feat: sync deployed v1.0.0-rc.38-custom.42

Sync production source including routing, billing, Astra detection and public uptime. Update custom build instructions and license packaging.

relaykit/relayconvert/internal/claude_messages/to_oai_chat_req.go

@@ -1,6 +1,7 @@
 package claudemessages
 
 import (
+	"context"
 	"fmt"
 	"strings"
 
@@ -23,7 +24,7 @@ type openRouterRequestReasoning struct {
 	Exclude   bool   `json:"exclude,omitempty"`
 }
 
-func ClaudeMessagesRequestToOpenAIChat(claudeRequest dto.ClaudeRequest, info convmeta.Meta) (*dto.GeneralOpenAIRequest, error) {
+func ClaudeMessagesRequestToOpenAIChat(ctx context.Context, claudeRequest dto.ClaudeRequest, info convmeta.Meta) (*dto.GeneralOpenAIRequest, error) {
 	openAIRequest := dto.GeneralOpenAIRequest{
 		Model:       claudeRequest.Model,
 		Temperature: claudeRequest.Temperature,
@@ -40,7 +41,7 @@ func ClaudeMessagesRequestToOpenAIChat(claudeRequest dto.ClaudeRequest, info con
 	if claudeRequest.Stream != nil {
 		openAIRequest.Stream = kitutil.GetPointer(*claudeRequest.Stream)
 	}
-	reasoningIntent, effectiveEffort, err := claudeRequestReasoningIntent(&claudeRequest, info)
+	reasoningIntent, effectiveEffort, err := claudeRequestReasoningIntent(ctx, &claudeRequest, info)
 	if err != nil {
 		return nil, reasoning.AsClientError(err)
 	}

relaykit/relayconvert/internal/claude_messages/to_oai_responses_req.go

@@ -1,6 +1,7 @@
 package claudemessages
 
 import (
+	"context"
 	"encoding/json"
 	"errors"
 	"fmt"
@@ -8,11 +9,12 @@ import (
 
 	"github.com/QuantumNous/new-api/relaykit/dto"
 	"github.com/QuantumNous/new-api/relaykit/relayconvert/convmeta"
+	"github.com/QuantumNous/new-api/relaykit/relayconvert/internal/convdiag"
 	kitutil "github.com/QuantumNous/new-api/relaykit/relayconvert/kitutil"
 	"github.com/QuantumNous/new-api/relaykit/relayconvert/reasoning"
 )
 
-func ClaudeMessagesRequestToOpenAIResponses(claudeRequest dto.ClaudeRequest, info convmeta.Meta) (*dto.OpenAIResponsesRequest, error) {
+func ClaudeMessagesRequestToOpenAIResponses(ctx context.Context, claudeRequest dto.ClaudeRequest, info convmeta.Meta) (*dto.OpenAIResponsesRequest, error) {
 	if strings.TrimSpace(claudeRequest.Model) == "" {
 		return nil, errors.New("model is required")
 	}
@@ -67,7 +69,7 @@ func ClaudeMessagesRequestToOpenAIResponses(claudeRequest dto.ClaudeRequest, inf
 		responsesRequest.MaxOutputTokens = &maxOutputTokens
 	}
 
-	reasoningIntent, effectiveEffort, err := claudeRequestReasoningIntent(&claudeRequest, info)
+	reasoningIntent, effectiveEffort, err := claudeRequestReasoningIntent(ctx, &claudeRequest, info)
 	if err != nil {
 		return nil, reasoning.AsClientError(err)
 	}
@@ -81,20 +83,22 @@ func ClaudeMessagesRequestToOpenAIResponses(claudeRequest dto.ClaudeRequest, inf
 	return responsesRequest, nil
 }
 
-func claudeRequestReasoningIntent(claudeRequest *dto.ClaudeRequest, info convmeta.Meta) (reasoning.Intent, reasoning.Effort, error) {
-	reasoningIntent, err := reasoning.FromClaude(claudeRequest)
+func claudeRequestReasoningIntent(ctx context.Context, claudeRequest *dto.ClaudeRequest, info convmeta.Meta) (reasoning.Intent, reasoning.Effort, error) {
+	reasoningIntent, diagnostics, err := reasoning.FromClaude(claudeRequest)
 	if err != nil {
 		return reasoning.Intent{}, "", err
 	}
+	convdiag.Add(ctx, diagnostics...)
 	sourceModel := claudeRequest.Model
 	if info != nil && info.GetOriginModelName() != "" {
 		sourceModel = info.GetOriginModelName()
 	}
 	if suffix := reasoning.IntentFromState(convmeta.ReasoningStateOf(info)); !suffix.IsEmpty() {
-		reasoningIntent, err = reasoning.MergeExplicitAndSuffix(reasoningIntent, suffix, sourceModel)
+		reasoningIntent, diagnostics, err = reasoning.MergeExplicitAndSuffix(reasoningIntent, suffix, sourceModel)
 		if err != nil {
 			return reasoning.Intent{}, "", err
 		}
+		convdiag.Add(ctx, diagnostics...)
 	}
 	reasoningIntent = reasoning.ResolveClaudeDefault(sourceModel, reasoningIntent)
 	return reasoningIntent, reasoning.EffectiveEffort(reasoningIntent), nil

relaykit/relayconvert/internal/gemini_chat/to_oai_chat_req.go

@@ -1,17 +1,19 @@
 package geminichat
 
 import (
+	"context"
 	"fmt"
 	"strings"
 
 	"github.com/QuantumNous/new-api/relaykit/dto"
 	"github.com/QuantumNous/new-api/relaykit/relayconvert/convmeta"
+	"github.com/QuantumNous/new-api/relaykit/relayconvert/internal/convdiag"
 	"github.com/QuantumNous/new-api/relaykit/relayconvert/internal/jsonutil"
 	kitutil "github.com/QuantumNous/new-api/relaykit/relayconvert/kitutil"
 	"github.com/QuantumNous/new-api/relaykit/relayconvert/reasoning"
 )
 
-func GeminiGenerateContentRequestToOpenAIChat(geminiRequest *dto.GeminiChatRequest, info convmeta.Meta) (*dto.GeneralOpenAIRequest, error) {
+func GeminiGenerateContentRequestToOpenAIChat(ctx context.Context, geminiRequest *dto.GeminiChatRequest, info convmeta.Meta) (*dto.GeneralOpenAIRequest, error) {
 	modelName := ""
 	isStream := false
 	if info != nil {
@@ -22,30 +24,42 @@ func GeminiGenerateContentRequestToOpenAIChat(geminiRequest *dto.GeminiChatReque
 		Model:  modelName,
 		Stream: kitutil.GetPointer(isStream),
 	}
-	reasoningIntent, err := reasoning.FromGemini(geminiRequest)
-	if err != nil {
-		return nil, reasoning.AsClientError(err)
-	}
 	sourceModelName := modelName
 	if info != nil && info.GetOriginModelName() != "" {
 		sourceModelName = info.GetOriginModelName()
 	}
-	baseSourceModel := sourceModelName
 	opts := convmeta.OptionsOf(info)
 	preserveSuffix := opts.ShouldPreserveThinkingSuffix(sourceModelName)
+	// Capability lookups need the unsuffixed Gemini model the client addressed.
+	// The host already folded any alias into ReasoningState; mirror its parser
+	// so gemini-pro-latest-thinking still resolves to gemini-pro-latest.
+	baseSourceModel := sourceModelName
 	if !preserveSuffix {
-		if suffix := reasoning.IntentFromState(convmeta.ReasoningStateOf(info)); !suffix.IsEmpty() {
-			reasoningIntent, err = reasoning.MergeExplicitAndSuffix(reasoningIntent, suffix, sourceModelName)
-			if err != nil {
-				return nil, reasoning.AsClientError(err)
-			}
+		baseSourceModel = reasoning.ParseModelModifiers(sourceModelName).Base
+		if base, _, found, err := reasoning.ParseGeminiModelSuffix(baseSourceModel, opts.Gemini.ThinkingAdapterEnabled); err == nil && found {
+			baseSourceModel = base
 		}
 	}
 	if baseSourceModel != "" && geminiRequest.GenerationConfig.ThinkingConfig != nil {
-		_, err = reasoning.ValidateGeminiThinkingConfig(baseSourceModel, geminiRequest.GenerationConfig.ThinkingConfig)
+		_, diagnostics, err := reasoning.NormalizeGeminiThinkingConfig(baseSourceModel, &geminiRequest.GenerationConfig)
 		if err != nil {
 			return nil, reasoning.AsClientError(err)
 		}
+		convdiag.Add(ctx, diagnostics...)
+	}
+	reasoningIntent, diagnostics, err := reasoning.FromGemini(geminiRequest)
+	if err != nil {
+		return nil, reasoning.AsClientError(err)
+	}
+	convdiag.Add(ctx, diagnostics...)
+	if !preserveSuffix {
+		if suffix := reasoning.IntentFromState(convmeta.ReasoningStateOf(info)); !suffix.IsEmpty() {
+			reasoningIntent, diagnostics, err = reasoning.MergeExplicitAndSuffix(reasoningIntent, suffix, sourceModelName)
+			if err != nil {
+				return nil, reasoning.AsClientError(err)
+			}
+			convdiag.Add(ctx, diagnostics...)
+		}
 	}
 	reasoningIntent = reasoning.ResolveGeminiDefault(baseSourceModel, reasoningIntent)
 	effectiveEffort := reasoning.EffectiveEffort(reasoningIntent)

relaykit/relayconvert/internal/oai_chat/to_claude_messages_req.go

@@ -8,6 +8,7 @@ import (
 
 	"github.com/QuantumNous/new-api/relaykit/dto"
 	"github.com/QuantumNous/new-api/relaykit/relayconvert/convmeta"
+	"github.com/QuantumNous/new-api/relaykit/relayconvert/internal/convdiag"
 	relaymedia "github.com/QuantumNous/new-api/relaykit/relayconvert/internal/media"
 	sharedclaude "github.com/QuantumNous/new-api/relaykit/relayconvert/internal/shared/claude"
 	kitutil "github.com/QuantumNous/new-api/relaykit/relayconvert/kitutil"
@@ -94,10 +95,11 @@ func OpenAIChatRequestToClaudeMessages(c context.Context, info convmeta.Meta, te
 		}
 	}
 
-	sourceReasoning, err := reasoning.FromOpenAIChat(&textRequest)
+	sourceReasoning, diagnostics, err := reasoning.FromOpenAIChat(&textRequest)
 	if err != nil {
 		return nil, reasoning.AsClientError(err)
 	}
+	convdiag.Add(c, diagnostics...)
 	if err := sharedclaude.ApplyReasoning(c, &claudeRequest, info, sourceReasoning, true); err != nil {
 		return nil, reasoning.AsClientError(err)
 	}

relaykit/relayconvert/internal/oai_chat/to_gemini_chat_req.go

@@ -137,7 +137,7 @@ func OpenAIChatRequestToGeminiGenerateContent(c context.Context, textRequest dto
 		}
 	}
 
-	if err := sharedgemini.ApplyThinkingConfig(&geminiRequest, info, textRequest); err != nil {
+	if err := sharedgemini.ApplyThinkingConfig(c, &geminiRequest, info, textRequest); err != nil {
 		return nil, reasoning.AsClientError(err)
 	}
 

relaykit/relayconvert/internal/oai_chat/to_oai_responses_req.go

@@ -1,12 +1,14 @@
 package oaichat
 
 import (
+	"context"
 	"encoding/json"
 	"errors"
 	"fmt"
 	"strings"
 
 	"github.com/QuantumNous/new-api/relaykit/dto"
+	"github.com/QuantumNous/new-api/relaykit/relayconvert/internal/convdiag"
 	kitutil "github.com/QuantumNous/new-api/relaykit/relayconvert/kitutil"
 	"github.com/QuantumNous/new-api/relaykit/relayconvert/reasoning"
 	"github.com/samber/lo"
@@ -74,7 +76,7 @@ func convertChatResponseFormatToResponsesText(reqFormat *dto.ResponseFormat) jso
 	return textRaw
 }
 
-func ChatCompletionsRequestToResponsesRequest(req *dto.GeneralOpenAIRequest) (*dto.OpenAIResponsesRequest, error) {
+func ChatCompletionsRequestToResponsesRequest(ctx context.Context, req *dto.GeneralOpenAIRequest) (*dto.OpenAIResponsesRequest, error) {
 	if req == nil {
 		return nil, errors.New("request is nil")
 	}
@@ -412,10 +414,11 @@ func ChatCompletionsRequestToResponsesRequest(req *dto.GeneralOpenAIRequest) (*d
 		out.MaxOutputTokens = lo.ToPtr(maxOutputTokens)
 	}
 
-	reasoningIntent, err := reasoning.FromOpenAIChat(req)
+	reasoningIntent, diagnostics, err := reasoning.FromOpenAIChat(req)
 	if err != nil {
 		return nil, reasoning.AsClientError(err)
 	}
+	convdiag.Add(ctx, diagnostics...)
 	if err := reasoning.ApplyToOpenAIResponses(out, reasoningIntent); err != nil {
 		return nil, reasoning.AsClientError(err)
 	}

relaykit/relayconvert/internal/oai_chat/to_oai_responses_req_test.go

@@ -1,6 +1,7 @@
 package oaichat
 
 import (
+	"context"
 	"encoding/json"
 	"testing"
 
@@ -28,7 +29,7 @@ func TestChatCompletionsRequestToResponsesRequestInstructionsAndTools(t *testing
 		},
 	}
 
-	got, err := ChatCompletionsRequestToResponsesRequest(req)
+	got, err := ChatCompletionsRequestToResponsesRequest(context.Background(), req)
 	require.NoError(t, err)
 
 	assert.Equal(t, "gpt-test", got.Model)
@@ -42,7 +43,7 @@ func TestChatCompletionsRequestToResponsesRequestInstructionsAndTools(t *testing
 func TestChatCompletionsRequestToResponsesRequestPreservesPromptCacheKey(t *testing.T) {
 	t.Run("present", func(t *testing.T) {
 		key := "session-\"quoted\"\\path\n世界"
-		got, err := ChatCompletionsRequestToResponsesRequest(&dto.GeneralOpenAIRequest{
+		got, err := ChatCompletionsRequestToResponsesRequest(context.Background(), &dto.GeneralOpenAIRequest{
 			Model:          "gpt-test",
 			Messages:       []dto.Message{{Role: "user", Content: "hello"}},
 			PromptCacheKey: key,
@@ -59,7 +60,7 @@ func TestChatCompletionsRequestToResponsesRequestPreservesPromptCacheKey(t *test
 	})
 
 	t.Run("absent", func(t *testing.T) {
-		got, err := ChatCompletionsRequestToResponsesRequest(&dto.GeneralOpenAIRequest{
+		got, err := ChatCompletionsRequestToResponsesRequest(context.Background(), &dto.GeneralOpenAIRequest{
 			Model:    "gpt-test",
 			Messages: []dto.Message{{Role: "user", Content: "hello"}},
 		})
@@ -92,7 +93,7 @@ func TestChatCompletionsRequestToResponsesRequestPreservesQwenThinkingBudget(t *
 				},
 			}
 
-			got, err := ChatCompletionsRequestToResponsesRequest(req)
+			got, err := ChatCompletionsRequestToResponsesRequest(context.Background(), req)
 			require.NoError(t, err)
 			assert.Equal(t, tt.budget, got.ThinkingBudget)
 
@@ -108,7 +109,7 @@ func TestChatCompletionsRequestToResponsesRequestPreservesQwenThinkingBudget(t *
 }
 
 func TestChatCompletionsRequestToResponsesRequestRejectsMultipleChoices(t *testing.T) {
-	_, err := ChatCompletionsRequestToResponsesRequest(&dto.GeneralOpenAIRequest{
+	_, err := ChatCompletionsRequestToResponsesRequest(context.Background(), &dto.GeneralOpenAIRequest{
 		Model: "gpt-test",
 		N:     lo.ToPtr(2),
 	})
@@ -147,7 +148,7 @@ func TestChatCompletionsRequestToResponsesRequestPreservesPenalties(t *testing.T
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			got, err := ChatCompletionsRequestToResponsesRequest(&dto.GeneralOpenAIRequest{
+			got, err := ChatCompletionsRequestToResponsesRequest(context.Background(), &dto.GeneralOpenAIRequest{
 				Model:            "gpt-test",
 				Messages:         []dto.Message{{Role: "user", Content: "hello"}},
 				FrequencyPenalty: tt.frequency,

relaykit/relayconvert/internal/oai_chat/to_oai_responses_resp.go

@@ -42,7 +42,7 @@ func ChatCompletionsResponseToResponsesResponse(resp *dto.OpenAITextResponse, id
 	out := &dto.OpenAIResponsesResponse{
 		ID:        id,
 		Object:    "response",
-		CreatedAt: chatCreatedAt(resp.Created),
+		CreatedAt: dto.IntValue(chatCreatedAt(resp.Created)),
 		Status:    []byte(`"completed"`),
 		Model:     resp.Model,
 		Output:    make([]dto.ResponsesOutput, 0),

relaykit/relayconvert/internal/oai_chat/to_oai_responses_stream_resp.go

@@ -665,7 +665,7 @@ func (s *ChatToResponsesStreamState) finalResponse() *dto.OpenAIResponsesRespons
 	return &dto.OpenAIResponsesResponse{
 		ID:                s.ID,
 		Object:            "response",
-		CreatedAt:         int(s.Created),
+		CreatedAt:         dto.IntValue(s.Created),
 		Status:            []byte(fmt.Sprintf("%q", s.status)),
 		IncompleteDetails: s.incompleteDetails,
 		Model:             s.Model,
@@ -678,7 +678,7 @@ func (s *ChatToResponsesStreamState) createdResponse() *dto.OpenAIResponsesRespo
 	return &dto.OpenAIResponsesResponse{
 		ID:        s.ID,
 		Object:    "response",
-		CreatedAt: int(s.Created),
+		CreatedAt: dto.IntValue(s.Created),
 		Status:    []byte(`"in_progress"`),
 		Model:     s.Model,
 		Output:    []dto.ResponsesOutput{},

relaykit/relayconvert/internal/oai_responses/to_claude_messages_req.go

@@ -7,6 +7,7 @@ import (
 
 	"github.com/QuantumNous/new-api/relaykit/dto"
 	"github.com/QuantumNous/new-api/relaykit/relayconvert/convmeta"
+	"github.com/QuantumNous/new-api/relaykit/relayconvert/internal/convdiag"
 	relaymedia "github.com/QuantumNous/new-api/relaykit/relayconvert/internal/media"
 	sharedclaude "github.com/QuantumNous/new-api/relaykit/relayconvert/internal/shared/claude"
 	kitutil "github.com/QuantumNous/new-api/relaykit/relayconvert/kitutil"
@@ -56,10 +57,11 @@ func OpenAIResponsesRequestToClaudeMessages(c context.Context, info convmeta.Met
 	if toolChoice != nil || RawJSONPresent(req.ParallelToolCalls) {
 		claudeRequest.ToolChoice = sharedclaude.MapOpenAIToolChoice(toolChoice, ParallelToolCalls(req.ParallelToolCalls))
 	}
-	sourceReasoning, err := reasoning.FromOpenAIResponses(req)
+	sourceReasoning, diagnostics, err := reasoning.FromOpenAIResponses(req)
 	if err != nil {
 		return nil, reasoning.AsClientError(err)
 	}
+	convdiag.Add(c, diagnostics...)
 	if err := sharedclaude.ApplyReasoning(c, claudeRequest, info, sourceReasoning, true); err != nil {
 		return nil, reasoning.AsClientError(err)
 	}

relaykit/relayconvert/internal/oai_responses/to_gemini_chat_req.go

@@ -8,6 +8,7 @@ import (
 
 	"github.com/QuantumNous/new-api/relaykit/dto"
 	"github.com/QuantumNous/new-api/relaykit/relayconvert/convmeta"
+	"github.com/QuantumNous/new-api/relaykit/relayconvert/internal/convdiag"
 	relaymedia "github.com/QuantumNous/new-api/relaykit/relayconvert/internal/media"
 	sharedgemini "github.com/QuantumNous/new-api/relaykit/relayconvert/internal/shared/gemini"
 	kitutil "github.com/QuantumNous/new-api/relaykit/relayconvert/kitutil"
@@ -61,17 +62,18 @@ func OpenAIResponsesRequestToGeminiChat(c context.Context, req *dto.OpenAIRespon
 	if err := applyResponsesTextToGemini(req.Text, geminiRequest); err != nil {
 		return nil, err
 	}
-	reasoningIntent, err := reasoning.FromOpenAIResponses(req)
+	reasoningIntent, diagnostics, err := reasoning.FromOpenAIResponses(req)
 	if err != nil {
 		return nil, reasoning.AsClientError(err)
 	}
+	convdiag.Add(c, diagnostics...)
 	var reasoningPivot dto.GeneralOpenAIRequest
 	if err := reasoning.ApplyToOpenAIChat(&reasoningPivot, reasoningIntent); err != nil {
 		return nil, reasoning.AsClientError(err)
 	}
 	reasoningPivot.Model = req.Model
 	reasoningPivot.MaxCompletionTokens = req.MaxOutputTokens
-	if err := sharedgemini.ApplyThinkingConfig(geminiRequest, info, reasoningPivot); err != nil {
+	if err := sharedgemini.ApplyThinkingConfig(c, geminiRequest, info, reasoningPivot); err != nil {
 		return nil, reasoning.AsClientError(err)
 	}
 

relaykit/relayconvert/internal/oai_responses/to_oai_chat_req.go

@@ -1,12 +1,14 @@
 package oairesponses
 
 import (
+	"context"
 	"encoding/json"
 	"errors"
 	"fmt"
 	"strings"
 
 	"github.com/QuantumNous/new-api/relaykit/dto"
+	"github.com/QuantumNous/new-api/relaykit/relayconvert/internal/convdiag"
 	kitutil "github.com/QuantumNous/new-api/relaykit/relayconvert/kitutil"
 	"github.com/QuantumNous/new-api/relaykit/relayconvert/reasoning"
 )
@@ -25,7 +27,7 @@ const (
 	ResponsesInputTypeCustomToolOutput   = responsesInputTypeCustomToolOutput
 )
 
-func ResponsesRequestToChatCompletionsRequest(req *dto.OpenAIResponsesRequest) (*dto.GeneralOpenAIRequest, error) {
+func ResponsesRequestToChatCompletionsRequest(ctx context.Context, req *dto.OpenAIResponsesRequest) (*dto.GeneralOpenAIRequest, error) {
 	if req == nil {
 		return nil, errors.New("request is nil")
 	}
@@ -86,9 +88,12 @@ func ResponsesRequestToChatCompletionsRequest(req *dto.OpenAIResponsesRequest) (
 		return nil, fmt.Errorf("invalid presence_penalty: %w", err)
 	}
 
-	if reasoningIntent, err := reasoning.FromOpenAIResponses(req); err != nil {
+	reasoningIntent, diagnostics, err := reasoning.FromOpenAIResponses(req)
+	if err != nil {
 		return nil, reasoning.AsClientError(err)
-	} else if err := reasoning.ApplyToOpenAIChat(out, reasoningIntent); err != nil {
+	}
+	convdiag.Add(ctx, diagnostics...)
+	if err := reasoning.ApplyToOpenAIChat(out, reasoningIntent); err != nil {
 		return nil, reasoning.AsClientError(err)
 	}
 	if req.ServiceTier != "" {

relaykit/relayconvert/internal/oai_responses/to_oai_chat_req_test.go

@@ -1,6 +1,7 @@
 package oairesponses
 
 import (
+	"context"
 	"encoding/json"
 	"testing"
 
@@ -19,7 +20,7 @@ func TestResponsesRequestToChatCompletionsRequestInstructionsAndScalarInput(t *t
 	maxOutputTokens := uint(128)
 	parallelToolCalls := true
 
-	got, err := ResponsesRequestToChatCompletionsRequest(&dto.OpenAIResponsesRequest{
+	got, err := ResponsesRequestToChatCompletionsRequest(context.Background(), &dto.OpenAIResponsesRequest{
 		Model:                "gpt-test",
 		Instructions:         mustRawMessage(t, "system rules"),
 		Input:                mustRawMessage(t, "hello"),
@@ -68,7 +69,7 @@ func TestResponsesRequestToChatCompletionsRequestPreservesQwenThinkingBudget(t *
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			got, err := ResponsesRequestToChatCompletionsRequest(&dto.OpenAIResponsesRequest{
+			got, err := ResponsesRequestToChatCompletionsRequest(context.Background(), &dto.OpenAIResponsesRequest{
 				Model:          "qwen-plus",
 				Input:          mustRawMessage(t, "hello"),
 				EnableThinking: json.RawMessage(`true`),
@@ -89,7 +90,7 @@ func TestResponsesRequestToChatCompletionsRequestPreservesQwenThinkingBudget(t *
 }
 
 func TestResponsesRequestToChatCompletionsRequestMultimodalInput(t *testing.T) {
-	got, err := ResponsesRequestToChatCompletionsRequest(&dto.OpenAIResponsesRequest{
+	got, err := ResponsesRequestToChatCompletionsRequest(context.Background(), &dto.OpenAIResponsesRequest{
 		Model: "gpt-test",
 		Input: mustRawMessage(t, []map[string]any{
 			{
@@ -123,7 +124,7 @@ func TestResponsesRequestToChatCompletionsRequestMultimodalInput(t *testing.T) {
 }
 
 func TestResponsesRequestToChatCompletionsRequestAssistantTextAndFunctionCallCoexist(t *testing.T) {
-	got, err := ResponsesRequestToChatCompletionsRequest(&dto.OpenAIResponsesRequest{
+	got, err := ResponsesRequestToChatCompletionsRequest(context.Background(), &dto.OpenAIResponsesRequest{
 		Model: "gpt-test",
 		Input: mustRawMessage(t, []map[string]any{
 			{
@@ -162,7 +163,7 @@ func TestResponsesRequestToChatCompletionsRequestAssistantTextAndFunctionCallCoe
 }
 
 func TestResponsesRequestToChatCompletionsRequestOnlyFunctionCallCreatesAssistant(t *testing.T) {
-	got, err := ResponsesRequestToChatCompletionsRequest(&dto.OpenAIResponsesRequest{
+	got, err := ResponsesRequestToChatCompletionsRequest(context.Background(), &dto.OpenAIResponsesRequest{
 		Model: "gpt-test",
 		Input: mustRawMessage(t, []map[string]any{
 			{
@@ -184,7 +185,7 @@ func TestResponsesRequestToChatCompletionsRequestOnlyFunctionCallCreatesAssistan
 }
 
 func TestResponsesRequestToChatCompletionsRequestToolsToolChoiceAndTextFormat(t *testing.T) {
-	got, err := ResponsesRequestToChatCompletionsRequest(&dto.OpenAIResponsesRequest{
+	got, err := ResponsesRequestToChatCompletionsRequest(context.Background(), &dto.OpenAIResponsesRequest{
 		Model: "gpt-test",
 		Input: mustRawMessage(t, "hello"),
 		Tools: mustRawMessage(t, []map[string]any{
@@ -233,7 +234,7 @@ func TestResponsesRequestToChatCompletionsRequestToolsToolChoiceAndTextFormat(t
 }
 
 func TestResponsesRequestToChatCompletionsRequestCustomToolCallPreservesRawShape(t *testing.T) {
-	got, err := ResponsesRequestToChatCompletionsRequest(&dto.OpenAIResponsesRequest{
+	got, err := ResponsesRequestToChatCompletionsRequest(context.Background(), &dto.OpenAIResponsesRequest{
 		Model: "gpt-test",
 		Input: mustRawMessage(t, []map[string]any{
 			{
@@ -287,7 +288,7 @@ func TestResponsesRequestToChatCompletionsRequestRejectsStatefulFields(t *testin
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			_, err := ResponsesRequestToChatCompletionsRequest(tt.req)
+			_, err := ResponsesRequestToChatCompletionsRequest(context.Background(), tt.req)
 			require.Error(t, err)
 			assert.Contains(t, err.Error(), tt.want)
 			assert.Contains(t, err.Error(), "stateful fields")
@@ -326,7 +327,7 @@ func TestResponsesRequestToChatCompletionsRequestPreservesPenalties(t *testing.T
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			got, err := ResponsesRequestToChatCompletionsRequest(&dto.OpenAIResponsesRequest{
+			got, err := ResponsesRequestToChatCompletionsRequest(context.Background(), &dto.OpenAIResponsesRequest{
 				Model:            "gpt-test",
 				Input:            mustRawMessage(t, "hello"),
 				FrequencyPenalty: tt.frequencyRaw,
@@ -341,7 +342,7 @@ func TestResponsesRequestToChatCompletionsRequestPreservesPenalties(t *testing.T
 }
 
 func TestResponsesRequestToChatCompletionsRequestRejectsMalformedPenalty(t *testing.T) {
-	_, err := ResponsesRequestToChatCompletionsRequest(&dto.OpenAIResponsesRequest{
+	_, err := ResponsesRequestToChatCompletionsRequest(context.Background(), &dto.OpenAIResponsesRequest{
 		Model:            "gpt-test",
 		Input:            mustRawMessage(t, "hello"),
 		FrequencyPenalty: json.RawMessage(`"not-a-number"`),

relaykit/relayconvert/internal/oai_responses/to_oai_chat_resp.go

@@ -66,7 +66,7 @@ func ResponsesResponseToChatCompletionsResponse(resp *dto.OpenAIResponsesRespons
 
 	usage := UsageFromResponsesUsage(resp.Usage)
 
-	created := resp.CreatedAt
+	created := int64(resp.CreatedAt)
 
 	var toolCalls []dto.ToolCallResponse
 	if len(resp.Output) > 0 {
@@ -218,12 +218,7 @@ func usageFromResponsesUsage(src *dto.Usage, createBillingSnapshot bool) *dto.Us
 		usage.TotalTokens = usage.PromptTokens + usage.CompletionTokens
 	}
 	if src.InputTokensDetails != nil {
-		usage.PromptTokensDetails.CachedTokens = src.InputTokensDetails.CachedTokens
-		usage.PromptTokensDetails.CachedCreationTokens = src.InputTokensDetails.CachedCreationTokens
-		usage.PromptTokensDetails.CacheWriteTokens = src.InputTokensDetails.CacheWriteTokens
-		usage.PromptTokensDetails.TextTokens = src.InputTokensDetails.TextTokens
-		usage.PromptTokensDetails.ImageTokens = src.InputTokensDetails.ImageTokens
-		usage.PromptTokensDetails.AudioTokens = src.InputTokensDetails.AudioTokens
+		usage.PromptTokensDetails = src.InputTokensDetails.Clone()
 	}
 	if src.CompletionTokenDetails.ReasoningTokens != 0 ||
 		src.CompletionTokenDetails.TextTokens != 0 ||

relaykit/relayconvert/internal/oai_responses/to_oai_chat_resp_test.go

@@ -9,6 +9,49 @@ import (
 	"github.com/stretchr/testify/require"
 )
 
+func TestNormalizeResponsesUsagePreservesImageCacheDetails(t *testing.T) {
+	for _, tc := range []struct {
+		name        string
+		imageTokens int
+	}{
+		{name: "image cache", imageTokens: 200},
+		{name: "explicit zero image cache", imageTokens: 0},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			imageTokens, audioTokens := tc.imageTokens, 0
+			source := &dto.Usage{
+				InputTokens:  1000,
+				OutputTokens: 100,
+				InputTokensDetails: &dto.InputTokenDetails{
+					CachedTokens:         300,
+					CachedCreationTokens: 20,
+					CacheWriteTokens:     10,
+					TextTokens:           400,
+					ImageTokens:          600,
+					CachedTokensDetails: &dto.CachedTokenDetails{
+						ImageTokens: &imageTokens,
+						AudioTokens: &audioTokens,
+					},
+				},
+			}
+
+			usage := NormalizeResponsesUsage(source)
+			assert.Equal(t, *source.InputTokensDetails, usage.PromptTokensDetails)
+			details := usage.PromptTokensDetails.CachedTokensDetails
+			require.NotNil(t, details)
+			require.NotNil(t, details.ImageTokens)
+			require.NotNil(t, details.AudioTokens)
+			assert.Nil(t, details.TextTokens, "missing modalities must remain absent")
+
+			imageTokens, audioTokens = 999, 888
+			source.InputTokensDetails.ImageTokens = 9999
+			assert.Equal(t, tc.imageTokens, *details.ImageTokens, "later source updates must not change billable image cache")
+			assert.Zero(t, *details.AudioTokens, "explicit zero must survive normalization and source mutation")
+			assert.Equal(t, 600, usage.PromptTokensDetails.ImageTokens)
+		})
+	}
+}
+
 func TestResponsesResponseToChatCompletionsPreservesTextAndToolCalls(t *testing.T) {
 	resp := &dto.OpenAIResponsesResponse{
 		ID:        "resp_1",

relaykit/relayconvert/internal/shared/claude/reasoning.go

@@ -52,21 +52,24 @@ func ApplyReasoning(ctx context.Context, req *dto.ClaudeRequest, info convmeta.M
 		return nil
 	}
 
-	native, err := reasoning.FromClaude(req)
+	native, diagnostics, err := reasoning.FromClaude(req)
 	if err != nil {
 		return err
 	}
-	explicit, err := reasoning.MergeExplicit(native, source, req.Model)
+	convdiag.Add(ctx, diagnostics...)
+	explicit, diagnostics, err := reasoning.MergeExplicit(native, source, req.Model)
 	if err != nil {
 		return err
 	}
+	convdiag.Add(ctx, diagnostics...)
 	if info != nil && !reasoning.IsKnownClaudeModel(capabilityModel) && reasoning.IsKnownClaudeModel(info.GetOriginModelName()) {
 		capabilityModel = info.GetOriginModelName()
 	}
-	intent, err := reasoning.MergeExplicitAndSuffix(explicit, suffix, req.Model)
+	intent, diagnostics, err := reasoning.MergeExplicitAndSuffix(explicit, suffix, req.Model)
 	if err != nil {
 		return err
 	}
+	convdiag.Add(ctx, diagnostics...)
 	knownClaudeModel := reasoning.IsKnownClaudeModel(capabilityModel)
 	if !knownClaudeModel && intent.Mode == reasoning.ModeAdaptive {
 		// Cross-protocol pivots cannot safely assume that an unknown
@@ -80,8 +83,8 @@ func ApplyReasoning(ctx context.Context, req *dto.ClaudeRequest, info convmeta.M
 	}
 	if req.MaxTokens == nil && intent.HasStrength() {
 		// Adapter-provided defaults may be raised to accommodate an exact
-		// cross-protocol budget. Explicit client max_tokens values are never
-		// expanded and remain subject to the renderer's strict validation.
+		// cross-protocol budget. Values too small for manual thinking are
+		// raised again by the renderer, which reports the change.
 		minimum := uint(1280)
 		if configuredDefault, configured := opts.Claude.DefaultMaxTokensFor(capabilityModel); configured && configuredDefault > 0 {
 			minimum = uint(configuredDefault)
@@ -108,6 +111,9 @@ func ApplyReasoning(ctx context.Context, req *dto.ClaudeRequest, info convmeta.M
 	}
 	convdiag.Add(ctx, rendered.Diagnostics...)
 	req.Model = baseModel
+	if rendered.MaxTokens != nil {
+		req.MaxTokens = rendered.MaxTokens
+	}
 	if rendered.Thinking != nil {
 		req.Thinking = rendered.Thinking
 	}

relaykit/relayconvert/internal/shared/gemini/request.go

@@ -1,13 +1,16 @@
 package gemini
 
 import (
+	"context"
 	"fmt"
 	"strconv"
 	"strings"
 
 	"github.com/QuantumNous/new-api/relaykit/dto"
 	"github.com/QuantumNous/new-api/relaykit/relayconvert/convmeta"
+	"github.com/QuantumNous/new-api/relaykit/relayconvert/internal/convdiag"
 	"github.com/QuantumNous/new-api/relaykit/relayconvert/reasoning"
+	"github.com/QuantumNous/new-api/relaykit/types"
 )
 
 var SupportedMimeTypes = map[string]bool{
@@ -73,7 +76,12 @@ func AttachFirstTextThoughtSignature(opts *convmeta.Options, parts []dto.GeminiP
 	return false
 }
 
-func ApplyThinkingConfig(geminiRequest *dto.GeminiChatRequest, info convmeta.Meta, oaiRequest ...dto.GeneralOpenAIRequest) error {
+// ApplyThinkingConfig resolves every reasoning control that can reach a
+// Gemini request (provider-native thinkingConfig, generic OpenAI fields, and
+// host model-name aliases) into the target model's dialect. Precedence is
+// model name, then provider-native config, then generic fields; each override
+// and each capability coercion is reported through convdiag on ctx.
+func ApplyThinkingConfig(ctx context.Context, geminiRequest *dto.GeminiChatRequest, info convmeta.Meta, oaiRequest ...dto.GeneralOpenAIRequest) error {
 	opts := convmeta.OptionsOf(info)
 	if geminiRequest == nil {
 		return nil
@@ -86,11 +94,13 @@ func ApplyThinkingConfig(geminiRequest *dto.GeminiChatRequest, info convmeta.Met
 		if modelName == "" {
 			modelName = oaiRequest[0].Model
 		}
+		var diagnostics []types.ConversionDiagnostic
 		var err error
-		source, err = reasoning.FromOpenAIChat(&oaiRequest[0])
+		source, diagnostics, err = reasoning.FromOpenAIChat(&oaiRequest[0])
 		if err != nil {
 			return err
 		}
+		convdiag.Add(ctx, diagnostics...)
 	}
 
 	baseModel := modelName
@@ -110,6 +120,12 @@ func ApplyThinkingConfig(geminiRequest *dto.GeminiChatRequest, info convmeta.Met
 			effort := ""
 			if config := geminiRequest.GenerationConfig.ThinkingConfig; config != nil {
 				effort = config.ThinkingLevel
+				// Gemini accepts thinkingLevel case-insensitively; record the
+				// canonical effort so logs match other protocols. Unknown values
+				// stay as sent because this path does not validate.
+				if canonical, err := reasoning.ParseEffort(effort); err == nil {
+					effort = string(canonical)
+				}
 				if effort == "" && config.ThinkingBudget != nil {
 					effort = string(reasoning.EffortFromBudget(*config.ThinkingBudget))
 				}
@@ -118,33 +134,47 @@ func ApplyThinkingConfig(geminiRequest *dto.GeminiChatRequest, info convmeta.Met
 		}
 		return nil
 	}
-	native, err := reasoning.FromGemini(geminiRequest)
+	// Rewrite the provider-native config into the target model's dialect
+	// first, so a budget on Gemini 3 or a level on Gemini 2.5 is converted
+	// rather than rejected and later comparisons see what will be sent.
+	nativeEffort, diagnostics, err := reasoning.NormalizeGeminiThinkingConfig(baseModel, &geminiRequest.GenerationConfig)
 	if err != nil {
 		return err
 	}
-	source = reasoning.ResolveGeminiEnabledDefault(baseModel, source, geminiRequest.GenerationConfig.MaxOutputTokens)
+	convdiag.Add(ctx, diagnostics...)
+	native, diagnostics, err := reasoning.FromGemini(geminiRequest)
+	if err != nil {
+		return err
+	}
+	convdiag.Add(ctx, diagnostics...)
 	if native.HasStrength() && source.HasStrength() {
-		equivalent, compareErr := reasoning.EquivalentGeminiStrength(baseModel, native, source)
-		if compareErr != nil {
-			return compareErr
+		// Native Gemini configuration is the lossless representation and wins
+		// over the generic OpenAI fields. Only a generic effort or budget can
+		// disagree with it; a bare enable states no strength to compare.
+		if source.Effort != "" || source.BudgetTokens != nil {
+			equivalent, compareErr := reasoning.EquivalentGeminiStrength(baseModel, native, source)
+			if compareErr != nil {
+				return compareErr
+			}
+			if !equivalent {
+				convdiag.Add(ctx, types.ConversionDiagnostic{
+					Code:     "native_overrode_standard",
+					Path:     "generationConfig.thinkingConfig",
+					Message:  fmt.Sprintf("model %q: Gemini thinking_config effort %q overrides the standard reasoning effort %q", modelName, reasoning.EffectiveEffort(native), reasoning.EffectiveEffort(source)),
+					Severity: types.ConversionDiagnosticWarning,
+				})
+			}
 		}
-		if !equivalent {
-			nativeEffort := reasoning.EffectiveEffort(native)
-			sourceEffort := reasoning.EffectiveEffort(source)
-			return fmt.Errorf("%w for model %q: Gemini thinking_config effort %q differs from standard effort %q", reasoning.ErrEffortConflict, modelName, nativeEffort, sourceEffort)
-		}
-		// Native Gemini configuration is the lossless representation. Once the
-		// two controls are equivalent, retain only portable visibility metadata
-		// from the standard representation.
 		if native.IncludeThoughts == nil {
 			native.IncludeThoughts = source.IncludeThoughts
 		}
 		source = reasoning.Intent{}
 	}
-	explicit, err := reasoning.MergeExplicit(native, source, modelName)
+	explicit, diagnostics, err := reasoning.MergeExplicit(native, source, modelName)
 	if err != nil {
 		return err
 	}
+	convdiag.Add(ctx, diagnostics...)
 	if explicit.HasStrength() && suffix.HasStrength() {
 		equivalent, compareErr := reasoning.EquivalentGeminiStrength(baseModel, explicit, suffix)
 		if compareErr != nil {
@@ -157,22 +187,21 @@ func ApplyThinkingConfig(geminiRequest *dto.GeminiChatRequest, info convmeta.Met
 			suffix = reasoning.Intent{}
 		}
 	}
-	requested, err := reasoning.MergeExplicitAndSuffix(explicit, suffix, modelName)
+	requested, diagnostics, err := reasoning.MergeExplicitAndSuffix(explicit, suffix, modelName)
 	if err != nil {
 		return err
 	}
+	convdiag.Add(ctx, diagnostics...)
 	requested = reasoning.ResolveGeminiEnabledDefault(baseModel, requested, geminiRequest.GenerationConfig.MaxOutputTokens)
 
 	if native.HasStrength() && !suffix.HasStrength() {
+		// The normalized native config is what goes upstream; only portable
+		// visibility metadata is taken from the standard representation.
 		if explicit.IncludeThoughts != nil {
 			geminiRequest.GenerationConfig.ThinkingConfig.IncludeThoughts = explicit.IncludeThoughts
 		}
-		effort, err := reasoning.ValidateGeminiThinkingConfig(baseModel, geminiRequest.GenerationConfig.ThinkingConfig)
-		if err != nil {
-			return err
-		}
-		if info != nil && effort != "" {
-			info.SetReasoningEffort(string(effort))
+		if info != nil && nativeEffort != "" {
+			info.SetReasoningEffort(string(nativeEffort))
 		}
 		return nil
 	}
@@ -188,6 +217,7 @@ func ApplyThinkingConfig(geminiRequest *dto.GeminiChatRequest, info convmeta.Met
 	if err != nil {
 		return err
 	}
+	convdiag.Add(ctx, rendered.Diagnostics...)
 	geminiRequest.GenerationConfig.ThinkingConfig = rendered.Config
 	if info != nil && rendered.EffectiveEffort != "" {
 		info.SetReasoningEffort(string(rendered.EffectiveEffort))