diff --git a/backend/internal/service/gateway_service.go b/backend/internal/service/gateway_service.go index 0b50162a..4544ec82 100644 --- a/backend/internal/service/gateway_service.go +++ b/backend/internal/service/gateway_service.go @@ -491,6 +491,7 @@ type ForwardResult struct { RequestID string Usage ClaudeUsage Model string + UpstreamModel string // Actual upstream model after mapping (empty = no mapping) Stream bool Duration time.Duration FirstTokenMs *int // 首字时间（流式请求） @@ -3989,7 +3990,13 @@ func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *A passthroughModel = mappedModel } } - return s.forwardAnthropicAPIKeyPassthrough(ctx, c, account, passthroughBody, passthroughModel, parsed.Stream, startTime) + return s.forwardAnthropicAPIKeyPassthroughWithInput(ctx, c, account, anthropicPassthroughForwardInput{ + Body: passthroughBody, + RequestModel: passthroughModel, + OriginalModel: parsed.Model, + RequestStream: parsed.Stream, + StartTime: startTime, + }) } if account != nil && account.IsBedrock() { @@ -4513,6 +4520,7 @@ func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *A RequestID: resp.Header.Get("x-request-id"), Usage: *usage, Model: originalModel, // 使用原始模型用于计费和日志 + UpstreamModel: mappedModel, Stream: reqStream, Duration: time.Since(startTime), FirstTokenMs: firstTokenMs, @@ -4520,14 +4528,38 @@ func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *A }, nil } +type anthropicPassthroughForwardInput struct { + Body []byte + RequestModel string + OriginalModel string + RequestStream bool + StartTime time.Time +} + func (s *GatewayService) forwardAnthropicAPIKeyPassthrough( ctx context.Context, c *gin.Context, account *Account, body []byte, reqModel string, + originalModel string, reqStream bool, startTime time.Time, +) (*ForwardResult, error) { + return s.forwardAnthropicAPIKeyPassthroughWithInput(ctx, c, account, anthropicPassthroughForwardInput{ + Body: body, + RequestModel: reqModel, + OriginalModel: originalModel, + RequestStream: reqStream, + StartTime: startTime, + }) +} + +func (s *GatewayService) forwardAnthropicAPIKeyPassthroughWithInput( + ctx context.Context, + c *gin.Context, + account *Account, + input anthropicPassthroughForwardInput, ) (*ForwardResult, error) { token, tokenType, err := s.GetAccessToken(ctx, account) if err != nil { @@ -4543,19 +4575,19 @@ func (s *GatewayService) forwardAnthropicAPIKeyPassthrough( } logger.LegacyPrintf("service.gateway", "[Anthropic 自动透传] 命中 API Key 透传分支: account=%d name=%s model=%s stream=%v", - account.ID, account.Name, reqModel, reqStream) + account.ID, account.Name, input.RequestModel, input.RequestStream) if c != nil { c.Set("anthropic_passthrough", true) } // 重试间复用同一请求体，避免每次 string(body) 产生额外分配。 - setOpsUpstreamRequestBody(c, body) + setOpsUpstreamRequestBody(c, input.Body) var resp *http.Response retryStart := time.Now() for attempt := 1; attempt <= maxRetryAttempts; attempt++ { - upstreamCtx, releaseUpstreamCtx := detachStreamUpstreamContext(ctx, reqStream) - upstreamReq, err := s.buildUpstreamRequestAnthropicAPIKeyPassthrough(upstreamCtx, c, account, body, token) + upstreamCtx, releaseUpstreamCtx := detachStreamUpstreamContext(ctx, input.RequestStream) + upstreamReq, err := s.buildUpstreamRequestAnthropicAPIKeyPassthrough(upstreamCtx, c, account, input.Body, token) releaseUpstreamCtx() if err != nil { return nil, err @@ -4713,8 +4745,8 @@ func (s *GatewayService) forwardAnthropicAPIKeyPassthrough( var usage *ClaudeUsage var firstTokenMs *int var clientDisconnect bool - if reqStream { - streamResult, err := s.handleStreamingResponseAnthropicAPIKeyPassthrough(ctx, resp, c, account, startTime, reqModel) + if input.RequestStream { + streamResult, err := s.handleStreamingResponseAnthropicAPIKeyPassthrough(ctx, resp, c, account, input.StartTime, input.RequestModel) if err != nil { return nil, err } @@ -4734,9 +4766,10 @@ func (s *GatewayService) forwardAnthropicAPIKeyPassthrough( return &ForwardResult{ RequestID: resp.Header.Get("x-request-id"), Usage: *usage, - Model: reqModel, - Stream: reqStream, - Duration: time.Since(startTime), + Model: input.OriginalModel, + UpstreamModel: input.RequestModel, + Stream: input.RequestStream, + Duration: time.Since(input.StartTime), FirstTokenMs: firstTokenMs, ClientDisconnect: clientDisconnect, }, nil @@ -5241,6 +5274,7 @@ func (s *GatewayService) forwardBedrock( RequestID: resp.Header.Get("x-amzn-requestid"), Usage: *usage, Model: reqModel, + UpstreamModel: mappedModel, Stream: reqStream, Duration: time.Since(startTime), FirstTokenMs: firstTokenMs, @@ -7529,6 +7563,7 @@ func (s *GatewayService) RecordUsage(ctx context.Context, input *RecordUsageInpu AccountID: account.ID, RequestID: requestID, Model: result.Model, + UpstreamModel: optionalNonEqualStringPtr(result.UpstreamModel, result.Model), ReasoningEffort: result.ReasoningEffort, InboundEndpoint: optionalTrimmedStringPtr(input.InboundEndpoint), UpstreamEndpoint: optionalTrimmedStringPtr(input.UpstreamEndpoint), @@ -7710,6 +7745,7 @@ func (s *GatewayService) RecordUsageWithLongContext(ctx context.Context, input * AccountID: account.ID, RequestID: requestID, Model: result.Model, + UpstreamModel: optionalNonEqualStringPtr(result.UpstreamModel, result.Model), ReasoningEffort: result.ReasoningEffort, InboundEndpoint: optionalTrimmedStringPtr(input.InboundEndpoint), UpstreamEndpoint: optionalTrimmedStringPtr(input.UpstreamEndpoint), diff --git a/backend/internal/service/openai_gateway_chat_completions.go b/backend/internal/service/openai_gateway_chat_completions.go index 9529f6be..7202f7cb 100644 --- a/backend/internal/service/openai_gateway_chat_completions.go +++ b/backend/internal/service/openai_gateway_chat_completions.go @@ -277,12 +277,13 @@ func (s *OpenAIGatewayService) handleChatBufferedStreamingResponse( c.JSON(http.StatusOK, chatResp) return &OpenAIForwardResult{ - RequestID: requestID, - Usage: usage, - Model: originalModel, - BillingModel: mappedModel, - Stream: false, - Duration: time.Since(startTime), + RequestID: requestID, + Usage: usage, + Model: originalModel, + BillingModel: mappedModel, + UpstreamModel: mappedModel, + Stream: false, + Duration: time.Since(startTime), }, nil } @@ -324,13 +325,14 @@ func (s *OpenAIGatewayService) handleChatStreamingResponse( resultWithUsage := func() *OpenAIForwardResult { return &OpenAIForwardResult{ - RequestID: requestID, - Usage: usage, - Model: originalModel, - BillingModel: mappedModel, - Stream: true, - Duration: time.Since(startTime), - FirstTokenMs: firstTokenMs, + RequestID: requestID, + Usage: usage, + Model: originalModel, + BillingModel: mappedModel, + UpstreamModel: mappedModel, + Stream: true, + Duration: time.Since(startTime), + FirstTokenMs: firstTokenMs, } } diff --git a/backend/internal/service/openai_gateway_messages.go b/backend/internal/service/openai_gateway_messages.go index 58714571..6a29823a 100644 --- a/backend/internal/service/openai_gateway_messages.go +++ b/backend/internal/service/openai_gateway_messages.go @@ -299,12 +299,13 @@ func (s *OpenAIGatewayService) handleAnthropicBufferedStreamingResponse( c.JSON(http.StatusOK, anthropicResp) return &OpenAIForwardResult{ - RequestID: requestID, - Usage: usage, - Model: originalModel, - BillingModel: mappedModel, - Stream: false, - Duration: time.Since(startTime), + RequestID: requestID, + Usage: usage, + Model: originalModel, + BillingModel: mappedModel, + UpstreamModel: mappedModel, + Stream: false, + Duration: time.Since(startTime), }, nil } @@ -347,13 +348,14 @@ func (s *OpenAIGatewayService) handleAnthropicStreamingResponse( // resultWithUsage builds the final result snapshot. resultWithUsage := func() *OpenAIForwardResult { return &OpenAIForwardResult{ - RequestID: requestID, - Usage: usage, - Model: originalModel, - BillingModel: mappedModel, - Stream: true, - Duration: time.Since(startTime), - FirstTokenMs: firstTokenMs, + RequestID: requestID, + Usage: usage, + Model: originalModel, + BillingModel: mappedModel, + UpstreamModel: mappedModel, + Stream: true, + Duration: time.Since(startTime), + FirstTokenMs: firstTokenMs, } } diff --git a/backend/internal/service/openai_gateway_service.go b/backend/internal/service/openai_gateway_service.go index c8876edb..cf902c20 100644 --- a/backend/internal/service/openai_gateway_service.go +++ b/backend/internal/service/openai_gateway_service.go @@ -216,6 +216,9 @@ type OpenAIForwardResult struct { // This is set by the Anthropic Messages conversion path where // the mapped upstream model differs from the client-facing model. BillingModel string + // UpstreamModel is the actual model sent to the upstream provider after mapping. + // Empty when no mapping was applied (requested model was used as-is). + UpstreamModel string // ServiceTier records the OpenAI Responses API service tier, e.g. "priority" / "flex". // Nil means the request did not specify a recognized tier. ServiceTier *string @@ -2128,6 +2131,7 @@ func (s *OpenAIGatewayService) Forward(ctx context.Context, c *gin.Context, acco firstTokenMs, wsAttempts, ) + wsResult.UpstreamModel = mappedModel return wsResult, nil } s.writeOpenAIWSFallbackErrorResponse(c, account, wsErr) @@ -2263,6 +2267,7 @@ func (s *OpenAIGatewayService) Forward(ctx context.Context, c *gin.Context, acco RequestID: resp.Header.Get("x-request-id"), Usage: *usage, Model: originalModel, + UpstreamModel: mappedModel, ServiceTier: serviceTier, ReasoningEffort: reasoningEffort, Stream: reqStream, @@ -4134,7 +4139,8 @@ func (s *OpenAIGatewayService) RecordUsage(ctx context.Context, input *OpenAIRec APIKeyID: apiKey.ID, AccountID: account.ID, RequestID: requestID, - Model: billingModel, + Model: result.Model, + UpstreamModel: optionalNonEqualStringPtr(result.UpstreamModel, result.Model), ServiceTier: result.ServiceTier, ReasoningEffort: result.ReasoningEffort, InboundEndpoint: optionalTrimmedStringPtr(input.InboundEndpoint), @@ -4700,11 +4706,3 @@ func normalizeOpenAIReasoningEffort(raw string) string { return "" } } - -func optionalTrimmedStringPtr(raw string) *string { - trimmed := strings.TrimSpace(raw) - if trimmed == "" { - return nil - } - return &trimmed -} diff --git a/backend/internal/service/usage_log.go b/backend/internal/service/usage_log.go index 7f1bef7f..5a498676 100644 --- a/backend/internal/service/usage_log.go +++ b/backend/internal/service/usage_log.go @@ -98,6 +98,9 @@ type UsageLog struct { AccountID int64 RequestID string Model string + // UpstreamModel is the actual model sent to the upstream provider after mapping. + // Nil means no mapping was applied (requested model was used as-is). + UpstreamModel *string // ServiceTier records the OpenAI service tier used for billing, e.g. "priority" / "flex". ServiceTier *string // ReasoningEffort is the request's reasoning effort level. diff --git a/backend/internal/service/usage_log_helpers.go b/backend/internal/service/usage_log_helpers.go new file mode 100644 index 00000000..2ab51849 --- /dev/null +++ b/backend/internal/service/usage_log_helpers.go @@ -0,0 +1,21 @@ +package service + +import "strings" + +func optionalTrimmedStringPtr(raw string) *string { + trimmed := strings.TrimSpace(raw) + if trimmed == "" { + return nil + } + return &trimmed +} + +// optionalNonEqualStringPtr returns a pointer to value if it is non-empty and +// differs from compare; otherwise nil. Used to store upstream_model only when +// it differs from the requested model. +func optionalNonEqualStringPtr(value, compare string) *string { + if value == "" || value == compare { + return nil + } + return &value +}