feat: publish v1.0.0-rc.36-custom.1

relay/channel/claude/adaptor.go

@@ -0,0 +1,172 @@
+package claude
+
+import (
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+	"net/url"
+
+	"github.com/QuantumNous/new-api/relay/channel"
+	relaycommon "github.com/QuantumNous/new-api/relay/common"
+	"github.com/QuantumNous/new-api/relaykit/dto"
+	"github.com/QuantumNous/new-api/relaykit/relayconvert"
+	"github.com/QuantumNous/new-api/relaykit/types"
+	"github.com/QuantumNous/new-api/service"
+	"github.com/QuantumNous/new-api/setting/model_setting"
+
+	"github.com/gin-gonic/gin"
+)
+
+type Adaptor struct {
+}
+
+func (a *Adaptor) ConvertGeminiRequest(c *gin.Context, info *relaycommon.RelayInfo, request *dto.GeminiChatRequest) (any, error) {
+	if request == nil {
+		return nil, errors.New("request is nil")
+	}
+	result, err := service.ConvertRequest(c, info, types.RelayFormatClaude, request)
+	if err != nil {
+		return nil, err
+	}
+	return result.Value, nil
+}
+
+func (a *Adaptor) ConvertClaudeRequest(c *gin.Context, info *relaycommon.RelayInfo, request *dto.ClaudeRequest) (any, error) {
+	if request.MaxTokens != nil && *request.MaxTokens == 0 {
+		request.MaxTokens = nil
+	}
+	if err := relayconvert.ApplyClaudeThinkingModel(request, info); err != nil {
+		return nil, err
+	}
+	if request.MaxTokens == nil {
+		defaultMaxTokens := uint(model_setting.GetClaudeSettings().GetDefaultMaxTokens(request.Model))
+		request.MaxTokens = &defaultMaxTokens
+	}
+	// ApplyClaudeThinkingModel no longer rewrites request.Model. Do not write
+	// a still-suffixed name back over the entry-normalized UpstreamModelName
+	// (AWS/Vertex look up getAwsModelID / claudeModelMap from that field).
+	if info.UpstreamModelName == "" {
+		info.UpstreamModelName = request.Model
+	}
+	return request, nil
+}
+
+func (a *Adaptor) ConvertAudioRequest(c *gin.Context, info *relaycommon.RelayInfo, request dto.AudioRequest) (io.Reader, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
+func (a *Adaptor) ConvertImageRequest(c *gin.Context, info *relaycommon.RelayInfo, request dto.ImageRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
+func (a *Adaptor) Init(info *relaycommon.RelayInfo) {
+}
+
+func (a *Adaptor) GetRequestURL(info *relaycommon.RelayInfo) (string, error) {
+	requestURL := fmt.Sprintf("%s/v1/messages", info.ChannelBaseUrl)
+	if !shouldAppendClaudeBetaQuery(info) {
+		return requestURL, nil
+	}
+
+	parsedURL, err := url.Parse(requestURL)
+	if err != nil {
+		return "", err
+	}
+	query := parsedURL.Query()
+	query.Set("beta", "true")
+	parsedURL.RawQuery = query.Encode()
+	return parsedURL.String(), nil
+}
+
+func shouldAppendClaudeBetaQuery(info *relaycommon.RelayInfo) bool {
+	if info == nil {
+		return false
+	}
+	if info.IsClaudeBetaQuery {
+		return true
+	}
+	if info.ChannelOtherSettings.ClaudeBetaQuery {
+		return true
+	}
+	return false
+}
+
+func CommonClaudeHeadersOperation(c *gin.Context, req *http.Header, info *relaycommon.RelayInfo) {
+	// common headers operation
+	anthropicBeta := c.Request.Header.Get("anthropic-beta")
+	if anthropicBeta != "" {
+		req.Set("anthropic-beta", anthropicBeta)
+	}
+	model_setting.GetClaudeSettings().WriteHeaders(info.OriginModelName, req)
+}
+
+func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Header, info *relaycommon.RelayInfo) error {
+	channel.SetupApiRequestHeader(info, c, req)
+	req.Set("x-api-key", info.ApiKey)
+	anthropicVersion := c.Request.Header.Get("anthropic-version")
+	if anthropicVersion == "" {
+		anthropicVersion = "2023-06-01"
+	}
+	req.Set("anthropic-version", anthropicVersion)
+	CommonClaudeHeadersOperation(c, req, info)
+	return nil
+}
+
+func (a *Adaptor) ConvertOpenAIRequest(c *gin.Context, info *relaycommon.RelayInfo, request *dto.GeneralOpenAIRequest) (any, error) {
+	if request == nil {
+		return nil, errors.New("request is nil")
+	}
+	result, err := service.ConvertRequest(c, info, types.RelayFormatClaude, request)
+	if err != nil {
+		return nil, err
+	}
+	return result.Value, nil
+}
+
+func (a *Adaptor) ConvertRerankRequest(c *gin.Context, relayMode int, request dto.RerankRequest) (any, error) {
+	return nil, nil
+}
+
+func (a *Adaptor) ConvertEmbeddingRequest(c *gin.Context, info *relaycommon.RelayInfo, request dto.EmbeddingRequest) (any, error) {
+	//TODO implement me
+	return nil, errors.New("not implemented")
+}
+
+func (a *Adaptor) ConvertOpenAIResponsesRequest(c *gin.Context, info *relaycommon.RelayInfo, request dto.OpenAIResponsesRequest) (any, error) {
+	result, err := service.ConvertRequest(c, info, types.RelayFormatClaude, &request)
+	if err != nil {
+		return nil, err
+	}
+	claudeRequest, ok := result.Value.(*dto.ClaudeRequest)
+	if !ok {
+		return nil, fmt.Errorf("expected Anthropic Messages request, got %T", result.Value)
+	}
+	return claudeRequest, 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) {
+	info.FinalRequestRelayFormat = types.RelayFormatClaude
+	if info.RelayFormat == types.RelayFormatOpenAIResponses && info.IsStream {
+		return ClaudeResponsesStreamHandler(c, resp, info)
+	}
+	if info.IsStream {
+		return ClaudeStreamHandler(c, resp, info)
+	} else {
+		return ClaudeHandler(c, resp, info)
+	}
+}
+
+func (a *Adaptor) GetModelList() []string {
+	return ModelList
+}
+
+func (a *Adaptor) GetChannelName() string {
+	return ChannelName
+}

relay/channel/claude/adaptor_test.go

@@ -0,0 +1,225 @@
+package claude
+
+import (
+	"net/http/httptest"
+	"testing"
+
+	"github.com/QuantumNous/new-api/common"
+	relaycommon "github.com/QuantumNous/new-api/relay/common"
+	"github.com/QuantumNous/new-api/relay/helper"
+	"github.com/QuantumNous/new-api/relaykit/dto"
+	"github.com/QuantumNous/new-api/setting/model_setting"
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestConvertClaudeRequestTreatsZeroMaxTokensAsUnset(t *testing.T) {
+	zero := uint(0)
+	req := &dto.ClaudeRequest{
+		Model:     "claude-sonnet-4-5",
+		MaxTokens: &zero,
+		Messages: []dto.ClaudeMessage{
+			{Role: "user", Content: "hello"},
+		},
+	}
+	info := &relaycommon.RelayInfo{
+		ChannelMeta: &relaycommon.ChannelMeta{
+			UpstreamModelName: "claude-sonnet-4-5",
+		},
+	}
+
+	out, err := (&Adaptor{}).ConvertClaudeRequest(nil, info, req)
+	require.NoError(t, err)
+	converted, ok := out.(*dto.ClaudeRequest)
+	require.True(t, ok)
+	require.NotNil(t, converted.MaxTokens)
+	assert.Equal(t, uint(model_setting.GetClaudeSettings().GetDefaultMaxTokens(req.Model)), *converted.MaxTokens)
+}
+
+func TestConvertClaudeRequestPreservesNativeClaudeCodeThinking(t *testing.T) {
+	budget := 10000
+	maxTokens := uint(20000)
+	temperature := 0.7
+	topP := 0.9
+	req := &dto.ClaudeRequest{
+		Model:        "claude-opus-4-8",
+		MaxTokens:    &maxTokens,
+		Temperature:  &temperature,
+		TopP:         &topP,
+		Thinking:     &dto.Thinking{Type: "enabled", BudgetTokens: &budget},
+		OutputConfig: []byte(`{"effort":"high"}`),
+		Messages: []dto.ClaudeMessage{
+			{Role: "user", Content: "hello"},
+		},
+	}
+	info := &relaycommon.RelayInfo{
+		OriginModelName: req.Model,
+		ChannelMeta: &relaycommon.ChannelMeta{
+			UpstreamModelName: req.Model,
+		},
+	}
+
+	out, err := (&Adaptor{}).ConvertClaudeRequest(nil, info, req)
+	require.NoError(t, err)
+	converted, ok := out.(*dto.ClaudeRequest)
+	require.True(t, ok)
+	require.NotNil(t, converted.Thinking)
+	assert.Equal(t, "enabled", converted.Thinking.Type)
+	require.NotNil(t, converted.Thinking.BudgetTokens)
+	assert.Equal(t, budget, *converted.Thinking.BudgetTokens)
+	assert.Equal(t, temperature, *converted.Temperature)
+	assert.Equal(t, topP, *converted.TopP)
+	assert.JSONEq(t, `{"effort":"high"}`, string(converted.OutputConfig))
+	assert.Empty(t, info.ConversionDiagnostics())
+}
+
+func TestConvertClaudeRequestZeroMaxTokensStillRaisesThinkingBudget(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	c, _ := gin.CreateTestContext(httptest.NewRecorder())
+
+	zero := uint(0)
+	original := &dto.ClaudeRequest{
+		Model:     "claude-3-7-sonnet-thinking",
+		MaxTokens: &zero,
+		Messages: []dto.ClaudeMessage{
+			{Role: "user", Content: "hello"},
+		},
+	}
+	info := &relaycommon.RelayInfo{
+		OriginModelName: "claude-3-7-sonnet-thinking",
+		Request:         original,
+		ChannelMeta: &relaycommon.ChannelMeta{
+			UpstreamModelName: "claude-3-7-sonnet-thinking",
+		},
+	}
+	outbound, err := common.DeepCopy(original)
+	require.NoError(t, err)
+	require.NoError(t, helper.ModelMappedHelper(c, info, outbound))
+	err = helper.ApplyReasoningModelSuffix(nil, info, outbound)
+	require.NoError(t, err)
+
+	out, err := (&Adaptor{}).ConvertClaudeRequest(nil, info, outbound)
+	require.NoError(t, err)
+	converted, ok := out.(*dto.ClaudeRequest)
+	require.True(t, ok)
+	assert.Equal(t, "claude-3-7-sonnet", converted.Model)
+	require.NotNil(t, converted.Thinking)
+	require.NotNil(t, converted.MaxTokens)
+	assert.Greater(t, *converted.MaxTokens, uint(1024))
+}
+
+func TestConvertClaudeRequestDoesNotOverwriteTrimmedUpstreamModelName(t *testing.T) {
+	req := &dto.ClaudeRequest{
+		Model: "claude-3-7-sonnet-thinking",
+		Messages: []dto.ClaudeMessage{
+			{Role: "user", Content: "hello"},
+		},
+	}
+	info := &relaycommon.RelayInfo{
+		ChannelMeta: &relaycommon.ChannelMeta{
+			UpstreamModelName: "claude-3-7-sonnet",
+		},
+	}
+
+	_, err := (&Adaptor{}).ConvertClaudeRequest(nil, info, req)
+	require.NoError(t, err)
+	assert.Equal(t, "claude-3-7-sonnet", info.UpstreamModelName)
+}
+
+func geminiToClaudeInfo() *relaycommon.RelayInfo {
+	return &relaycommon.RelayInfo{
+		OriginModelName: "claude-3-7-sonnet",
+		ChannelMeta: &relaycommon.ChannelMeta{
+			UpstreamModelName: "claude-3-7-sonnet",
+		},
+	}
+}
+
+func TestConvertGeminiRequestMapsSystemInstructionToolsAndMultimodal(t *testing.T) {
+	req := &dto.GeminiChatRequest{
+		Contents: []dto.GeminiChatContent{
+			{
+				Role: "user",
+				Parts: []dto.GeminiPart{
+					{Text: "What is in this image?"},
+					{InlineData: &dto.GeminiInlineData{MimeType: "image/png", Data: "aGVsbG8="}},
+				},
+			},
+		},
+		SystemInstructions: &dto.GeminiChatContent{
+			Parts: []dto.GeminiPart{{Text: "You are a helpful assistant."}},
+		},
+	}
+	req.SetTools([]dto.GeminiChatTool{
+		{
+			FunctionDeclarations: []dto.FunctionRequest{
+				{
+					Name:        "lookup",
+					Description: "Lookup data",
+					Parameters: map[string]any{
+						"type":       "object",
+						"properties": map[string]any{"q": map[string]any{"type": "string"}},
+					},
+				},
+			},
+		},
+	})
+
+	out, err := (&Adaptor{}).ConvertGeminiRequest(nil, geminiToClaudeInfo(), req)
+	require.NoError(t, err)
+	converted, ok := out.(*dto.ClaudeRequest)
+	require.True(t, ok)
+
+	system := converted.ParseSystem()
+	require.NotEmpty(t, system)
+	assert.Contains(t, system[0].GetText(), "You are a helpful assistant.")
+	require.NotEmpty(t, converted.Messages)
+	assert.Equal(t, "user", converted.Messages[0].Role)
+
+	blocks, parseErr := converted.Messages[0].ParseContent()
+	require.NoError(t, parseErr)
+	var foundImage bool
+	for _, block := range blocks {
+		if block.Type == "image" || (block.Source != nil && block.Source.Type == "base64") {
+			foundImage = true
+			break
+		}
+	}
+	assert.True(t, foundImage)
+
+	require.NotNil(t, converted.Tools)
+	tools, err := common.Marshal(converted.Tools)
+	require.NoError(t, err)
+	assert.Contains(t, string(tools), `"lookup"`)
+	require.NotNil(t, converted.MaxTokens)
+	assert.Greater(t, *converted.MaxTokens, uint(0))
+}
+
+func TestConvertGeminiRequestThinkingConfigUsesReasoningIntent(t *testing.T) {
+	budget := 1024
+	maxTokens := uint(4096)
+	req := &dto.GeminiChatRequest{
+		Contents: []dto.GeminiChatContent{
+			{Role: "user", Parts: []dto.GeminiPart{{Text: "think"}}},
+		},
+		GenerationConfig: dto.GeminiChatGenerationConfig{
+			MaxOutputTokens: &maxTokens,
+			ThinkingConfig:  &dto.GeminiThinkingConfig{ThinkingBudget: &budget},
+		},
+	}
+
+	out, err := (&Adaptor{}).ConvertGeminiRequest(nil, geminiToClaudeInfo(), req)
+	require.NoError(t, err)
+	converted, ok := out.(*dto.ClaudeRequest)
+	require.True(t, ok)
+	require.NotNil(t, converted.Thinking)
+	assert.Equal(t, "enabled", converted.Thinking.Type)
+	require.NotNil(t, converted.Thinking.BudgetTokens)
+	assert.Equal(t, 1024, *converted.Thinking.BudgetTokens)
+}
+
+func TestConvertGeminiRequestNilRequest(t *testing.T) {
+	_, err := (&Adaptor{}).ConvertGeminiRequest(nil, geminiToClaudeInfo(), nil)
+	require.Error(t, err)
+}

relay/channel/claude/constants.go

@@ -0,0 +1,45 @@
+package claude
+
+var ModelList = []string{
+	"claude-3-sonnet-20240229",
+	"claude-3-opus-20240229",
+	"claude-3-haiku-20240307",
+	"claude-3-5-haiku-20241022",
+	"claude-haiku-4-5-20251001",
+	"claude-3-5-sonnet-20240620",
+	"claude-3-5-sonnet-20241022",
+	"claude-3-7-sonnet-20250219",
+	"claude-3-7-sonnet-20250219-thinking",
+	"claude-sonnet-4-20250514",
+	"claude-sonnet-4-20250514-thinking",
+	"claude-opus-4-20250514",
+	"claude-opus-4-20250514-thinking",
+	"claude-opus-4-1-20250805",
+	"claude-opus-4-1-20250805-thinking",
+	"claude-sonnet-4-5-20250929",
+	"claude-sonnet-4-5-20250929-thinking",
+	"claude-opus-4-5-20251101",
+	"claude-opus-4-5-20251101-thinking",
+	"claude-opus-4-6",
+	"claude-opus-4-6-max",
+	"claude-opus-4-6-high",
+	"claude-opus-4-6-medium",
+	"claude-opus-4-6-low",
+	"claude-sonnet-4-6",
+	"claude-opus-4-7",
+	"claude-opus-4-7-max",
+	"claude-opus-4-7-xhigh",
+	"claude-opus-4-7-high",
+	"claude-opus-4-7-medium",
+	"claude-opus-4-7-low",
+	"claude-opus-4-7-thinking",
+	"claude-opus-4-8",
+	"claude-opus-4-8-max",
+	"claude-opus-4-8-xhigh",
+	"claude-opus-4-8-high",
+	"claude-opus-4-8-medium",
+	"claude-opus-4-8-low",
+	"claude-opus-4-8-thinking",
+}
+
+var ChannelName = "claude"

relay/channel/claude/dto.go

@@ -0,0 +1,95 @@
+package claude
+
+//
+//type ClaudeMetadata struct {
+//	UserId string `json:"user_id"`
+//}
+//
+//type ClaudeMediaMessage struct {
+//	Type        string               `json:"type"`
+//	Text        string               `json:"text,omitempty"`
+//	Source      *ClaudeMessageSource `json:"source,omitempty"`
+//	Usage       *ClaudeUsage         `json:"usage,omitempty"`
+//	StopReason  *string              `json:"stop_reason,omitempty"`
+//	PartialJson string               `json:"partial_json,omitempty"`
+//	Thinking    string               `json:"thinking,omitempty"`
+//	Signature   string               `json:"signature,omitempty"`
+//	Delta       string               `json:"delta,omitempty"`
+//	// tool_calls
+//	Id        string `json:"id,omitempty"`
+//	Name      string `json:"name,omitempty"`
+//	Input     any    `json:"input,omitempty"`
+//	Content   string `json:"content,omitempty"`
+//	ToolUseId string `json:"tool_use_id,omitempty"`
+//}
+//
+//type ClaudeMessageSource struct {
+//	Type      string `json:"type"`
+//	MediaType string `json:"media_type"`
+//	Data      string `json:"data"`
+//}
+//
+//type ClaudeMessage struct {
+//	Role    string `json:"role"`
+//	Content any    `json:"content"`
+//}
+//
+//type Tool struct {
+//	Name        string                 `json:"name"`
+//	Description string                 `json:"description,omitempty"`
+//	InputSchema map[string]interface{} `json:"input_schema"`
+//}
+//
+//type InputSchema struct {
+//	Type       string `json:"type"`
+//	Properties any    `json:"properties,omitempty"`
+//	Required   any    `json:"required,omitempty"`
+//}
+//
+//type ClaudeRequest struct {
+//	Model             string          `json:"model"`
+//	Prompt            string          `json:"prompt,omitempty"`
+//	System            string          `json:"system,omitempty"`
+//	Messages          []ClaudeMessage `json:"messages,omitempty"`
+//	MaxTokens         uint            `json:"max_tokens,omitempty"`
+//	MaxTokensToSample uint            `json:"max_tokens_to_sample,omitempty"`
+//	StopSequences     []string        `json:"stop_sequences,omitempty"`
+//	Temperature       *float64        `json:"temperature,omitempty"`
+//	TopP              float64         `json:"top_p,omitempty"`
+//	TopK              int             `json:"top_k,omitempty"`
+//	//ClaudeMetadata    `json:"metadata,omitempty"`
+//	Stream     bool      `json:"stream,omitempty"`
+//	Tools      any       `json:"tools,omitempty"`
+//	ToolChoice any       `json:"tool_choice,omitempty"`
+//	Thinking   *Thinking `json:"thinking,omitempty"`
+//}
+//
+//type Thinking struct {
+//	Type         string `json:"type"`
+//	BudgetTokens int    `json:"budget_tokens"`
+//}
+//
+//type ClaudeError struct {
+//	Type    string `json:"type"`
+//	Message string `json:"message"`
+//}
+//
+//type ClaudeResponse struct {
+//	Id           string               `json:"id"`
+//	Type         string               `json:"type"`
+//	Content      []ClaudeMediaMessage `json:"content"`
+//	Completion   string               `json:"completion"`
+//	StopReason   string               `json:"stop_reason"`
+//	Model        string               `json:"model"`
+//	Error        ClaudeError          `json:"error"`
+//	Usage        ClaudeUsage          `json:"usage"`
+//	Index        int                  `json:"index"` // stream only
+//	ContentBlock *ClaudeMediaMessage  `json:"content_block"`
+//	Delta        *ClaudeMediaMessage  `json:"delta"`   // stream only
+//	Message      *ClaudeResponse      `json:"message"` // stream only: message_start
+//}
+//
+//type ClaudeUsage struct {
+//	InputTokens  int `json:"input_tokens"`
+//	OutputTokens int `json:"output_tokens"`
+//}

relay/channel/claude/message_delta_usage_patch_test.go

@@ -0,0 +1,129 @@
+package claude
+
+import (
+	"testing"
+
+	relaycommon "github.com/QuantumNous/new-api/relay/common"
+	"github.com/QuantumNous/new-api/relaykit/dto"
+	"github.com/QuantumNous/new-api/setting/model_setting"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"github.com/tidwall/gjson"
+)
+
+func TestPatchClaudeMessageDeltaUsageDataPreserveUnknownFields(t *testing.T) {
+	originalData := `{"type":"message_delta","delta":{"stop_reason":"end_turn","stop_sequence":null},"usage":{"output_tokens":53},"vendor_meta":{"trace_id":"trace_001"}}`
+	usage := &dto.ClaudeUsage{
+		InputTokens:              100,
+		CacheReadInputTokens:     30,
+		CacheCreationInputTokens: 50,
+	}
+
+	patchedData := patchClaudeMessageDeltaUsageData(originalData, usage)
+
+	require.Equal(t, "message_delta", gjson.Get(patchedData, "type").String())
+	require.Equal(t, "end_turn", gjson.Get(patchedData, "delta.stop_reason").String())
+	require.Equal(t, "trace_001", gjson.Get(patchedData, "vendor_meta.trace_id").String())
+	require.EqualValues(t, 53, gjson.Get(patchedData, "usage.output_tokens").Int())
+	require.EqualValues(t, 100, gjson.Get(patchedData, "usage.input_tokens").Int())
+	require.EqualValues(t, 30, gjson.Get(patchedData, "usage.cache_read_input_tokens").Int())
+	require.EqualValues(t, 50, gjson.Get(patchedData, "usage.cache_creation_input_tokens").Int())
+}
+
+func TestPatchClaudeMessageDeltaUsageDataZeroValueChecks(t *testing.T) {
+	originalData := `{"type":"message_delta","usage":{"output_tokens":53,"input_tokens":9,"cache_read_input_tokens":0}}`
+	usage := &dto.ClaudeUsage{
+		InputTokens:              100,
+		CacheReadInputTokens:     30,
+		CacheCreationInputTokens: 0,
+	}
+
+	patchedData := patchClaudeMessageDeltaUsageData(originalData, usage)
+
+	require.EqualValues(t, 9, gjson.Get(patchedData, "usage.input_tokens").Int())
+	require.EqualValues(t, 30, gjson.Get(patchedData, "usage.cache_read_input_tokens").Int())
+	assert.False(t, gjson.Get(patchedData, "usage.cache_creation_input_tokens").Exists())
+}
+
+func TestShouldSkipClaudeMessageDeltaUsagePatch(t *testing.T) {
+	originGlobalPassThrough := model_setting.GetGlobalSettings().PassThroughRequestEnabled
+	t.Cleanup(func() {
+		model_setting.GetGlobalSettings().PassThroughRequestEnabled = originGlobalPassThrough
+	})
+
+	model_setting.GetGlobalSettings().PassThroughRequestEnabled = true
+	assert.True(t, shouldSkipClaudeMessageDeltaUsagePatch(&relaycommon.RelayInfo{}))
+
+	model_setting.GetGlobalSettings().PassThroughRequestEnabled = false
+	assert.True(t, shouldSkipClaudeMessageDeltaUsagePatch(&relaycommon.RelayInfo{
+		ChannelMeta: &relaycommon.ChannelMeta{ChannelSetting: dto.ChannelSettings{PassThroughBodyEnabled: true}},
+	}))
+	assert.False(t, shouldSkipClaudeMessageDeltaUsagePatch(&relaycommon.RelayInfo{
+		ChannelMeta: &relaycommon.ChannelMeta{ChannelSetting: dto.ChannelSettings{PassThroughBodyEnabled: false}},
+	}))
+}
+
+func TestBuildMessageDeltaPatchUsage(t *testing.T) {
+	t.Run("merge missing fields from claudeInfo", func(t *testing.T) {
+		claudeResponse := &dto.ClaudeResponse{Usage: &dto.ClaudeUsage{OutputTokens: 53}}
+		claudeInfo := &ClaudeResponseInfo{
+			Usage: &dto.Usage{
+				PromptTokens: 100,
+				PromptTokensDetails: dto.InputTokenDetails{
+					CachedTokens:         30,
+					CachedCreationTokens: 50,
+				},
+				ClaudeCacheCreation5mTokens: 10,
+				ClaudeCacheCreation1hTokens: 20,
+			},
+		}
+
+		usage := buildMessageDeltaPatchUsage(claudeResponse, claudeInfo)
+		require.NotNil(t, usage)
+		require.EqualValues(t, 100, usage.InputTokens)
+		require.EqualValues(t, 30, usage.CacheReadInputTokens)
+		require.EqualValues(t, 50, usage.CacheCreationInputTokens)
+		require.EqualValues(t, 53, usage.OutputTokens)
+		require.NotNil(t, usage.CacheCreation)
+		require.EqualValues(t, 30, usage.CacheCreation.Ephemeral5mInputTokens)
+		require.EqualValues(t, 20, usage.CacheCreation.Ephemeral1hInputTokens)
+	})
+
+	t.Run("keep upstream non-zero values", func(t *testing.T) {
+		claudeResponse := &dto.ClaudeResponse{Usage: &dto.ClaudeUsage{
+			InputTokens:              9,
+			CacheReadInputTokens:     7,
+			CacheCreationInputTokens: 6,
+		}}
+		claudeInfo := &ClaudeResponseInfo{Usage: &dto.Usage{
+			PromptTokens: 100,
+			PromptTokensDetails: dto.InputTokenDetails{
+				CachedTokens:         30,
+				CachedCreationTokens: 50,
+			},
+		}}
+
+		usage := buildMessageDeltaPatchUsage(claudeResponse, claudeInfo)
+		require.EqualValues(t, 9, usage.InputTokens)
+		require.EqualValues(t, 7, usage.CacheReadInputTokens)
+		require.EqualValues(t, 6, usage.CacheCreationInputTokens)
+	})
+
+	t.Run("default aggregate cache creation to 5m when split missing", func(t *testing.T) {
+		claudeResponse := &dto.ClaudeResponse{Usage: &dto.ClaudeUsage{
+			OutputTokens:             53,
+			CacheCreationInputTokens: 50,
+		}}
+		claudeInfo := &ClaudeResponseInfo{Usage: &dto.Usage{
+			PromptTokensDetails: dto.InputTokenDetails{
+				CachedCreationTokens: 50,
+			},
+		}}
+
+		usage := buildMessageDeltaPatchUsage(claudeResponse, claudeInfo)
+		require.NotNil(t, usage)
+		require.NotNil(t, usage.CacheCreation)
+		require.EqualValues(t, 50, usage.CacheCreation.Ephemeral5mInputTokens)
+		require.EqualValues(t, 0, usage.CacheCreation.Ephemeral1hInputTokens)
+	})
+}

relay/channel/claude/relay-claude.go

@@ -0,0 +1,415 @@
+package claude
+
+import (
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+
+	"github.com/QuantumNous/new-api/common"
+	"github.com/QuantumNous/new-api/constant"
+	"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/relaykit/dto"
+	"github.com/QuantumNous/new-api/relaykit/relayconvert"
+	"github.com/QuantumNous/new-api/relaykit/types"
+	"github.com/QuantumNous/new-api/service"
+	"github.com/QuantumNous/new-api/setting/model_setting"
+
+	"github.com/gin-gonic/gin"
+)
+
+func stopReasonClaude2OpenAI(reason string) string {
+	return relayconvert.StopReasonClaudeToOpenAI(reason)
+}
+
+func maybeMarkClaudeRefusal(c *gin.Context, stopReason string) {
+	if c == nil {
+		return
+	}
+	if strings.EqualFold(stopReason, "refusal") {
+		common.SetContextKey(c, constant.ContextKeyAdminRejectReason, "claude_stop_reason=refusal")
+	}
+}
+
+func StreamResponseClaude2OpenAI(claudeResponse *dto.ClaudeResponse) *dto.ChatCompletionsStreamResponse {
+	return relayconvert.StreamResponseClaude2OpenAI(claudeResponse)
+}
+
+func ResponseClaude2OpenAI(claudeResponse *dto.ClaudeResponse) *dto.OpenAITextResponse {
+	return relayconvert.ResponseClaude2OpenAI(claudeResponse)
+}
+
+type ClaudeResponseInfo = relayconvert.ClaudeResponseInfo
+
+func cacheCreationTokensForOpenAIUsage(usage *dto.Usage) int {
+	if usage == nil {
+		return 0
+	}
+	openAIUsage := relayconvert.UsageFromClaudeUsage(usage)
+	if openAIUsage == nil {
+		return 0
+	}
+	return openAIUsage.PromptTokens - usage.PromptTokens - usage.PromptTokensDetails.CachedTokens
+}
+
+func buildOpenAIStyleUsageFromClaudeUsage(usage *dto.Usage) dto.Usage {
+	mapped := relayconvert.UsageFromClaudeUsage(usage)
+	if mapped == nil {
+		return dto.Usage{}
+	}
+	return *mapped
+}
+
+func buildMessageDeltaPatchUsage(claudeResponse *dto.ClaudeResponse, claudeInfo *ClaudeResponseInfo) *dto.ClaudeUsage {
+	return relayconvert.BuildMessageDeltaPatchUsage(claudeResponse, claudeInfo)
+}
+
+func shouldSkipClaudeMessageDeltaUsagePatch(info *relaycommon.RelayInfo) bool {
+	if model_setting.GetGlobalSettings().PassThroughRequestEnabled {
+		return true
+	}
+	if info == nil {
+		return false
+	}
+	return info.ChannelSetting.PassThroughBodyEnabled
+}
+
+func patchClaudeMessageDeltaUsageData(data string, usage *dto.ClaudeUsage) string {
+	return relayconvert.PatchClaudeMessageDeltaUsageData(data, usage)
+}
+
+func FormatClaudeResponseInfo(claudeResponse *dto.ClaudeResponse, oaiResponse *dto.ChatCompletionsStreamResponse, claudeInfo *ClaudeResponseInfo) bool {
+	return relayconvert.FormatClaudeResponseInfo(claudeResponse, oaiResponse, claudeInfo)
+}
+
+func HandleStreamResponseData(c *gin.Context, info *relaycommon.RelayInfo, claudeInfo *ClaudeResponseInfo, data string) *types.NewAPIError {
+	var claudeResponse dto.ClaudeResponse
+	err := common.UnmarshalJsonStr(data, &claudeResponse)
+	if err != nil {
+		common.SysLog("error unmarshalling stream response: " + err.Error())
+		return types.NewError(err, types.ErrorCodeBadResponseBody)
+	}
+	if claudeError := claudeResponse.GetClaudeError(); claudeError != nil && claudeError.Type != "" {
+		return types.WithClaudeError(*claudeError, http.StatusInternalServerError)
+	}
+	if claudeResponse.StopReason != "" {
+		maybeMarkClaudeRefusal(c, claudeResponse.StopReason)
+	}
+	if claudeResponse.Delta != nil && claudeResponse.Delta.StopReason != nil {
+		maybeMarkClaudeRefusal(c, *claudeResponse.Delta.StopReason)
+	}
+	if info.RelayFormat == types.RelayFormatClaude {
+		FormatClaudeResponseInfo(&claudeResponse, nil, claudeInfo)
+
+		if claudeResponse.Type == "message_start" {
+			// message_start, 获取usage
+			if claudeResponse.Message != nil {
+				info.UpstreamModelName = claudeResponse.Message.Model
+			}
+		} else if claudeResponse.Type == "message_delta" {
+			// 确保 message_delta 的 usage 包含完整的 input_tokens 和 cache 相关字段
+			// 解决 AWS Bedrock 等上游返回的 message_delta 缺少这些字段的问题
+			if !shouldSkipClaudeMessageDeltaUsagePatch(info) {
+				data = patchClaudeMessageDeltaUsageData(data, buildMessageDeltaPatchUsage(&claudeResponse, claudeInfo))
+			}
+		}
+		countClaudeStreamBillableTools(c, info, &claudeResponse)
+		helper.ClaudeChunkData(c, claudeResponse, data)
+	} else if info.RelayFormat == types.RelayFormatOpenAI {
+		state, err := claudeToChatStreamState(info)
+		if err != nil {
+			return types.NewError(err, types.ErrorCodeBadResponseBody)
+		}
+		response, err := state.ConvertChunk(&claudeResponse)
+		if err != nil {
+			return types.NewError(err, types.ErrorCodeBadResponseBody)
+		}
+
+		if !FormatClaudeResponseInfo(&claudeResponse, response, claudeInfo) {
+			return nil
+		}
+
+		countClaudeStreamBillableTools(c, info, &claudeResponse)
+
+		if response == nil {
+			return nil
+		}
+		err = helper.ObjectData(c, response)
+		if err != nil {
+			logger.LogError(c, "send_stream_response_failed: "+err.Error())
+		}
+	} else if info.RelayFormat == types.RelayFormatGemini {
+		state, err := claudeToGeminiStreamState(info)
+		if err != nil {
+			return types.NewError(err, types.ErrorCodeBadResponseBody)
+		}
+		results, err := service.ConvertStreamResponseChunk(c, info, state, &claudeResponse)
+		if err != nil {
+			return types.NewError(err, types.ErrorCodeBadResponseBody)
+		}
+		if !FormatClaudeResponseInfo(&claudeResponse, nil, claudeInfo) {
+			return nil
+		}
+		countClaudeStreamBillableTools(c, info, &claudeResponse)
+		if sendErr := sendGeminiStreamResults(c, results); sendErr != nil {
+			return sendErr
+		}
+	}
+	return nil
+}
+
+func claudeToChatStreamState(info *relaycommon.RelayInfo) (*relayconvert.ClaudeToChatStreamState, error) {
+	if info != nil && info.ClaudeToChatStreamState != nil {
+		state, ok := info.ClaudeToChatStreamState.(*relayconvert.ClaudeToChatStreamState)
+		if !ok || state == nil {
+			return nil, fmt.Errorf("invalid Claude-to-Chat stream state %T", info.ClaudeToChatStreamState)
+		}
+		return state, nil
+	}
+
+	state := relayconvert.NewClaudeToChatStreamState()
+	if info != nil {
+		info.ClaudeToChatStreamState = state
+	}
+	return state, nil
+}
+
+func claudeToGeminiStreamState(info *relaycommon.RelayInfo) (*relayconvert.ResponseStreamState, error) {
+	if info != nil && info.ChatToGeminiStreamState != nil {
+		state, ok := info.ChatToGeminiStreamState.(*relayconvert.ResponseStreamState)
+		if !ok || state == nil {
+			return nil, fmt.Errorf("invalid Claude-to-Gemini stream state %T", info.ChatToGeminiStreamState)
+		}
+		return state, nil
+	}
+
+	state, err := relayconvert.NewResponseStreamState(types.RelayFormatClaude, types.RelayFormatGemini, relayconvert.ResponseStreamOptions{})
+	if err != nil {
+		return nil, err
+	}
+	if info != nil {
+		info.ChatToGeminiStreamState = state
+	}
+	return state, nil
+}
+
+func sendGeminiStreamResults(c *gin.Context, results []relayconvert.ResponseResult) *types.NewAPIError {
+	for _, result := range results {
+		geminiResponse, ok := result.Value.(*dto.GeminiChatResponse)
+		if !ok {
+			return types.NewError(fmt.Errorf("expected Gemini stream response, got %T", result.Value), types.ErrorCodeBadResponseBody)
+		}
+		if geminiResponse == nil {
+			continue
+		}
+		data, err := common.Marshal(geminiResponse)
+		if err != nil {
+			return types.NewError(err, types.ErrorCodeBadResponseBody)
+		}
+		c.Render(-1, common.CustomEvent{Data: "data: " + string(data)})
+		_ = helper.FlushWriter(c)
+	}
+	return nil
+}
+
+func countClaudeStreamBillableTools(c *gin.Context, info *relaycommon.RelayInfo, claudeResponse *dto.ClaudeResponse) {
+	if claudeResponse == nil {
+		return
+	}
+	if claudeResponse.Type == "content_block_start" &&
+		claudeResponse.ContentBlock != nil &&
+		claudeResponse.ContentBlock.Type == "tool_use" {
+		info.CountBillableToolCall(dto.BuildInCallToolUse, claudeResponse.ContentBlock.Name)
+	}
+	if claudeResponse.Type == "message_delta" &&
+		claudeResponse.Usage != nil &&
+		claudeResponse.Usage.ServerToolUse != nil &&
+		claudeResponse.Usage.ServerToolUse.WebSearchRequests > 0 {
+		c.Set("claude_web_search_requests", claudeResponse.Usage.ServerToolUse.WebSearchRequests)
+	}
+}
+
+func HandleStreamFinalResponse(c *gin.Context, info *relaycommon.RelayInfo, claudeInfo *ClaudeResponseInfo) {
+	if claudeInfo.Usage.PromptTokens == 0 {
+		//上游出错
+	}
+	if claudeInfo.Usage.CompletionTokens == 0 || !claudeInfo.Done {
+		if common.DebugEnabled {
+			common.SysLog("claude response usage is not complete, maybe upstream error")
+		}
+		// 只补缺失字段，不整份覆盖——保留 message_start 已拿到的 cache 字段
+		fallback := service.ResponseText2Usage(c, claudeInfo.ResponseText.String(), info.UpstreamModelName, info.GetEstimatePromptTokens())
+		if claudeInfo.Usage.CompletionTokens == 0 ||
+			(!claudeInfo.Done && fallback.CompletionTokens > claudeInfo.Usage.CompletionTokens) {
+			claudeInfo.Usage.CompletionTokens = fallback.CompletionTokens
+		}
+		if claudeInfo.Usage.PromptTokens == 0 {
+			claudeInfo.Usage.PromptTokens = fallback.PromptTokens
+		}
+		claudeInfo.Usage.TotalTokens = claudeInfo.Usage.PromptTokens + claudeInfo.Usage.CompletionTokens
+	}
+	if claudeInfo.Usage != nil {
+		claudeInfo.Usage.UsageSemantic = "anthropic"
+	}
+	relayconvert.FinalizeClaudeStreamBillingUsage(claudeInfo)
+
+	if info.RelayFormat == types.RelayFormatClaude {
+		//
+	} else if info.RelayFormat == types.RelayFormatOpenAI {
+		if info.ShouldIncludeUsage {
+			openAIUsage := buildOpenAIStyleUsageFromClaudeUsage(claudeInfo.Usage)
+			response := helper.GenerateFinalUsageResponse(claudeInfo.ResponseId, claudeInfo.Created, info.UpstreamModelName, openAIUsage)
+			err := helper.ObjectData(c, response)
+			if err != nil {
+				common.SysLog("send final response failed: " + err.Error())
+			}
+		}
+		helper.Done(c)
+	} else if info.RelayFormat == types.RelayFormatGemini {
+		state, err := claudeToGeminiStreamState(info)
+		if err != nil {
+			common.SysLog("error creating Gemini stream state: " + err.Error())
+			return
+		}
+		results, err := service.FinalizeStreamResponse(c, info, state)
+		if err != nil {
+			common.SysLog("error finalizing Gemini stream response: " + err.Error())
+			return
+		}
+		if sendErr := sendGeminiStreamResults(c, results); sendErr != nil {
+			common.SysLog("send final Gemini stream response failed: " + sendErr.Error())
+		}
+	}
+}
+
+func ClaudeStreamHandler(c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (*dto.Usage, *types.NewAPIError) {
+	claudeInfo := &ClaudeResponseInfo{
+		ResponseId:   helper.GetResponseID(c),
+		Created:      common.GetTimestamp(),
+		Model:        info.UpstreamModelName,
+		ResponseText: strings.Builder{},
+		Usage:        &dto.Usage{},
+	}
+	var err *types.NewAPIError
+	helper.StreamScannerHandler(c, resp, info, func(data string, sr *helper.StreamResult) {
+		err = HandleStreamResponseData(c, info, claudeInfo, data)
+		if err != nil {
+			sr.Stop(err)
+		}
+	})
+	if err != nil {
+		return nil, err
+	}
+
+	HandleStreamFinalResponse(c, info, claudeInfo)
+	return claudeInfo.Usage, nil
+}
+
+func HandleClaudeResponseData(c *gin.Context, info *relaycommon.RelayInfo, claudeInfo *ClaudeResponseInfo, httpResp *http.Response, data []byte) *types.NewAPIError {
+	var claudeResponse dto.ClaudeResponse
+	err := common.Unmarshal(data, &claudeResponse)
+	if err != nil {
+		return types.NewError(err, types.ErrorCodeBadResponseBody)
+	}
+	if claudeError := claudeResponse.GetClaudeError(); claudeError != nil && claudeError.Type != "" {
+		return types.WithClaudeError(*claudeError, http.StatusInternalServerError)
+	}
+	maybeMarkClaudeRefusal(c, claudeResponse.StopReason)
+	if claudeInfo.Usage == nil {
+		claudeInfo.Usage = &dto.Usage{}
+	}
+	if claudeResponse.Usage != nil {
+		claudeInfo.Usage.PromptTokens = claudeResponse.Usage.InputTokens
+		claudeInfo.Usage.CompletionTokens = claudeResponse.Usage.OutputTokens
+		claudeInfo.Usage.TotalTokens = claudeResponse.Usage.InputTokens + claudeResponse.Usage.OutputTokens
+		claudeInfo.Usage.UsageSemantic = "anthropic"
+		claudeInfo.Usage.BillingUsage = dto.CloneBillingUsage(claudeResponse.Usage.BillingUsage)
+		if claudeInfo.Usage.BillingUsage == nil {
+			claudeInfo.Usage.BillingUsage = dto.NewClaudeMessagesBillingUsage(claudeResponse.Usage)
+		}
+		claudeInfo.Usage.PromptTokensDetails.CachedTokens = claudeResponse.Usage.CacheReadInputTokens
+		claudeInfo.Usage.PromptTokensDetails.CachedCreationTokens = claudeResponse.Usage.CacheCreationInputTokens
+		claudeInfo.Usage.ClaudeCacheCreation5mTokens = claudeResponse.Usage.GetCacheCreation5mTokens()
+		claudeInfo.Usage.ClaudeCacheCreation1hTokens = claudeResponse.Usage.GetCacheCreation1hTokens()
+	}
+	var responseData []byte
+	switch info.RelayFormat {
+	case types.RelayFormatOpenAI:
+		openaiResponse := ResponseClaude2OpenAI(&claudeResponse)
+		openaiResponse.Usage = buildOpenAIStyleUsageFromClaudeUsage(claudeInfo.Usage)
+		responseData, err = common.Marshal(openaiResponse)
+		if err != nil {
+			return types.NewError(err, types.ErrorCodeBadResponseBody)
+		}
+	case types.RelayFormatOpenAIResponses:
+		convertResult, err := service.ConvertResponse(c, info, types.RelayFormatOpenAIResponses, &claudeResponse)
+		if err != nil {
+			return types.NewError(err, types.ErrorCodeBadResponseBody)
+		}
+		responsesResponse, ok := convertResult.Value.(*dto.OpenAIResponsesResponse)
+		if !ok {
+			return types.NewError(fmt.Errorf("expected OpenAI Responses response, got %T", convertResult.Value), types.ErrorCodeBadResponseBody)
+		}
+		if responseID := helper.GetResponseID(c); responseID != "" {
+			responsesResponse.ID = responseID
+		}
+		responseData, err = common.Marshal(responsesResponse)
+		if err != nil {
+			return types.NewError(err, types.ErrorCodeBadResponseBody)
+		}
+	case types.RelayFormatClaude:
+		responseData = data
+	case types.RelayFormatGemini:
+		{
+			convertResult, convertErr := service.ConvertResponse(c, info, types.RelayFormatGemini, &claudeResponse)
+			if convertErr != nil {
+				return types.NewError(convertErr, types.ErrorCodeBadResponseBody)
+			}
+			geminiResponse, ok := convertResult.Value.(*dto.GeminiChatResponse)
+			if !ok {
+				return types.NewError(fmt.Errorf("expected Gemini generateContent response, got %T", convertResult.Value), types.ErrorCodeBadResponseBody)
+			}
+			responseData, err = common.Marshal(geminiResponse)
+			if err != nil {
+				return types.NewError(err, types.ErrorCodeBadResponseBody)
+			}
+		}
+	}
+
+	if claudeResponse.Usage != nil && claudeResponse.Usage.ServerToolUse != nil && claudeResponse.Usage.ServerToolUse.WebSearchRequests > 0 {
+		c.Set("claude_web_search_requests", claudeResponse.Usage.ServerToolUse.WebSearchRequests)
+	}
+
+	for _, block := range claudeResponse.Content {
+		if block.Type == "tool_use" {
+			info.CountBillableToolCall(dto.BuildInCallToolUse, block.Name)
+		}
+	}
+
+	service.IOCopyBytesGracefully(c, httpResp, responseData)
+	return nil
+}
+
+func ClaudeHandler(c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (*dto.Usage, *types.NewAPIError) {
+	defer service.CloseResponseBodyGracefully(resp)
+
+	claudeInfo := &ClaudeResponseInfo{
+		ResponseId:   helper.GetResponseID(c),
+		Created:      common.GetTimestamp(),
+		Model:        info.UpstreamModelName,
+		ResponseText: strings.Builder{},
+		Usage:        &dto.Usage{},
+	}
+	responseBody, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, types.NewError(err, types.ErrorCodeBadResponseBody)
+	}
+	logger.LogDebug(c, "responseBody: %s", responseBody)
+	handleErr := HandleClaudeResponseData(c, info, claudeInfo, resp, responseBody)
+	if handleErr != nil {
+		return nil, handleErr
+	}
+	return claudeInfo.Usage, nil
+}

relay/channel/claude/relay_claude_test.go

@@ -0,0 +1,402 @@
+package claude
+
+import (
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"github.com/QuantumNous/new-api/common"
+	relaycommon "github.com/QuantumNous/new-api/relay/common"
+	"github.com/QuantumNous/new-api/relay/helper"
+	"github.com/QuantumNous/new-api/relaykit/dto"
+	"github.com/QuantumNous/new-api/relaykit/relayconvert"
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func commonPointer[T any](value T) *T {
+	return &value
+}
+
+func TestResponseOpenAI2ClaudeToolUseInputIsObject(t *testing.T) {
+	tests := []struct {
+		name string
+		args string
+		want map[string]any
+	}{
+		{name: "object", args: `{"q":"x"}`, want: map[string]any{"q": "x"}},
+		{name: "empty", args: "", want: map[string]any{}},
+		{name: "invalid", args: "{", want: map[string]any{}},
+		{name: "null", args: "null", want: map[string]any{}},
+		{name: "array", args: `["x"]`, want: map[string]any{}},
+		{name: "string", args: `"x"`, want: map[string]any{}},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			msg := dto.Message{Role: "assistant"}
+			msg.SetToolCalls([]dto.ToolCallRequest{
+				{
+					ID:   "call_1",
+					Type: "function",
+					Function: dto.FunctionRequest{
+						Name:      "lookup",
+						Arguments: tt.args,
+					},
+				},
+			})
+			resp := relayconvert.ResponseOpenAI2Claude(&dto.OpenAITextResponse{
+				Id:    "chatcmpl_1",
+				Model: "gpt-test",
+				Choices: []dto.OpenAITextResponseChoice{
+					{Message: msg, FinishReason: "tool_calls"},
+				},
+			}, nil)
+
+			require.Len(t, resp.Content, 1)
+			assert.Equal(t, "tool_use", resp.Content[0].Type)
+			assert.Equal(t, tt.want, resp.Content[0].Input)
+		})
+	}
+}
+
+func TestFormatClaudeResponseInfo_MessageStart(t *testing.T) {
+	claudeInfo := &ClaudeResponseInfo{
+		Usage: &dto.Usage{},
+	}
+	claudeResponse := &dto.ClaudeResponse{
+		Type: "message_start",
+		Message: &dto.ClaudeMediaMessage{
+			Id:    "msg_123",
+			Model: "claude-3-5-sonnet",
+			Usage: &dto.ClaudeUsage{
+				InputTokens:              100,
+				OutputTokens:             1,
+				CacheCreationInputTokens: 50,
+				CacheReadInputTokens:     30,
+			},
+		},
+	}
+
+	ok := FormatClaudeResponseInfo(claudeResponse, nil, claudeInfo)
+	if !ok {
+		t.Fatal("expected true")
+	}
+	if claudeInfo.Usage.PromptTokens != 100 {
+		t.Errorf("PromptTokens = %d, want 100", claudeInfo.Usage.PromptTokens)
+	}
+	if claudeInfo.Usage.PromptTokensDetails.CachedTokens != 30 {
+		t.Errorf("CachedTokens = %d, want 30", claudeInfo.Usage.PromptTokensDetails.CachedTokens)
+	}
+	if claudeInfo.Usage.PromptTokensDetails.CachedCreationTokens != 50 {
+		t.Errorf("CachedCreationTokens = %d, want 50", claudeInfo.Usage.PromptTokensDetails.CachedCreationTokens)
+	}
+	if claudeInfo.ResponseId != "msg_123" {
+		t.Errorf("ResponseId = %s, want msg_123", claudeInfo.ResponseId)
+	}
+	if claudeInfo.Model != "claude-3-5-sonnet" {
+		t.Errorf("Model = %s, want claude-3-5-sonnet", claudeInfo.Model)
+	}
+}
+
+func TestFormatClaudeResponseInfo_MessageDelta_FullUsage(t *testing.T) {
+	// message_start 先积累 usage
+	claudeInfo := &ClaudeResponseInfo{
+		Usage: &dto.Usage{
+			PromptTokens: 100,
+			PromptTokensDetails: dto.InputTokenDetails{
+				CachedTokens:         30,
+				CachedCreationTokens: 50,
+			},
+			CompletionTokens: 1,
+		},
+	}
+
+	// message_delta 带完整 usage（原生 Anthropic 场景）
+	claudeResponse := &dto.ClaudeResponse{
+		Type: "message_delta",
+		Usage: &dto.ClaudeUsage{
+			InputTokens:              100,
+			OutputTokens:             200,
+			CacheCreationInputTokens: 50,
+			CacheReadInputTokens:     30,
+		},
+	}
+
+	ok := FormatClaudeResponseInfo(claudeResponse, nil, claudeInfo)
+	if !ok {
+		t.Fatal("expected true")
+	}
+	if claudeInfo.Usage.PromptTokens != 100 {
+		t.Errorf("PromptTokens = %d, want 100", claudeInfo.Usage.PromptTokens)
+	}
+	if claudeInfo.Usage.CompletionTokens != 200 {
+		t.Errorf("CompletionTokens = %d, want 200", claudeInfo.Usage.CompletionTokens)
+	}
+	if claudeInfo.Usage.TotalTokens != 300 {
+		t.Errorf("TotalTokens = %d, want 300", claudeInfo.Usage.TotalTokens)
+	}
+	if !claudeInfo.Done {
+		t.Error("expected Done = true")
+	}
+}
+
+func TestFormatClaudeResponseInfo_MessageDelta_OnlyOutputTokens(t *testing.T) {
+	// 模拟 Bedrock: message_start 已积累 usage
+	claudeInfo := &ClaudeResponseInfo{
+		Usage: &dto.Usage{
+			PromptTokens: 100,
+			PromptTokensDetails: dto.InputTokenDetails{
+				CachedTokens:         30,
+				CachedCreationTokens: 50,
+			},
+			CompletionTokens:            1,
+			ClaudeCacheCreation5mTokens: 10,
+			ClaudeCacheCreation1hTokens: 20,
+		},
+	}
+
+	// Bedrock 的 message_delta 只有 output_tokens，缺少 input_tokens 和 cache 字段
+	claudeResponse := &dto.ClaudeResponse{
+		Type: "message_delta",
+		Usage: &dto.ClaudeUsage{
+			OutputTokens: 200,
+			// InputTokens, CacheCreationInputTokens, CacheReadInputTokens 都是 0
+		},
+	}
+
+	ok := FormatClaudeResponseInfo(claudeResponse, nil, claudeInfo)
+	if !ok {
+		t.Fatal("expected true")
+	}
+	// PromptTokens 应保持 message_start 的值（因为 message_delta 的 InputTokens=0，不更新）
+	if claudeInfo.Usage.PromptTokens != 100 {
+		t.Errorf("PromptTokens = %d, want 100", claudeInfo.Usage.PromptTokens)
+	}
+	if claudeInfo.Usage.CompletionTokens != 200 {
+		t.Errorf("CompletionTokens = %d, want 200", claudeInfo.Usage.CompletionTokens)
+	}
+	if claudeInfo.Usage.TotalTokens != 300 {
+		t.Errorf("TotalTokens = %d, want 300", claudeInfo.Usage.TotalTokens)
+	}
+	// cache 字段应保持 message_start 的值
+	if claudeInfo.Usage.PromptTokensDetails.CachedTokens != 30 {
+		t.Errorf("CachedTokens = %d, want 30", claudeInfo.Usage.PromptTokensDetails.CachedTokens)
+	}
+	if claudeInfo.Usage.PromptTokensDetails.CachedCreationTokens != 50 {
+		t.Errorf("CachedCreationTokens = %d, want 50", claudeInfo.Usage.PromptTokensDetails.CachedCreationTokens)
+	}
+	if claudeInfo.Usage.ClaudeCacheCreation5mTokens != 10 {
+		t.Errorf("ClaudeCacheCreation5mTokens = %d, want 10", claudeInfo.Usage.ClaudeCacheCreation5mTokens)
+	}
+	if claudeInfo.Usage.ClaudeCacheCreation1hTokens != 20 {
+		t.Errorf("ClaudeCacheCreation1hTokens = %d, want 20", claudeInfo.Usage.ClaudeCacheCreation1hTokens)
+	}
+	if !claudeInfo.Done {
+		t.Error("expected Done = true")
+	}
+}
+
+func TestFormatClaudeResponseInfo_NilClaudeInfo(t *testing.T) {
+	claudeResponse := &dto.ClaudeResponse{Type: "message_start"}
+	ok := FormatClaudeResponseInfo(claudeResponse, nil, nil)
+	if ok {
+		t.Error("expected false for nil claudeInfo")
+	}
+}
+
+func TestFormatClaudeResponseInfo_ContentBlockDelta(t *testing.T) {
+	text := "hello"
+	claudeInfo := &ClaudeResponseInfo{
+		Usage:        &dto.Usage{},
+		ResponseText: strings.Builder{},
+	}
+	claudeResponse := &dto.ClaudeResponse{
+		Type: "content_block_delta",
+		Delta: &dto.ClaudeMediaMessage{
+			Text: &text,
+		},
+	}
+
+	ok := FormatClaudeResponseInfo(claudeResponse, nil, claudeInfo)
+	if !ok {
+		t.Fatal("expected true")
+	}
+	if claudeInfo.ResponseText.String() != "hello" {
+		t.Errorf("ResponseText = %q, want %q", claudeInfo.ResponseText.String(), "hello")
+	}
+}
+
+func TestBuildOpenAIStyleUsageFromClaudeUsage(t *testing.T) {
+	usage := &dto.Usage{
+		PromptTokens:     100,
+		CompletionTokens: 20,
+		PromptTokensDetails: dto.InputTokenDetails{
+			CachedTokens:         30,
+			CachedCreationTokens: 50,
+		},
+		ClaudeCacheCreation5mTokens: 10,
+		ClaudeCacheCreation1hTokens: 20,
+		UsageSemantic:               "anthropic",
+	}
+
+	openAIUsage := buildOpenAIStyleUsageFromClaudeUsage(usage)
+
+	if openAIUsage.PromptTokens != 180 {
+		t.Fatalf("PromptTokens = %d, want 180", openAIUsage.PromptTokens)
+	}
+	if openAIUsage.InputTokens != 180 {
+		t.Fatalf("InputTokens = %d, want 180", openAIUsage.InputTokens)
+	}
+	if openAIUsage.TotalTokens != 200 {
+		t.Fatalf("TotalTokens = %d, want 200", openAIUsage.TotalTokens)
+	}
+	if openAIUsage.UsageSemantic != "openai" {
+		t.Fatalf("UsageSemantic = %s, want openai", openAIUsage.UsageSemantic)
+	}
+	if openAIUsage.UsageSource != "anthropic" {
+		t.Fatalf("UsageSource = %s, want anthropic", openAIUsage.UsageSource)
+	}
+}
+
+func TestBuildOpenAIStyleUsageFromClaudeUsagePreservesCacheCreationRemainder(t *testing.T) {
+	tests := []struct {
+		name                    string
+		cachedCreationTokens    int
+		cacheCreationTokens5m   int
+		cacheCreationTokens1h   int
+		expectedTotalInputToken int
+	}{
+		{
+			name:                    "prefers aggregate when it includes remainder",
+			cachedCreationTokens:    50,
+			cacheCreationTokens5m:   10,
+			cacheCreationTokens1h:   20,
+			expectedTotalInputToken: 180,
+		},
+		{
+			name:                    "falls back to split tokens when aggregate missing",
+			cachedCreationTokens:    0,
+			cacheCreationTokens5m:   10,
+			cacheCreationTokens1h:   20,
+			expectedTotalInputToken: 160,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			usage := &dto.Usage{
+				PromptTokens:     100,
+				CompletionTokens: 20,
+				PromptTokensDetails: dto.InputTokenDetails{
+					CachedTokens:         30,
+					CachedCreationTokens: tt.cachedCreationTokens,
+				},
+				ClaudeCacheCreation5mTokens: tt.cacheCreationTokens5m,
+				ClaudeCacheCreation1hTokens: tt.cacheCreationTokens1h,
+				UsageSemantic:               "anthropic",
+			}
+
+			openAIUsage := buildOpenAIStyleUsageFromClaudeUsage(usage)
+
+			if openAIUsage.PromptTokens != tt.expectedTotalInputToken {
+				t.Fatalf("PromptTokens = %d, want %d", openAIUsage.PromptTokens, tt.expectedTotalInputToken)
+			}
+			if openAIUsage.InputTokens != tt.expectedTotalInputToken {
+				t.Fatalf("InputTokens = %d, want %d", openAIUsage.InputTokens, tt.expectedTotalInputToken)
+			}
+		})
+	}
+}
+
+func TestBuildOpenAIStyleUsageFromClaudeUsageDefaultsAggregateCacheCreationTo5m(t *testing.T) {
+	usage := &dto.Usage{
+		PromptTokens:     100,
+		CompletionTokens: 20,
+		PromptTokensDetails: dto.InputTokenDetails{
+			CachedTokens:         30,
+			CachedCreationTokens: 50,
+		},
+		UsageSemantic: "anthropic",
+	}
+
+	openAIUsage := buildOpenAIStyleUsageFromClaudeUsage(usage)
+
+	require.Equal(t, 50, openAIUsage.ClaudeCacheCreation5mTokens)
+	require.Equal(t, 0, openAIUsage.ClaudeCacheCreation1hTokens)
+}
+
+func applyOpenAIChatReasoningThroughHandlerOrder(t *testing.T, original dto.GeneralOpenAIRequest) (*dto.GeneralOpenAIRequest, *relaycommon.RelayInfo) {
+	t.Helper()
+	gin.SetMode(gin.TestMode)
+	c, _ := gin.CreateTestContext(httptest.NewRecorder())
+
+	info := &relaycommon.RelayInfo{
+		OriginModelName: original.Model,
+		Request:         &original,
+		ChannelMeta: &relaycommon.ChannelMeta{
+			UpstreamModelName: original.Model,
+		},
+	}
+	outbound, err := common.DeepCopy(&original)
+	require.NoError(t, err)
+	require.NoError(t, helper.ModelMappedHelper(c, info, outbound))
+	err = helper.ApplyReasoningModelSuffix(nil, info, outbound)
+	require.NoError(t, err)
+	return outbound, info
+}
+
+func TestOpenAIChatRequestToClaudeMessages_ClaudeOpus48HighUsesAdaptiveThinking(t *testing.T) {
+	original := dto.GeneralOpenAIRequest{
+		Model:       "claude-opus-4-8-high",
+		Temperature: commonPointer(0.7),
+		TopP:        commonPointer(0.9),
+		TopK:        commonPointer(40),
+		Messages: []dto.Message{
+			{
+				Role:    "user",
+				Content: "hello",
+			},
+		},
+	}
+
+	outbound, info := applyOpenAIChatReasoningThroughHandlerOrder(t, original)
+	claudeRequest, err := relayconvert.OpenAIChatRequestToClaudeMessages(nil, info, *outbound)
+	require.NoError(t, err)
+	require.Equal(t, "claude-opus-4-8", claudeRequest.Model)
+	require.NotNil(t, claudeRequest.Thinking)
+	require.Equal(t, "adaptive", claudeRequest.Thinking.Type)
+	require.Equal(t, "summarized", claudeRequest.Thinking.Display)
+	require.JSONEq(t, `{"effort":"high"}`, string(claudeRequest.OutputConfig))
+	require.Nil(t, claudeRequest.Temperature)
+	require.Nil(t, claudeRequest.TopP)
+	require.Nil(t, claudeRequest.TopK)
+}
+
+func TestOpenAIChatRequestToClaudeMessages_ClaudeOpus48ThinkingUsesAdaptiveHighEffort(t *testing.T) {
+	original := dto.GeneralOpenAIRequest{
+		Model:       "claude-opus-4-8-thinking",
+		Temperature: commonPointer(0.7),
+		TopP:        commonPointer(0.9),
+		TopK:        commonPointer(40),
+		Messages: []dto.Message{
+			{
+				Role:    "user",
+				Content: "hello",
+			},
+		},
+	}
+
+	outbound, info := applyOpenAIChatReasoningThroughHandlerOrder(t, original)
+	claudeRequest, err := relayconvert.OpenAIChatRequestToClaudeMessages(nil, info, *outbound)
+	require.NoError(t, err)
+	require.Equal(t, "claude-opus-4-8", claudeRequest.Model)
+	require.NotNil(t, claudeRequest.Thinking)
+	require.Equal(t, "adaptive", claudeRequest.Thinking.Type)
+	require.Equal(t, "summarized", claudeRequest.Thinking.Display)
+	require.JSONEq(t, `{"effort":"high"}`, string(claudeRequest.OutputConfig))
+	require.Nil(t, claudeRequest.Temperature)
+	require.Nil(t, claudeRequest.TopP)
+	require.Nil(t, claudeRequest.TopK)
+}

relay/channel/claude/relay_responses.go

@@ -0,0 +1,180 @@
+package claude
+
+import (
+	"fmt"
+	"net/http"
+	"strings"
+
+	"github.com/QuantumNous/new-api/common"
+	"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/relaykit/dto"
+	"github.com/QuantumNous/new-api/relaykit/relayconvert"
+	"github.com/QuantumNous/new-api/relaykit/types"
+	"github.com/QuantumNous/new-api/service"
+	"github.com/gin-gonic/gin"
+)
+
+func ClaudeResponsesStreamHandler(c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (*dto.Usage, *types.NewAPIError) {
+	responseID := helper.GetResponseID(c)
+	created := common.GetTimestamp()
+	state, err := relayconvert.NewResponseStreamState(types.RelayFormatClaude, types.RelayFormatOpenAIResponses, relayconvert.ResponseStreamOptions{
+		ID:                 responseID,
+		Model:              info.UpstreamModelName,
+		Created:            created,
+		EmitSequenceNumber: true,
+	})
+	if err != nil {
+		return nil, types.NewOpenAIError(err, types.ErrorCodeBadResponse, http.StatusInternalServerError)
+	}
+	hostedBridge := relayconvert.NewClaudeHostedStreamBridge()
+
+	claudeInfo := &ClaudeResponseInfo{
+		ResponseId:   responseID,
+		Created:      created,
+		Model:        info.UpstreamModelName,
+		ResponseText: strings.Builder{},
+		Usage:        &dto.Usage{},
+	}
+	var streamErr *types.NewAPIError
+	// streamFailed means a Responses-native terminal error was sent successfully.
+	// In that case the scanner stops without a transport error and the partial
+	// upstream usage remains billable.
+	streamFailed := false
+
+	sendResponsesEvent := func(eventType string, payload dto.ResponsesStreamResponse) bool {
+		payload.Type = eventType
+		data, err := common.Marshal(payload)
+		if err != nil {
+			streamErr = types.NewOpenAIError(err, types.ErrorCodeJsonMarshalFailed, http.StatusInternalServerError)
+			return false
+		}
+		if err := helper.ResponseChunkData(c, dto.ResponsesStreamResponse{Type: eventType}, string(data)); err != nil {
+			streamErr = types.NewOpenAIError(err, types.ErrorCodeBadResponse, http.StatusInternalServerError)
+			return false
+		}
+		return true
+	}
+	sendResult := func(result relayconvert.ResponseResult) bool {
+		event, ok := result.Value.(relayconvert.ChatToResponsesStreamEvent)
+		if !ok {
+			streamErr = types.NewOpenAIError(
+				fmt.Errorf("expected OpenAI Responses stream event, got %T", result.Value),
+				types.ErrorCodeBadResponse,
+				http.StatusInternalServerError,
+			)
+			return false
+		}
+		return sendResponsesEvent(event.Type, event.Payload)
+	}
+	failResponsesStream := func(err error) bool {
+		failureResults, handled := state.FailResponsesStream("server_error", err.Error(), "")
+		if !handled {
+			return false
+		}
+		for _, result := range failureResults {
+			if !sendResult(result) {
+				return true
+			}
+		}
+		streamFailed = true
+		return true
+	}
+
+	helper.StreamScannerHandler(c, resp, info, func(data string, sr *helper.StreamResult) {
+		var claudeResponse dto.ClaudeResponse
+		if err := common.UnmarshalJsonStr(data, &claudeResponse); err != nil {
+			logger.LogError(c, "failed to unmarshal Claude stream event: "+err.Error())
+			if failResponsesStream(err) {
+				// A nil streamErr here is intentional: the protocol-level failure
+				// event was delivered, so only the scanner needs to stop.
+				sr.Stop(streamErr)
+				return
+			}
+			streamErr = types.NewOpenAIError(err, types.ErrorCodeBadResponseBody, http.StatusInternalServerError)
+			sr.Stop(streamErr)
+			return
+		}
+		if claudeError := claudeResponse.GetClaudeError(); claudeError != nil && claudeError.Type != "" {
+			if failResponsesStream(fmt.Errorf("%s", claudeError.Message)) {
+				sr.Stop(streamErr)
+				return
+			}
+			streamErr = types.WithClaudeError(*claudeError, http.StatusInternalServerError)
+			sr.Stop(streamErr)
+			return
+		}
+
+		if claudeResponse.StopReason != "" {
+			maybeMarkClaudeRefusal(c, claudeResponse.StopReason)
+		}
+		if claudeResponse.Delta != nil && claudeResponse.Delta.StopReason != nil {
+			maybeMarkClaudeRefusal(c, *claudeResponse.Delta.StopReason)
+		}
+		if claudeResponse.Type == "message_start" && claudeResponse.Message != nil {
+			info.UpstreamModelName = claudeResponse.Message.Model
+		}
+		FormatClaudeResponseInfo(&claudeResponse, nil, claudeInfo)
+		countClaudeStreamBillableTools(c, info, &claudeResponse)
+		hostedEvents, consumed, err := hostedBridge.Convert(&claudeResponse, state)
+		if err != nil {
+			if failResponsesStream(err) {
+				sr.Stop(streamErr)
+				return
+			}
+			streamErr = types.NewOpenAIError(err, types.ErrorCodeBadResponse, http.StatusInternalServerError)
+			sr.Stop(streamErr)
+			return
+		}
+		for _, event := range hostedEvents {
+			if !sendResponsesEvent(event.Type, event.Payload) {
+				sr.Stop(streamErr)
+				return
+			}
+		}
+		if consumed {
+			return
+		}
+
+		results, err := service.ConvertStreamResponseChunk(c, info, state, &claudeResponse)
+		if err != nil {
+			if failResponsesStream(err) {
+				sr.Stop(streamErr)
+				return
+			}
+			streamErr = types.NewOpenAIError(err, types.ErrorCodeBadResponse, http.StatusInternalServerError)
+			sr.Stop(streamErr)
+			return
+		}
+		for _, result := range results {
+			if !sendResult(result) {
+				sr.Stop(streamErr)
+				return
+			}
+		}
+	})
+	if streamErr != nil {
+		return nil, streamErr
+	}
+	if streamFailed {
+		return claudeInfo.Usage, nil
+	}
+
+	HandleStreamFinalResponse(c, info, claudeInfo)
+	openAIUsage := buildOpenAIStyleUsageFromClaudeUsage(claudeInfo.Usage)
+	state.SetUsage(&openAIUsage)
+	finalResults, err := service.FinalizeStreamResponse(c, info, state)
+	if err != nil {
+		if failResponsesStream(err) {
+			return claudeInfo.Usage, streamErr
+		}
+		return nil, types.NewOpenAIError(err, types.ErrorCodeBadResponse, http.StatusInternalServerError)
+	}
+	for _, result := range finalResults {
+		if !sendResult(result) {
+			return nil, streamErr
+		}
+	}
+	return claudeInfo.Usage, nil
+}

relay/channel/claude/tool_billing_test.go

@@ -0,0 +1,76 @@
+package claude
+
+import (
+	"net/http/httptest"
+	"testing"
+
+	relaycommon "github.com/QuantumNous/new-api/relay/common"
+	"github.com/QuantumNous/new-api/relaykit/dto"
+	"github.com/QuantumNous/new-api/relaykit/types"
+	"github.com/QuantumNous/new-api/setting/operation_setting"
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestHandleClaudeResponseDataCountsToolUse(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	operation_setting.SetToolPriceForTest("lookup_fn", 3.0)
+	t.Cleanup(func() {
+		operation_setting.DeleteToolPriceForTest("lookup_fn")
+	})
+
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	info := &relaycommon.RelayInfo{
+		OriginModelName: "claude-3-7-sonnet",
+		RelayFormat:     types.RelayFormatClaude,
+	}
+	claudeInfo := &ClaudeResponseInfo{Usage: &dto.Usage{}}
+
+	data := []byte(`{
+		"type":"message",
+		"content":[
+			{"type":"text","text":"hi"},
+			{"type":"tool_use","id":"tu1","name":"lookup_fn","input":{}},
+			{"type":"server_tool_use","id":"stu1","name":"web_search","input":{}}
+		],
+		"usage":{"input_tokens":1,"output_tokens":1}
+	}`)
+
+	err := HandleClaudeResponseData(c, info, claudeInfo, nil, data)
+	require.Nil(t, err)
+	require.NotNil(t, info.ResponsesUsageInfo)
+	require.Contains(t, info.ResponsesUsageInfo.BuiltInTools, "lookup_fn")
+	assert.Equal(t, 1, info.ResponsesUsageInfo.BuiltInTools["lookup_fn"].CallCount)
+	assert.NotContains(t, info.ResponsesUsageInfo.BuiltInTools, "web_search")
+}
+
+func TestCountClaudeStreamBillableToolsSetsWebSearchRequests(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+	info := &relaycommon.RelayInfo{OriginModelName: "claude-3-7-sonnet"}
+
+	countClaudeStreamBillableTools(c, info, &dto.ClaudeResponse{
+		Type: "message_delta",
+		Usage: &dto.ClaudeUsage{
+			ServerToolUse: &dto.ClaudeServerToolUse{WebSearchRequests: 3},
+		},
+	})
+	assert.Equal(t, 3, c.GetInt("claude_web_search_requests"))
+
+	operation_setting.SetToolPriceForTest("stream_fn", 2.0)
+	t.Cleanup(func() {
+		operation_setting.DeleteToolPriceForTest("stream_fn")
+	})
+	countClaudeStreamBillableTools(c, info, &dto.ClaudeResponse{
+		Type: "content_block_start",
+		ContentBlock: &dto.ClaudeMediaMessage{
+			Type: "tool_use",
+			Name: "stream_fn",
+		},
+	})
+	require.Contains(t, info.ResponsesUsageInfo.BuiltInTools, "stream_fn")
+	assert.Equal(t, 1, info.ResponsesUsageInfo.BuiltInTools["stream_fn"].CallCount)
+}