{"path":"model","mode":"trim_prefix"}]}
+	input := []byte(`{"model":"gpt-4"}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path": "model",
+				"mode": "trim_prefix",
+			},
+		},
+	}
+
+	_, err := ApplyParamOverride(input, override, nil)
+	if err == nil {
+		t.Fatalf("expected error, got nil")
+	}
+}
+
+func TestApplyParamOverrideReplace(t *testing.T) {
+	// replace example:
+	// {"operations":[{"path":"model","mode":"replace","from":"openai/","to":""}]}
+	input := []byte(`{"model":"openai/gpt-4o-mini","temperature":0.7}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path": "model",
+				"mode": "replace",
+				"from": "openai/",
+				"to":   "",
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"model":"gpt-4o-mini","temperature":0.7}`, string(out))
+}
+
+func TestApplyParamOverrideRegexReplace(t *testing.T) {
+	// regex_replace example:
+	// {"operations":[{"path":"model","mode":"regex_replace","from":"^gpt-","to":"openai/gpt-"}]}
+	input := []byte(`{"model":"gpt-4o-mini","temperature":0.7}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path": "model",
+				"mode": "regex_replace",
+				"from": "^gpt-",
+				"to":   "openai/gpt-",
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"model":"openai/gpt-4o-mini","temperature":0.7}`, string(out))
+}
+
+func TestApplyParamOverrideReplaceRequiresFrom(t *testing.T) {
+	// replace requires from example:
+	// {"operations":[{"path":"model","mode":"replace"}]}
+	input := []byte(`{"model":"gpt-4"}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path": "model",
+				"mode": "replace",
+			},
+		},
+	}
+
+	_, err := ApplyParamOverride(input, override, nil)
+	if err == nil {
+		t.Fatalf("expected error, got nil")
+	}
+}
+
+func TestApplyParamOverrideRegexReplaceRequiresPattern(t *testing.T) {
+	// regex_replace requires from(pattern) example:
+	// {"operations":[{"path":"model","mode":"regex_replace"}]}
+	input := []byte(`{"model":"gpt-4"}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path": "model",
+				"mode": "regex_replace",
+			},
+		},
+	}
+
+	_, err := ApplyParamOverride(input, override, nil)
+	if err == nil {
+		t.Fatalf("expected error, got nil")
+	}
+}
+
+func TestApplyParamOverrideDelete(t *testing.T) {
+	input := []byte(`{"model":"gpt-4","temperature":0.7}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path": "temperature",
+				"mode": "delete",
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+
+	var got map[string]interface{}
+	if err := json.Unmarshal(out, &got); err != nil {
+		t.Fatalf("failed to unmarshal output JSON: %v", err)
+	}
+	if _, exists := got["temperature"]; exists {
+		t.Fatalf("expected temperature to be deleted")
+	}
+}
+
+func TestApplyParamOverrideSet(t *testing.T) {
+	input := []byte(`{"model":"gpt-4","temperature":0.7}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path":  "temperature",
+				"mode":  "set",
+				"value": 0.1,
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"model":"gpt-4","temperature":0.1}`, string(out))
+}
+
+func TestApplyParamOverrideSetKeepOrigin(t *testing.T) {
+	input := []byte(`{"model":"gpt-4","temperature":0.7}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path":        "temperature",
+				"mode":        "set",
+				"value":       0.1,
+				"keep_origin": true,
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"model":"gpt-4","temperature":0.7}`, string(out))
+}
+
+func TestApplyParamOverrideMove(t *testing.T) {
+	input := []byte(`{"model":"gpt-4","meta":{"x":1}}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"mode": "move",
+				"from": "model",
+				"to":   "meta.model",
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"meta":{"x":1,"model":"gpt-4"}}`, string(out))
+}
+
+func TestApplyParamOverrideMoveMissingSource(t *testing.T) {
+	input := []byte(`{"meta":{"x":1}}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"mode": "move",
+				"from": "model",
+				"to":   "meta.model",
+			},
+		},
+	}
+
+	_, err := ApplyParamOverride(input, override, nil)
+	if err == nil {
+		t.Fatalf("expected error, got nil")
+	}
+}
+
+func TestApplyParamOverridePrependAppendString(t *testing.T) {
+	input := []byte(`{"model":"gpt-4"}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path":  "model",
+				"mode":  "prepend",
+				"value": "openai/",
+			},
+			map[string]interface{}{
+				"path":  "model",
+				"mode":  "append",
+				"value": "-latest",
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"model":"openai/gpt-4-latest"}`, string(out))
+}
+
+func TestApplyParamOverridePrependAppendArray(t *testing.T) {
+	input := []byte(`{"arr":[1,2]}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path":  "arr",
+				"mode":  "prepend",
+				"value": 0,
+			},
+			map[string]interface{}{
+				"path":  "arr",
+				"mode":  "append",
+				"value": []interface{}{3, 4},
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"arr":[0,1,2,3,4]}`, string(out))
+}
+
+func TestApplyParamOverrideAppendObjectMergeKeepOrigin(t *testing.T) {
+	input := []byte(`{"obj":{"a":1}}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path":        "obj",
+				"mode":        "append",
+				"keep_origin": true,
+				"value": map[string]interface{}{
+					"a": 2,
+					"b": 3,
+				},
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"obj":{"a":1,"b":3}}`, string(out))
+}
+
+func TestApplyParamOverrideAppendObjectMergeOverride(t *testing.T) {
+	input := []byte(`{"obj":{"a":1}}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path": "obj",
+				"mode": "append",
+				"value": map[string]interface{}{
+					"a": 2,
+					"b": 3,
+				},
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"obj":{"a":2,"b":3}}`, string(out))
+}
+
+func TestApplyParamOverrideConditionORDefault(t *testing.T) {
+	input := []byte(`{"model":"gpt-4","temperature":0.7}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path":  "temperature",
+				"mode":  "set",
+				"value": 0.1,
+				"conditions": []interface{}{
+					map[string]interface{}{
+						"path":  "model",
+						"mode":  "prefix",
+						"value": "gpt",
+					},
+					map[string]interface{}{
+						"path":  "model",
+						"mode":  "prefix",
+						"value": "claude",
+					},
+				},
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"model":"gpt-4","temperature":0.1}`, string(out))
+}
+
+func TestApplyParamOverrideConditionAND(t *testing.T) {
+	input := []byte(`{"model":"gpt-4","temperature":0.7}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path":  "temperature",
+				"mode":  "set",
+				"value": 0.1,
+				"logic": "AND",
+				"conditions": []interface{}{
+					map[string]interface{}{
+						"path":  "model",
+						"mode":  "prefix",
+						"value": "gpt",
+					},
+					map[string]interface{}{
+						"path":  "temperature",
+						"mode":  "gt",
+						"value": 0.5,
+					},
+				},
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"model":"gpt-4","temperature":0.1}`, string(out))
+}
+
+func TestApplyParamOverrideConditionInvert(t *testing.T) {
+	input := []byte(`{"model":"gpt-4","temperature":0.7}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path":  "temperature",
+				"mode":  "set",
+				"value": 0.1,
+				"conditions": []interface{}{
+					map[string]interface{}{
+						"path":   "model",
+						"mode":   "prefix",
+						"value":  "gpt",
+						"invert": true,
+					},
+				},
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"model":"gpt-4","temperature":0.7}`, string(out))
+}
+
+func TestApplyParamOverrideConditionPassMissingKey(t *testing.T) {
+	input := []byte(`{"temperature":0.7}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path":  "temperature",
+				"mode":  "set",
+				"value": 0.1,
+				"conditions": []interface{}{
+					map[string]interface{}{
+						"path":             "model",
+						"mode":             "prefix",
+						"value":            "gpt",
+						"pass_missing_key": true,
+					},
+				},
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"temperature":0.1}`, string(out))
+}
+
+func TestApplyParamOverrideConditionFromContext(t *testing.T) {
+	input := []byte(`{"temperature":0.7}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path":  "temperature",
+				"mode":  "set",
+				"value": 0.1,
+				"conditions": []interface{}{
+					map[string]interface{}{
+						"path":  "model",
+						"mode":  "prefix",
+						"value": "gpt",
+					},
+				},
+			},
+		},
+	}
+	ctx := map[string]interface{}{
+		"model": "gpt-4",
+	}
+
+	out, err := ApplyParamOverride(input, override, ctx)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"temperature":0.1}`, string(out))
+}
+
+func TestApplyParamOverrideNegativeIndexPath(t *testing.T) {
+	input := []byte(`{"arr":[{"model":"a"},{"model":"b"}]}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path":  "arr.-1.model",
+				"mode":  "set",
+				"value": "c",
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"arr":[{"model":"a"},{"model":"c"}]}`, string(out))
+}
+
+func TestApplyParamOverrideRegexReplaceInvalidPattern(t *testing.T) {
+	// regex_replace invalid pattern example:
+	// {"operations":[{"path":"model","mode":"regex_replace","from":"(","to":"x"}]}
+	input := []byte(`{"model":"gpt-4"}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path": "model",
+				"mode": "regex_replace",
+				"from": "(",
+				"to":   "x",
+			},
+		},
+	}
+
+	_, err := ApplyParamOverride(input, override, nil)
+	if err == nil {
+		t.Fatalf("expected error, got nil")
+	}
+}
+
+func TestApplyParamOverrideCopy(t *testing.T) {
+	// copy example:
+	// {"operations":[{"mode":"copy","from":"model","to":"original_model"}]}
+	input := []byte(`{"model":"gpt-4","temperature":0.7}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"mode": "copy",
+				"from": "model",
+				"to":   "original_model",
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"model":"gpt-4","original_model":"gpt-4","temperature":0.7}`, string(out))
+}
+
+func TestApplyParamOverrideCopyMissingSource(t *testing.T) {
+	// copy missing source example:
+	// {"operations":[{"mode":"copy","from":"model","to":"original_model"}]}
+	input := []byte(`{"temperature":0.7}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"mode": "copy",
+				"from": "model",
+				"to":   "original_model",
+			},
+		},
+	}
+
+	_, err := ApplyParamOverride(input, override, nil)
+	if err == nil {
+		t.Fatalf("expected error, got nil")
+	}
+}
+
+func TestApplyParamOverrideCopyRequiresFromTo(t *testing.T) {
+	// copy requires from/to example:
+	// {"operations":[{"mode":"copy"}]}
+	input := []byte(`{"model":"gpt-4"}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"mode": "copy",
+			},
+		},
+	}
+
+	_, err := ApplyParamOverride(input, override, nil)
+	if err == nil {
+		t.Fatalf("expected error, got nil")
+	}
+}
+
+func TestApplyParamOverrideEnsurePrefix(t *testing.T) {
+	// ensure_prefix example:
+	// {"operations":[{"path":"model","mode":"ensure_prefix","value":"openai/"}]}
+	input := []byte(`{"model":"gpt-4"}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path":  "model",
+				"mode":  "ensure_prefix",
+				"value": "openai/",
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"model":"openai/gpt-4"}`, string(out))
+}
+
+func TestApplyParamOverrideEnsurePrefixNoop(t *testing.T) {
+	// ensure_prefix no-op example:
+	// {"operations":[{"path":"model","mode":"ensure_prefix","value":"openai/"}]}
+	input := []byte(`{"model":"openai/gpt-4"}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path":  "model",
+				"mode":  "ensure_prefix",
+				"value": "openai/",
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"model":"openai/gpt-4"}`, string(out))
+}
+
+func TestApplyParamOverrideEnsureSuffix(t *testing.T) {
+	// ensure_suffix example:
+	// {"operations":[{"path":"model","mode":"ensure_suffix","value":"-latest"}]}
+	input := []byte(`{"model":"gpt-4"}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path":  "model",
+				"mode":  "ensure_suffix",
+				"value": "-latest",
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"model":"gpt-4-latest"}`, string(out))
+}
+
+func TestApplyParamOverrideEnsureSuffixNoop(t *testing.T) {
+	// ensure_suffix no-op example:
+	// {"operations":[{"path":"model","mode":"ensure_suffix","value":"-latest"}]}
+	input := []byte(`{"model":"gpt-4-latest"}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path":  "model",
+				"mode":  "ensure_suffix",
+				"value": "-latest",
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"model":"gpt-4-latest"}`, string(out))
+}
+
+func TestApplyParamOverrideEnsureRequiresValue(t *testing.T) {
+	// ensure_prefix requires value example:
+	// {"operations":[{"path":"model","mode":"ensure_prefix"}]}
+	input := []byte(`{"model":"gpt-4"}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path": "model",
+				"mode": "ensure_prefix",
+			},
+		},
+	}
+
+	_, err := ApplyParamOverride(input, override, nil)
+	if err == nil {
+		t.Fatalf("expected error, got nil")
+	}
+}
+
+func TestApplyParamOverrideTrimSpace(t *testing.T) {
+	// trim_space example:
+	// {"operations":[{"path":"model","mode":"trim_space"}]}
+	input := []byte("{\"model\":\"  gpt-4 \\n\"}")
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path": "model",
+				"mode": "trim_space",
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"model":"gpt-4"}`, string(out))
+}
+
+func TestApplyParamOverrideToLower(t *testing.T) {
+	// to_lower example:
+	// {"operations":[{"path":"model","mode":"to_lower"}]}
+	input := []byte(`{"model":"GPT-4"}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path": "model",
+				"mode": "to_lower",
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"model":"gpt-4"}`, string(out))
+}
+
+func TestApplyParamOverrideToUpper(t *testing.T) {
+	// to_upper example:
+	// {"operations":[{"path":"model","mode":"to_upper"}]}
+	input := []byte(`{"model":"gpt-4"}`)
+	override := map[string]interface{}{
+		"operations": []interface{}{
+			map[string]interface{}{
+				"path": "model",
+				"mode": "to_upper",
+			},
+		},
+	}
+
+	out, err := ApplyParamOverride(input, override, nil)
+	if err != nil {
+		t.Fatalf("ApplyParamOverride returned error: %v", err)
+	}
+	assertJSONEqual(t, `{"model":"GPT-4"}`, string(out))
+}
+
+func assertJSONEqual(t *testing.T, want, got string) {
+	t.Helper()
+
+	var wantObj interface{}
+	var gotObj interface{}
+
+	if err := json.Unmarshal([]byte(want), &wantObj); err != nil {
+		t.Fatalf("failed to unmarshal want JSON: %v", err)
+	}
+	if err := json.Unmarshal([]byte(got), &gotObj); err != nil {
+		t.Fatalf("failed to unmarshal got JSON: %v", err)
+	}
+
+	if !reflect.DeepEqual(wantObj, gotObj) {
+		t.Fatalf("json not equal\nwant: %s\ngot:  %s", want, got)
+	}
+}

relay/common/relay_info.go

@@ -115,11 +115,16 @@ type RelayInfo struct {
 	SendResponseCount      int
 	FinalPreConsumedQuota  int  // 最终预消耗的配额
 	IsClaudeBetaQuery      bool // /v1/messages?beta=true
+	IsChannelTest          bool // channel test request
 
 	PriceData types.PriceData
 
 	Request dto.Request
 
+	// RequestConversionChain records request format conversions in order, e.g.
+	// ["openai", "openai_responses"] or ["openai", "claude"].
+	RequestConversionChain []types.RelayFormat
+
 	ThinkingContentInfo
 	TokenCountMeta
 	*ClaudeConvertInfo
@@ -273,6 +278,7 @@ var streamSupportedChannels = map[int]bool{
 	constant.ChannelTypeZhipu_v4:   true,
 	constant.ChannelTypeAli:        true,
 	constant.ChannelTypeSubmodel:   true,
+	constant.ChannelTypeCodex:      true,
 }
 
 func GenRelayInfoWs(c *gin.Context, ws *websocket.Conn) *RelayInfo {
@@ -446,38 +452,83 @@ func genBaseRelayInfo(c *gin.Context, request dto.Request) *RelayInfo {
 }
 
 func GenRelayInfo(c *gin.Context, relayFormat types.RelayFormat, request dto.Request, ws *websocket.Conn) (*RelayInfo, error) {
+	var info *RelayInfo
+	var err error
 	switch relayFormat {
 	case types.RelayFormatOpenAI:
-		return GenRelayInfoOpenAI(c, request), nil
+		info = GenRelayInfoOpenAI(c, request)
 	case types.RelayFormatOpenAIAudio:
-		return GenRelayInfoOpenAIAudio(c, request), nil
+		info = GenRelayInfoOpenAIAudio(c, request)
 	case types.RelayFormatOpenAIImage:
-		return GenRelayInfoImage(c, request), nil
+		info = GenRelayInfoImage(c, request)
 	case types.RelayFormatOpenAIRealtime:
-		return GenRelayInfoWs(c, ws), nil
+		info = GenRelayInfoWs(c, ws)
 	case types.RelayFormatClaude:
-		return GenRelayInfoClaude(c, request), nil
+		info = GenRelayInfoClaude(c, request)
 	case types.RelayFormatRerank:
 		if request, ok := request.(*dto.RerankRequest); ok {
-			return GenRelayInfoRerank(c, request), nil
+			info = GenRelayInfoRerank(c, request)
+			break
 		}
-		return nil, errors.New("request is not a RerankRequest")
+		err = errors.New("request is not a RerankRequest")
 	case types.RelayFormatGemini:
-		return GenRelayInfoGemini(c, request), nil
+		info = GenRelayInfoGemini(c, request)
 	case types.RelayFormatEmbedding:
-		return GenRelayInfoEmbedding(c, request), nil
+		info = GenRelayInfoEmbedding(c, request)
 	case types.RelayFormatOpenAIResponses:
 		if request, ok := request.(*dto.OpenAIResponsesRequest); ok {
-			return GenRelayInfoResponses(c, request), nil
+			info = GenRelayInfoResponses(c, request)
+			break
 		}
-		return nil, errors.New("request is not a OpenAIResponsesRequest")
+		err = errors.New("request is not a OpenAIResponsesRequest")
 	case types.RelayFormatTask:
-		return genBaseRelayInfo(c, nil), nil
+		info = genBaseRelayInfo(c, nil)
 	case types.RelayFormatMjProxy:
-		return genBaseRelayInfo(c, nil), nil
+		info = genBaseRelayInfo(c, nil)
 	default:
-		return nil, errors.New("invalid relay format")
+		err = errors.New("invalid relay format")
 	}
+
+	if err != nil {
+		return nil, err
+	}
+	if info == nil {
+		return nil, errors.New("failed to build relay info")
+	}
+
+	info.InitRequestConversionChain()
+	return info, nil
+}
+
+func (info *RelayInfo) InitRequestConversionChain() {
+	if info == nil {
+		return
+	}
+	if len(info.RequestConversionChain) > 0 {
+		return
+	}
+	if info.RelayFormat == "" {
+		return
+	}
+	info.RequestConversionChain = []types.RelayFormat{info.RelayFormat}
+}
+
+func (info *RelayInfo) AppendRequestConversion(format types.RelayFormat) {
+	if info == nil {
+		return
+	}
+	if format == "" {
+		return
+	}
+	if len(info.RequestConversionChain) == 0 {
+		info.RequestConversionChain = []types.RelayFormat{format}
+		return
+	}
+	last := info.RequestConversionChain[len(info.RequestConversionChain)-1]
+	if last == format {
+		return
+	}
+	info.RequestConversionChain = append(info.RequestConversionChain, format)
 }
 
 //func (info *RelayInfo) SetPromptTokens(promptTokens int) {

relay/common/request_conversion.go

@@ -0,0 +1,40 @@
+package common
+
+import (
+	"github.com/QuantumNous/new-api/dto"
+	"github.com/QuantumNous/new-api/types"
+)
+
+func GuessRelayFormatFromRequest(req any) (types.RelayFormat, bool) {
+	switch req.(type) {
+	case *dto.GeneralOpenAIRequest, dto.GeneralOpenAIRequest:
+		return types.RelayFormatOpenAI, true
+	case *dto.OpenAIResponsesRequest, dto.OpenAIResponsesRequest:
+		return types.RelayFormatOpenAIResponses, true
+	case *dto.ClaudeRequest, dto.ClaudeRequest:
+		return types.RelayFormatClaude, true
+	case *dto.GeminiChatRequest, dto.GeminiChatRequest:
+		return types.RelayFormatGemini, true
+	case *dto.EmbeddingRequest, dto.EmbeddingRequest:
+		return types.RelayFormatEmbedding, true
+	case *dto.RerankRequest, dto.RerankRequest:
+		return types.RelayFormatRerank, true
+	case *dto.ImageRequest, dto.ImageRequest:
+		return types.RelayFormatOpenAIImage, true
+	case *dto.AudioRequest, dto.AudioRequest:
+		return types.RelayFormatOpenAIAudio, true
+	default:
+		return "", false
+	}
+}
+
+func AppendRequestConversionFromRequest(info *RelayInfo, req any) {
+	if info == nil {
+		return
+	}
+	format, ok := GuessRelayFormatFromRequest(req)
+	if !ok {
+		return
+	}
+	info.AppendRequestConversion(format)
+}

relay/compatible_handler.go

@@ -14,10 +14,12 @@ import (
 	"github.com/QuantumNous/new-api/logger"
 	"github.com/QuantumNous/new-api/model"
 	relaycommon "github.com/QuantumNous/new-api/relay/common"
+	relayconstant "github.com/QuantumNous/new-api/relay/constant"
 	"github.com/QuantumNous/new-api/relay/helper"
 	"github.com/QuantumNous/new-api/service"
 	"github.com/QuantumNous/new-api/setting/model_setting"
 	"github.com/QuantumNous/new-api/setting/operation_setting"
+	"github.com/QuantumNous/new-api/setting/ratio_setting"
 	"github.com/QuantumNous/new-api/types"
 
 	"github.com/shopspring/decimal"
@@ -72,9 +74,32 @@ func TextHelper(c *gin.Context, info *relaycommon.RelayInfo) (newAPIError *types
 		return types.NewError(fmt.Errorf("invalid api type: %d", info.ApiType), types.ErrorCodeInvalidApiType, types.ErrOptionWithSkipRetry())
 	}
 	adaptor.Init(info)
+
+	passThroughGlobal := model_setting.GetGlobalSettings().PassThroughRequestEnabled
+	if info.RelayMode == relayconstant.RelayModeChatCompletions &&
+		!passThroughGlobal &&
+		!info.ChannelSetting.PassThroughBodyEnabled &&
+		shouldChatCompletionsViaResponses(info) {
+		applySystemPromptIfNeeded(c, info, request)
+		usage, newApiErr := chatCompletionsViaResponses(c, info, adaptor, request)
+		if newApiErr != nil {
+			return newApiErr
+		}
+
+		var containAudioTokens = usage.CompletionTokenDetails.AudioTokens > 0 || usage.PromptTokensDetails.AudioTokens > 0
+		var containsAudioRatios = ratio_setting.ContainsAudioRatio(info.OriginModelName) || ratio_setting.ContainsAudioCompletionRatio(info.OriginModelName)
+
+		if containAudioTokens && containsAudioRatios {
+			service.PostAudioConsumeQuota(c, info, usage, "")
+		} else {
+			postConsumeQuota(c, info, usage)
+		}
+		return nil
+	}
+
 	var requestBody io.Reader
 
-	if model_setting.GetGlobalSettings().PassThroughRequestEnabled || info.ChannelSetting.PassThroughBodyEnabled {
+	if passThroughGlobal || info.ChannelSetting.PassThroughBodyEnabled {
 		body, err := common.GetRequestBody(c)
 		if err != nil {
 			return types.NewErrorWithStatusCode(err, types.ErrorCodeReadRequestBodyFailed, http.StatusBadRequest, types.ErrOptionWithSkipRetry())
@@ -88,6 +113,7 @@ func TextHelper(c *gin.Context, info *relaycommon.RelayInfo) (newAPIError *types
 		if err != nil {
 			return types.NewError(err, types.ErrorCodeConvertRequestFailed, types.ErrOptionWithSkipRetry())
 		}
+		relaycommon.AppendRequestConversionFromRequest(info, convertedRequest)
 
 		if info.ChannelSetting.SystemPrompt != "" {
 			// 如果有系统提示，则将其添加到请求中
@@ -181,22 +207,35 @@ func TextHelper(c *gin.Context, info *relaycommon.RelayInfo) (newAPIError *types
 		return newApiErr
 	}
 
-	if usage.(*dto.Usage).CompletionTokenDetails.AudioTokens > 0 || usage.(*dto.Usage).PromptTokensDetails.AudioTokens > 0 {
+	var containAudioTokens = usage.(*dto.Usage).CompletionTokenDetails.AudioTokens > 0 || usage.(*dto.Usage).PromptTokensDetails.AudioTokens > 0
+	var containsAudioRatios = ratio_setting.ContainsAudioRatio(info.OriginModelName) || ratio_setting.ContainsAudioCompletionRatio(info.OriginModelName)
+
+	if containAudioTokens && containsAudioRatios {
 		service.PostAudioConsumeQuota(c, info, usage.(*dto.Usage), "")
 	} else {
-		postConsumeQuota(c, info, usage.(*dto.Usage), "")
+		postConsumeQuota(c, info, usage.(*dto.Usage))
 	}
 	return nil
 }
 
-func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, usage *dto.Usage, extraContent string) {
+func shouldChatCompletionsViaResponses(info *relaycommon.RelayInfo) bool {
+	if info == nil {
+		return false
+	}
+	if info.RelayMode != relayconstant.RelayModeChatCompletions {
+		return false
+	}
+	return service.ShouldChatCompletionsUseResponsesGlobal(info.ChannelId, info.OriginModelName)
+}
+
+func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, usage *dto.Usage, extraContent ...string) {
 	if usage == nil {
 		usage = &dto.Usage{
 			PromptTokens:     relayInfo.GetEstimatePromptTokens(),
 			CompletionTokens: 0,
 			TotalTokens:      relayInfo.GetEstimatePromptTokens(),
 		}
-		extraContent += "（可能是请求出错）"
+		extraContent = append(extraContent, "上游无计费信息")
 	}
 	useTimeSeconds := time.Now().Unix() - relayInfo.StartTime.Unix()
 	promptTokens := usage.PromptTokens
@@ -246,8 +285,8 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, usage
 			dWebSearchQuota = decimal.NewFromFloat(webSearchPrice).
 				Mul(decimal.NewFromInt(int64(webSearchTool.CallCount))).
 				Div(decimal.NewFromInt(1000)).Mul(dGroupRatio).Mul(dQuotaPerUnit)
-			extraContent += fmt.Sprintf("Web Search 调用 %d 次，上下文大小 %s，调用花费 %s",
-				webSearchTool.CallCount, webSearchTool.SearchContextSize, dWebSearchQuota.String())
+			extraContent = append(extraContent, fmt.Sprintf("Web Search 调用 %d 次，上下文大小 %s，调用花费 %s",
+				webSearchTool.CallCount, webSearchTool.SearchContextSize, dWebSearchQuota.String()))
 		}
 	} else if strings.HasSuffix(modelName, "search-preview") {
 		// search-preview 模型不支持 response api
@@ -258,8 +297,8 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, usage
 		webSearchPrice = operation_setting.GetWebSearchPricePerThousand(modelName, searchContextSize)
 		dWebSearchQuota = decimal.NewFromFloat(webSearchPrice).
 			Div(decimal.NewFromInt(1000)).Mul(dGroupRatio).Mul(dQuotaPerUnit)
-		extraContent += fmt.Sprintf("Web Search 调用 1 次，上下文大小 %s，调用花费 %s",
-			searchContextSize, dWebSearchQuota.String())
+		extraContent = append(extraContent, fmt.Sprintf("Web Search 调用 1 次，上下文大小 %s，调用花费 %s",
+			searchContextSize, dWebSearchQuota.String()))
 	}
 	// claude web search tool 计费
 	var dClaudeWebSearchQuota decimal.Decimal
@@ -269,8 +308,8 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, usage
 		claudeWebSearchPrice = operation_setting.GetClaudeWebSearchPricePerThousand()
 		dClaudeWebSearchQuota = decimal.NewFromFloat(claudeWebSearchPrice).
 			Div(decimal.NewFromInt(1000)).Mul(dGroupRatio).Mul(dQuotaPerUnit).Mul(decimal.NewFromInt(int64(claudeWebSearchCallCount)))
-		extraContent += fmt.Sprintf("Claude Web Search 调用 %d 次，调用花费 %s",
-			claudeWebSearchCallCount, dClaudeWebSearchQuota.String())
+		extraContent = append(extraContent, fmt.Sprintf("Claude Web Search 调用 %d 次，调用花费 %s",
+			claudeWebSearchCallCount, dClaudeWebSearchQuota.String()))
 	}
 	// file search tool 计费
 	var dFileSearchQuota decimal.Decimal
@@ -281,8 +320,8 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, usage
 			dFileSearchQuota = decimal.NewFromFloat(fileSearchPrice).
 				Mul(decimal.NewFromInt(int64(fileSearchTool.CallCount))).
 				Div(decimal.NewFromInt(1000)).Mul(dGroupRatio).Mul(dQuotaPerUnit)
-			extraContent += fmt.Sprintf("File Search 调用 %d 次，调用花费 %s",
-				fileSearchTool.CallCount, dFileSearchQuota.String())
+			extraContent = append(extraContent, fmt.Sprintf("File Search 调用 %d 次，调用花费 %s",
+				fileSearchTool.CallCount, dFileSearchQuota.String()))
 		}
 	}
 	var dImageGenerationCallQuota decimal.Decimal
@@ -290,13 +329,14 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, usage
 	if ctx.GetBool("image_generation_call") {
 		imageGenerationCallPrice = operation_setting.GetGPTImage1PriceOnceCall(ctx.GetString("image_generation_call_quality"), ctx.GetString("image_generation_call_size"))
 		dImageGenerationCallQuota = decimal.NewFromFloat(imageGenerationCallPrice).Mul(dGroupRatio).Mul(dQuotaPerUnit)
-		extraContent += fmt.Sprintf("Image Generation Call 花费 %s", dImageGenerationCallQuota.String())
+		extraContent = append(extraContent, fmt.Sprintf("Image Generation Call 花费 %s", dImageGenerationCallQuota.String()))
 	}
 
 	var quotaCalculateDecimal decimal.Decimal
 
 	var audioInputQuota decimal.Decimal
 	var audioInputPrice float64
+	isClaudeUsageSemantic := relayInfo.ChannelType == constant.ChannelTypeAnthropic
 	if !relayInfo.PriceData.UsePrice {
 		baseTokens := dPromptTokens
 		// 减去 cached tokens
@@ -304,14 +344,14 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, usage
 		// OpenAI/OpenRouter 等 API 的 prompt_tokens 包含缓存 tokens，需要减去
 		var cachedTokensWithRatio decimal.Decimal
 		if !dCacheTokens.IsZero() {
-			if relayInfo.ChannelType != constant.ChannelTypeAnthropic {
+			if !isClaudeUsageSemantic {
 				baseTokens = baseTokens.Sub(dCacheTokens)
 			}
 			cachedTokensWithRatio = dCacheTokens.Mul(dCacheRatio)
 		}
 		var dCachedCreationTokensWithRatio decimal.Decimal
 		if !dCachedCreationTokens.IsZero() {
-			if relayInfo.ChannelType != constant.ChannelTypeAnthropic {
+			if !isClaudeUsageSemantic {
 				baseTokens = baseTokens.Sub(dCachedCreationTokens)
 			}
 			dCachedCreationTokensWithRatio = dCachedCreationTokens.Mul(dCachedCreationRatio)
@@ -331,7 +371,7 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, usage
 				// 重新计算 base tokens
 				baseTokens = baseTokens.Sub(dAudioTokens)
 				audioInputQuota = decimal.NewFromFloat(audioInputPrice).Div(decimal.NewFromInt(1000000)).Mul(dAudioTokens).Mul(dGroupRatio).Mul(dQuotaPerUnit)
-				extraContent += fmt.Sprintf("Audio Input 花费 %s", audioInputQuota.String())
+				extraContent = append(extraContent, fmt.Sprintf("Audio Input 花费 %s", audioInputQuota.String()))
 			}
 		}
 		promptQuota := baseTokens.Add(cachedTokensWithRatio).
@@ -356,17 +396,25 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, usage
 	// 添加 image generation call 计费
 	quotaCalculateDecimal = quotaCalculateDecimal.Add(dImageGenerationCallQuota)
 
+	if len(relayInfo.PriceData.OtherRatios) > 0 {
+		for key, otherRatio := range relayInfo.PriceData.OtherRatios {
+			dOtherRatio := decimal.NewFromFloat(otherRatio)
+			quotaCalculateDecimal = quotaCalculateDecimal.Mul(dOtherRatio)
+			extraContent = append(extraContent, fmt.Sprintf("其他倍率 %s: %f", key, otherRatio))
+		}
+	}
+
 	quota := int(quotaCalculateDecimal.Round(0).IntPart())
 	totalTokens := promptTokens + completionTokens
 
-	var logContent string
+	//var logContent string
 
 	// record all the consume log even if quota is 0
 	if totalTokens == 0 {
 		// in this case, must be some error happened
 		// we cannot just return, because we may have to return the pre-consumed quota
 		quota = 0
-		logContent += fmt.Sprintf("（可能是上游超时）")
+		extraContent = append(extraContent, "上游没有返回计费信息，无法扣费（可能是上游超时）")
 		logger.LogError(ctx, fmt.Sprintf("total tokens is 0, cannot consume quota, userId %d, channelId %d, "+
 			"tokenId %d, model %s， pre-consumed quota %d", relayInfo.UserId, relayInfo.ChannelId, relayInfo.TokenId, modelName, relayInfo.FinalPreConsumedQuota))
 	} else {
@@ -405,16 +453,19 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, usage
 	logModel := modelName
 	if strings.HasPrefix(logModel, "gpt-4-gizmo") {
 		logModel = "gpt-4-gizmo-*"
-		logContent += fmt.Sprintf("，模型 %s", modelName)
+		extraContent = append(extraContent, fmt.Sprintf("模型 %s", modelName))
 	}
 	if strings.HasPrefix(logModel, "gpt-4o-gizmo") {
 		logModel = "gpt-4o-gizmo-*"
-		logContent += fmt.Sprintf("，模型 %s", modelName)
-	}
-	if extraContent != "" {
-		logContent += ", " + extraContent
+		extraContent = append(extraContent, fmt.Sprintf("模型 %s", modelName))
 	}
+	logContent := strings.Join(extraContent, ", ")
 	other := service.GenerateTextOtherInfo(ctx, relayInfo, modelRatio, groupRatio, completionRatio, cacheTokens, cacheRatio, modelPrice, relayInfo.PriceData.GroupRatioInfo.GroupSpecialRatio)
+	// For chat-based calls to the Claude model, tagging is required. Using Claude's rendering logs, the two approaches handle input rendering differently.
+	if isClaudeUsageSemantic {
+		other["claude"] = true
+		other["usage_semantic"] = "anthropic"
+	}
 	if imageTokens != 0 {
 		other["image"] = true
 		other["image_ratio"] = imageRatio

relay/embedding_handler.go

@@ -45,6 +45,7 @@ func EmbeddingHelper(c *gin.Context, info *relaycommon.RelayInfo) (newAPIError *
 	if err != nil {
 		return types.NewError(err, types.ErrorCodeConvertRequestFailed, types.ErrOptionWithSkipRetry())
 	}
+	relaycommon.AppendRequestConversionFromRequest(info, convertedRequest)
 	jsonData, err := json.Marshal(convertedRequest)
 	if err != nil {
 		return types.NewError(err, types.ErrorCodeConvertRequestFailed, types.ErrOptionWithSkipRetry())
@@ -82,6 +83,6 @@ func EmbeddingHelper(c *gin.Context, info *relaycommon.RelayInfo) (newAPIError *
 		service.ResetStatusCode(newAPIError, statusCodeMappingStr)
 		return newAPIError
 	}
-	postConsumeQuota(c, info, usage.(*dto.Usage), "")
+	postConsumeQuota(c, info, usage.(*dto.Usage))
 	return nil
 }

relay/gemini_handler.go

@@ -149,6 +149,7 @@ func GeminiHelper(c *gin.Context, info *relaycommon.RelayInfo) (newAPIError *typ
 		if err != nil {
 			return types.NewError(err, types.ErrorCodeConvertRequestFailed, types.ErrOptionWithSkipRetry())
 		}
+		relaycommon.AppendRequestConversionFromRequest(info, convertedRequest)
 		jsonData, err := common.Marshal(convertedRequest)
 		if err != nil {
 			return types.NewError(err, types.ErrorCodeConvertRequestFailed, types.ErrOptionWithSkipRetry())
@@ -193,7 +194,7 @@ func GeminiHelper(c *gin.Context, info *relaycommon.RelayInfo) (newAPIError *typ
 		return openaiErr
 	}
 
-	postConsumeQuota(c, info, usage.(*dto.Usage), "")
+	postConsumeQuota(c, info, usage.(*dto.Usage))
 	return nil
 }
 
@@ -292,6 +293,6 @@ func GeminiEmbeddingHandler(c *gin.Context, info *relaycommon.RelayInfo) (newAPI
 		return openaiErr
 	}
 
-	postConsumeQuota(c, info, usage.(*dto.Usage), "")
+	postConsumeQuota(c, info, usage.(*dto.Usage))
 	return nil
 }

relay/image_handler.go

@@ -57,6 +57,7 @@ func ImageHelper(c *gin.Context, info *relaycommon.RelayInfo) (newAPIError *type
 		if err != nil {
 			return types.NewError(err, types.ErrorCodeConvertRequestFailed)
 		}
+		relaycommon.AppendRequestConversionFromRequest(info, convertedRequest)
 
 		switch convertedRequest.(type) {
 		case *bytes.Buffer:
@@ -124,12 +125,18 @@ func ImageHelper(c *gin.Context, info *relaycommon.RelayInfo) (newAPIError *type
 		quality = "hd"
 	}
 
-	var logContent string
+	var logContent []string
 
 	if len(request.Size) > 0 {
-		logContent = fmt.Sprintf("大小 %s, 品质 %s, 张数 %d", request.Size, quality, request.N)
+		logContent = append(logContent, fmt.Sprintf("大小 %s", request.Size))
+	}
+	if len(quality) > 0 {
+		logContent = append(logContent, fmt.Sprintf("品质 %s", quality))
+	}
+	if request.N > 0 {
+		logContent = append(logContent, fmt.Sprintf("生成数量 %d", request.N))
 	}
 
-	postConsumeQuota(c, info, usage.(*dto.Usage), logContent)
+	postConsumeQuota(c, info, usage.(*dto.Usage), logContent...)
 	return nil
 }

relay/relay_adaptor.go

@@ -11,6 +11,7 @@ import (
 	"github.com/QuantumNous/new-api/relay/channel/baidu_v2"
 	"github.com/QuantumNous/new-api/relay/channel/claude"
 	"github.com/QuantumNous/new-api/relay/channel/cloudflare"
+	"github.com/QuantumNous/new-api/relay/channel/codex"
 	"github.com/QuantumNous/new-api/relay/channel/cohere"
 	"github.com/QuantumNous/new-api/relay/channel/coze"
 	"github.com/QuantumNous/new-api/relay/channel/deepseek"
@@ -117,6 +118,8 @@ func GetAdaptor(apiType int) channel.Adaptor {
 		return &minimax.Adaptor{}
 	case constant.APITypeReplicate:
 		return &replicate.Adaptor{}
+	case constant.APITypeCodex:
+		return &codex.Adaptor{}
 	}
 	return nil
 }
@@ -148,7 +151,7 @@ func GetTaskAdaptor(platform constant.TaskPlatform) channel.TaskAdaptor {
 			return &taskvertex.TaskAdaptor{}
 		case constant.ChannelTypeVidu:
 			return &taskVidu.TaskAdaptor{}
-		case constant.ChannelTypeDoubaoVideo:
+		case constant.ChannelTypeDoubaoVideo, constant.ChannelTypeVolcEngine:
 			return &taskdoubao.TaskAdaptor{}
 		case constant.ChannelTypeSora, constant.ChannelTypeOpenAI:
 			return &tasksora.TaskAdaptor{}

relay/relay_task.go

@@ -150,6 +150,14 @@ func RelayTaskSubmit(c *gin.Context, info *relaycommon.RelayInfo) (taskErr *dto.
 		}
 	}
 
+	// 处理 auto 分组：从 context 获取实际选中的分组
+	// 当使用 auto 分组时，Distribute 中间件会将实际选中的分组存储在 ContextKeyAutoGroup 中
+	if autoGroup, exists := common.GetContextKey(c, constant.ContextKeyAutoGroup); exists {
+		if groupStr, ok := autoGroup.(string); ok && groupStr != "" {
+			info.UsingGroup = groupStr
+		}
+	}
+
 	// 预扣
 	groupRatio := ratio_setting.GetGroupRatio(info.UsingGroup)
 	var ratio float64

relay/rerank_handler.go

@@ -53,6 +53,7 @@ func RerankHelper(c *gin.Context, info *relaycommon.RelayInfo) (newAPIError *typ
 		if err != nil {
 			return types.NewError(err, types.ErrorCodeConvertRequestFailed, types.ErrOptionWithSkipRetry())
 		}
+		relaycommon.AppendRequestConversionFromRequest(info, convertedRequest)
 		jsonData, err := common.Marshal(convertedRequest)
 		if err != nil {
 			return types.NewError(err, types.ErrorCodeConvertRequestFailed, types.ErrOptionWithSkipRetry())
@@ -95,6 +96,6 @@ func RerankHelper(c *gin.Context, info *relaycommon.RelayInfo) (newAPIError *typ
 		service.ResetStatusCode(newAPIError, statusCodeMappingStr)
 		return newAPIError
 	}
-	postConsumeQuota(c, info, usage.(*dto.Usage), "")
+	postConsumeQuota(c, info, usage.(*dto.Usage))
 	return nil
 }

relay/responses_handler.go

@@ -53,6 +53,7 @@ func ResponsesHelper(c *gin.Context, info *relaycommon.RelayInfo) (newAPIError *
 		if err != nil {
 			return types.NewError(err, types.ErrorCodeConvertRequestFailed, types.ErrOptionWithSkipRetry())
 		}
+		relaycommon.AppendRequestConversionFromRequest(info, convertedRequest)
 		jsonData, err := common.Marshal(convertedRequest)
 		if err != nil {
 			return types.NewError(err, types.ErrorCodeConvertRequestFailed, types.ErrOptionWithSkipRetry())
@@ -107,7 +108,7 @@ func ResponsesHelper(c *gin.Context, info *relaycommon.RelayInfo) (newAPIError *
 	if strings.HasPrefix(info.OriginModelName, "gpt-4o-audio") {
 		service.PostAudioConsumeQuota(c, info, usage.(*dto.Usage), "")
 	} else {
-		postConsumeQuota(c, info, usage.(*dto.Usage), "")
+		postConsumeQuota(c, info, usage.(*dto.Usage))
 	}
 	return nil
 }