Merge pull request #6514 from StarryKira/codex/anthropic-reasoning-pricing

feat: add Anthropic reasoning effort and Fable 5.1 max pricing

backend/internal/handler/admin/channel_handler.go

@@ -57,21 +57,22 @@ type updateChannelRequest struct {
 }
 
 type channelModelPricingRequest struct {
-	Platform          string                     `json:"platform" binding:"omitempty,max=50"`
-	Models            []string                   `json:"models" binding:"required,min=1,max=100"`
-	BillingMode       string                     `json:"billing_mode" binding:"omitempty,oneof=token per_request image"`
-	InputPrice        *float64                   `json:"input_price" binding:"omitempty,min=0"`
-	OutputPrice       *float64                   `json:"output_price" binding:"omitempty,min=0"`
-	CacheWritePrice   *float64                   `json:"cache_write_price" binding:"omitempty,min=0"`
-	CacheWrite1hPrice *float64                   `json:"cache_write_1h_price" binding:"omitempty,min=0"`
-	CacheReadPrice    *float64                   `json:"cache_read_price" binding:"omitempty,min=0"`
-	FastMultiplier    *float64                   `json:"fast_multiplier" binding:"omitempty,gt=0"`
-	FlexMultiplier    *float64                   `json:"flex_multiplier" binding:"omitempty,gt=0"`
-	ImageInputPrice   *float64                   `json:"image_input_price" binding:"omitempty,min=0"`
-	ImageOutputPrice  *float64                   `json:"image_output_price" binding:"omitempty,min=0"`
-	PerRequestPrice   *float64                   `json:"per_request_price" binding:"omitempty,min=0"`
-	Intervals         []pricingIntervalRequest   `json:"intervals"`
-	TimePricing       *channelTimePricingRequest `json:"time_pricing"`
+	Platform                     string                     `json:"platform" binding:"omitempty,max=50"`
+	Models                       []string                   `json:"models" binding:"required,min=1,max=100"`
+	BillingMode                  string                     `json:"billing_mode" binding:"omitempty,oneof=token per_request image"`
+	InputPrice                   *float64                   `json:"input_price" binding:"omitempty,min=0"`
+	OutputPrice                  *float64                   `json:"output_price" binding:"omitempty,min=0"`
+	CacheWritePrice              *float64                   `json:"cache_write_price" binding:"omitempty,min=0"`
+	CacheWrite1hPrice            *float64                   `json:"cache_write_1h_price" binding:"omitempty,min=0"`
+	CacheReadPrice               *float64                   `json:"cache_read_price" binding:"omitempty,min=0"`
+	FastMultiplier               *float64                   `json:"fast_multiplier" binding:"omitempty,gt=0"`
+	FlexMultiplier               *float64                   `json:"flex_multiplier" binding:"omitempty,gt=0"`
+	MaxReasoningEffortMultiplier *float64                   `json:"max_reasoning_effort_multiplier" binding:"omitempty,gt=0"`
+	ImageInputPrice              *float64                   `json:"image_input_price" binding:"omitempty,min=0"`
+	ImageOutputPrice             *float64                   `json:"image_output_price" binding:"omitempty,min=0"`
+	PerRequestPrice              *float64                   `json:"per_request_price" binding:"omitempty,min=0"`
+	Intervals                    []pricingIntervalRequest   `json:"intervals"`
+	TimePricing                  *channelTimePricingRequest `json:"time_pricing"`
 }
 
 type channelTimePricingRequest struct {
@@ -129,22 +130,23 @@ type channelResponse struct {
 }
 
 type channelModelPricingResponse struct {
-	ID                int64                       `json:"id"`
-	Platform          string                      `json:"platform"`
-	Models            []string                    `json:"models"`
-	BillingMode       string                      `json:"billing_mode"`
-	InputPrice        *float64                    `json:"input_price"`
-	OutputPrice       *float64                    `json:"output_price"`
-	CacheWritePrice   *float64                    `json:"cache_write_price"`
-	CacheWrite1hPrice *float64                    `json:"cache_write_1h_price"`
-	CacheReadPrice    *float64                    `json:"cache_read_price"`
-	FastMultiplier    *float64                    `json:"fast_multiplier"`
-	FlexMultiplier    *float64                    `json:"flex_multiplier"`
-	ImageInputPrice   *float64                    `json:"image_input_price"`
-	ImageOutputPrice  *float64                    `json:"image_output_price"`
-	PerRequestPrice   *float64                    `json:"per_request_price"`
-	Intervals         []pricingIntervalResponse   `json:"intervals"`
-	TimePricing       *channelTimePricingResponse `json:"time_pricing"`
+	ID                           int64                       `json:"id"`
+	Platform                     string                      `json:"platform"`
+	Models                       []string                    `json:"models"`
+	BillingMode                  string                      `json:"billing_mode"`
+	InputPrice                   *float64                    `json:"input_price"`
+	OutputPrice                  *float64                    `json:"output_price"`
+	CacheWritePrice              *float64                    `json:"cache_write_price"`
+	CacheWrite1hPrice            *float64                    `json:"cache_write_1h_price"`
+	CacheReadPrice               *float64                    `json:"cache_read_price"`
+	FastMultiplier               *float64                    `json:"fast_multiplier"`
+	FlexMultiplier               *float64                    `json:"flex_multiplier"`
+	MaxReasoningEffortMultiplier *float64                    `json:"max_reasoning_effort_multiplier"`
+	ImageInputPrice              *float64                    `json:"image_input_price"`
+	ImageOutputPrice             *float64                    `json:"image_output_price"`
+	PerRequestPrice              *float64                    `json:"per_request_price"`
+	Intervals                    []pricingIntervalResponse   `json:"intervals"`
+	TimePricing                  *channelTimePricingResponse `json:"time_pricing"`
 }
 
 type channelTimePricingResponse struct {
@@ -258,22 +260,23 @@ func pricingToResponse(p *service.ChannelModelPricing) channelModelPricingRespon
 		intervals = append(intervals, intervalToResponse(iv))
 	}
 	return channelModelPricingResponse{
-		ID:                p.ID,
-		Platform:          platform,
-		Models:            models,
-		BillingMode:       billingMode,
-		InputPrice:        p.InputPrice,
-		OutputPrice:       p.OutputPrice,
-		CacheWritePrice:   p.CacheWritePrice,
-		CacheWrite1hPrice: p.CacheWrite1hPrice,
-		CacheReadPrice:    p.CacheReadPrice,
-		FastMultiplier:    p.FastMultiplier,
-		FlexMultiplier:    p.FlexMultiplier,
-		ImageInputPrice:   p.ImageInputPrice,
-		ImageOutputPrice:  p.ImageOutputPrice,
-		PerRequestPrice:   p.PerRequestPrice,
-		Intervals:         intervals,
-		TimePricing:       timePricingToResponse(p.TimePricing),
+		ID:                           p.ID,
+		Platform:                     platform,
+		Models:                       models,
+		BillingMode:                  billingMode,
+		InputPrice:                   p.InputPrice,
+		OutputPrice:                  p.OutputPrice,
+		CacheWritePrice:              p.CacheWritePrice,
+		CacheWrite1hPrice:            p.CacheWrite1hPrice,
+		CacheReadPrice:               p.CacheReadPrice,
+		FastMultiplier:               p.FastMultiplier,
+		FlexMultiplier:               p.FlexMultiplier,
+		MaxReasoningEffortMultiplier: p.MaxReasoningEffortMultiplier,
+		ImageInputPrice:              p.ImageInputPrice,
+		ImageOutputPrice:             p.ImageOutputPrice,
+		PerRequestPrice:              p.PerRequestPrice,
+		Intervals:                    intervals,
+		TimePricing:                  timePricingToResponse(p.TimePricing),
 	}
 }
 
@@ -350,27 +353,29 @@ func pricingRequestToService(reqs []channelModelPricingRequest, allowChannelMult
 				SortOrder:            iv.SortOrder,
 			})
 		}
-		var fastMultiplier, flexMultiplier *float64
+		var fastMultiplier, flexMultiplier, maxReasoningEffortMultiplier *float64
 		if allowChannelMultipliers {
 			fastMultiplier = r.FastMultiplier
 			flexMultiplier = r.FlexMultiplier
+			maxReasoningEffortMultiplier = r.MaxReasoningEffortMultiplier
 		}
 		result = append(result, service.ChannelModelPricing{
-			Platform:          platform,
-			Models:            r.Models,
-			BillingMode:       billingMode,
-			InputPrice:        r.InputPrice,
-			OutputPrice:       r.OutputPrice,
-			CacheWritePrice:   r.CacheWritePrice,
-			CacheWrite1hPrice: r.CacheWrite1hPrice,
-			CacheReadPrice:    r.CacheReadPrice,
-			FastMultiplier:    fastMultiplier,
-			FlexMultiplier:    flexMultiplier,
-			ImageInputPrice:   r.ImageInputPrice,
-			ImageOutputPrice:  r.ImageOutputPrice,
-			PerRequestPrice:   r.PerRequestPrice,
-			Intervals:         intervals,
-			TimePricing:       timePricingRequestToService(r.TimePricing),
+			Platform:                     platform,
+			Models:                       r.Models,
+			BillingMode:                  billingMode,
+			InputPrice:                   r.InputPrice,
+			OutputPrice:                  r.OutputPrice,
+			CacheWritePrice:              r.CacheWritePrice,
+			CacheWrite1hPrice:            r.CacheWrite1hPrice,
+			CacheReadPrice:               r.CacheReadPrice,
+			FastMultiplier:               fastMultiplier,
+			FlexMultiplier:               flexMultiplier,
+			MaxReasoningEffortMultiplier: maxReasoningEffortMultiplier,
+			ImageInputPrice:              r.ImageInputPrice,
+			ImageOutputPrice:             r.ImageOutputPrice,
+			PerRequestPrice:              r.PerRequestPrice,
+			Intervals:                    intervals,
+			TimePricing:                  timePricingRequestToService(r.TimePricing),
 		})
 	}
 	return result
@@ -615,14 +620,15 @@ func (h *ChannelHandler) GetModelDefaultPricing(c *gin.Context) {
 	}
 
 	response.Success(c, gin.H{
-		"found":                true,
-		"input_price":          pricing.InputPricePerToken,
-		"output_price":         pricing.OutputPricePerToken,
-		"cache_write_price":    cacheWritePrice,
-		"cache_write_1h_price": cacheWrite1hPrice,
-		"cache_read_price":     pricing.CacheReadPricePerToken,
-		"image_input_price":    pricing.ImageInputPricePerToken,
-		"image_output_price":   pricing.ImageOutputPricePerToken,
+		"found":                           true,
+		"input_price":                     pricing.InputPricePerToken,
+		"output_price":                    pricing.OutputPricePerToken,
+		"cache_write_price":               cacheWritePrice,
+		"cache_write_1h_price":            cacheWrite1hPrice,
+		"cache_read_price":                pricing.CacheReadPricePerToken,
+		"max_reasoning_effort_multiplier": pricing.MaxReasoningEffortMultiplier,
+		"image_input_price":               pricing.ImageInputPricePerToken,
+		"image_output_price":              pricing.ImageOutputPricePerToken,
 	})
 }
 

backend/internal/handler/admin/channel_handler_test.go

@@ -453,10 +453,11 @@ func TestPricingRequestToService_TimePricingNil(t *testing.T) {
 // 避免渠道倍率意外污染账号成本口径。
 func TestPricingRequestToService_MultipliersGatedByFlag(t *testing.T) {
 	req := channelModelPricingRequest{
-		Models:         []string{"gpt-5"},
-		BillingMode:    "token",
-		FastMultiplier: float64Ptr(2.5),
-		FlexMultiplier: float64Ptr(0.5),
+		Models:                       []string{"gpt-5"},
+		BillingMode:                  "token",
+		FastMultiplier:               float64Ptr(2.5),
+		FlexMultiplier:               float64Ptr(0.5),
+		MaxReasoningEffortMultiplier: float64Ptr(3),
 		Intervals: []pricingIntervalRequest{{
 			MinTokens:            272000,
 			InputMultiplier:      float64Ptr(2),
@@ -469,6 +470,7 @@ func TestPricingRequestToService_MultipliersGatedByFlag(t *testing.T) {
 	allowed := pricingRequestToService([]channelModelPricingRequest{req}, true)
 	require.Equal(t, float64Ptr(2.5), allowed[0].FastMultiplier)
 	require.Equal(t, float64Ptr(0.5), allowed[0].FlexMultiplier)
+	require.Equal(t, float64Ptr(3), allowed[0].MaxReasoningEffortMultiplier)
 	require.Equal(t, float64Ptr(2), allowed[0].Intervals[0].InputMultiplier)
 	require.Equal(t, float64Ptr(1.5), allowed[0].Intervals[0].OutputMultiplier)
 	require.Equal(t, float64Ptr(2), allowed[0].Intervals[0].CacheWriteMultiplier)
@@ -477,6 +479,7 @@ func TestPricingRequestToService_MultipliersGatedByFlag(t *testing.T) {
 	dropped := pricingRequestToService([]channelModelPricingRequest{req}, false)
 	require.Nil(t, dropped[0].FastMultiplier)
 	require.Nil(t, dropped[0].FlexMultiplier)
+	require.Nil(t, dropped[0].MaxReasoningEffortMultiplier)
 	require.Nil(t, dropped[0].Intervals[0].InputMultiplier)
 	require.Nil(t, dropped[0].Intervals[0].OutputMultiplier)
 	require.Nil(t, dropped[0].Intervals[0].CacheWriteMultiplier)
@@ -580,9 +583,10 @@ func TestGetModelDefaultPricing_ReturnsFable51CacheTTLs(t *testing.T) {
 	require.Equal(t, http.StatusOK, w.Code)
 	var body struct {
 		Data struct {
-			Found             bool     `json:"found"`
-			CacheWritePrice   float64  `json:"cache_write_price"`
-			CacheWrite1hPrice *float64 `json:"cache_write_1h_price"`
+			Found                        bool     `json:"found"`
+			CacheWritePrice              float64  `json:"cache_write_price"`
+			CacheWrite1hPrice            *float64 `json:"cache_write_1h_price"`
+			MaxReasoningEffortMultiplier *float64 `json:"max_reasoning_effort_multiplier"`
 		} `json:"data"`
 	}
 	require.NoError(t, json.Unmarshal(w.Body.Bytes(), &body))
@@ -590,6 +594,8 @@ func TestGetModelDefaultPricing_ReturnsFable51CacheTTLs(t *testing.T) {
 	require.InDelta(t, 12.5e-6, body.Data.CacheWritePrice, 1e-12)
 	require.NotNil(t, body.Data.CacheWrite1hPrice)
 	require.InDelta(t, 20e-6, *body.Data.CacheWrite1hPrice, 1e-12)
+	require.NotNil(t, body.Data.MaxReasoningEffortMultiplier)
+	require.Equal(t, 3.0, *body.Data.MaxReasoningEffortMultiplier)
 }
 
 func TestGetModelDefaultPricing_OmitsUnsupportedCache1hPrice(t *testing.T) {

backend/internal/handler/admin/group_handler.go

@@ -158,11 +158,11 @@ type CreateGroupRequest struct {
 	CodexModelsManifestConfig service.GroupCodexModelsManifestConfig `json:"codex_models_manifest_config"`
 	// 分组 RPM 上限（0 = 不限制）
 	RPMLimit int `json:"rpm_limit"`
-	// OpenAI/Codex 请求推理强度上限，空字符串表示不限制。
+	// Anthropic/OpenAI 请求推理强度上限，空字符串表示不限制。
 	MaxReasoningEffort string `json:"max_reasoning_effort"`
 	// 超过上限时的访问控制：downgrade（默认）或 deny。
 	MaxReasoningEffortOverLimit string `json:"max_reasoning_effort_over_limit"`
-	// OpenAI/Codex 推理强度映射，可按模型精确名、前缀或后缀限定。
+	// Anthropic/OpenAI 推理强度映射，可按模型精确名、前缀或后缀限定。
 	ReasoningEffortMappings []service.ReasoningEffortMapping `json:"reasoning_effort_mappings"`
 	// 从指定分组复制账号（创建后自动绑定）
 	CopyAccountsFromGroupIDs []int64 `json:"copy_accounts_from_group_ids"`
@@ -233,7 +233,7 @@ type UpdateGroupRequest struct {
 	CodexModelsManifestConfig *service.GroupCodexModelsManifestConfig `json:"codex_models_manifest_config"`
 	// 分组 RPM 上限（0 = 不限制）；nil 表示未提供不改动
 	RPMLimit *int `json:"rpm_limit"`
-	// OpenAI/Codex 请求推理强度上限；空字符串清除，nil 不修改。
+	// Anthropic/OpenAI 请求推理强度上限；空字符串清除，nil 不修改。
 	MaxReasoningEffort *string `json:"max_reasoning_effort"`
 	// 超过上限时的访问控制；空字符串视为 downgrade，nil 不修改。
 	MaxReasoningEffortOverLimit *string `json:"max_reasoning_effort_over_limit"`

backend/internal/handler/available_channel_handler.go

@@ -66,16 +66,17 @@ type userAvailableGroup struct {
 
 // userSupportedModelPricing 用户可见的定价字段白名单。
 type userSupportedModelPricing struct {
-	BillingMode       string                   `json:"billing_mode"`
-	InputPrice        *float64                 `json:"input_price"`
-	OutputPrice       *float64                 `json:"output_price"`
-	CacheWritePrice   *float64                 `json:"cache_write_price"`
-	CacheWrite1hPrice *float64                 `json:"cache_write_1h_price"`
-	CacheReadPrice    *float64                 `json:"cache_read_price"`
-	ImageInputPrice   *float64                 `json:"image_input_price"`
-	ImageOutputPrice  *float64                 `json:"image_output_price"`
-	PerRequestPrice   *float64                 `json:"per_request_price"`
-	Intervals         []userPricingIntervalDTO `json:"intervals"`
+	BillingMode                  string                   `json:"billing_mode"`
+	InputPrice                   *float64                 `json:"input_price"`
+	OutputPrice                  *float64                 `json:"output_price"`
+	CacheWritePrice              *float64                 `json:"cache_write_price"`
+	CacheWrite1hPrice            *float64                 `json:"cache_write_1h_price"`
+	CacheReadPrice               *float64                 `json:"cache_read_price"`
+	MaxReasoningEffortMultiplier *float64                 `json:"max_reasoning_effort_multiplier,omitempty"`
+	ImageInputPrice              *float64                 `json:"image_input_price"`
+	ImageOutputPrice             *float64                 `json:"image_output_price"`
+	PerRequestPrice              *float64                 `json:"per_request_price"`
+	Intervals                    []userPricingIntervalDTO `json:"intervals"`
 }
 
 // userPricingIntervalDTO 定价区间白名单（去掉内部 ID、SortOrder 等前端不渲染的字段）。
@@ -323,15 +324,16 @@ func toUserPricing(p *service.ChannelModelPricing) *userSupportedModelPricing {
 		billingMode = string(service.BillingModeToken)
 	}
 	return &userSupportedModelPricing{
-		BillingMode:       billingMode,
-		InputPrice:        p.InputPrice,
-		OutputPrice:       p.OutputPrice,
-		CacheWritePrice:   p.CacheWritePrice,
-		CacheWrite1hPrice: p.CacheWrite1hPrice,
-		CacheReadPrice:    p.CacheReadPrice,
-		ImageInputPrice:   p.ImageInputPrice,
-		ImageOutputPrice:  p.ImageOutputPrice,
-		PerRequestPrice:   p.PerRequestPrice,
-		Intervals:         intervals,
+		BillingMode:                  billingMode,
+		InputPrice:                   p.InputPrice,
+		OutputPrice:                  p.OutputPrice,
+		CacheWritePrice:              p.CacheWritePrice,
+		CacheWrite1hPrice:            p.CacheWrite1hPrice,
+		CacheReadPrice:               p.CacheReadPrice,
+		MaxReasoningEffortMultiplier: p.MaxReasoningEffortMultiplier,
+		ImageInputPrice:              p.ImageInputPrice,
+		ImageOutputPrice:             p.ImageOutputPrice,
+		PerRequestPrice:              p.PerRequestPrice,
+		Intervals:                    intervals,
 	}
 }

backend/internal/handler/composite_platform.go

@@ -64,13 +64,45 @@ func openAIReasoningEffortPolicyForRequest(c *gin.Context, apiKey *service.APIKe
 	if apiKey == nil || apiKey.Group == nil {
 		return "", nil, "", false
 	}
-	if apiKey.Group.Platform != service.PlatformOpenAI && apiKey.Group.Platform != service.PlatformComposite {
+	if apiKey.Group.Platform != service.PlatformAnthropic && apiKey.Group.Platform != service.PlatformOpenAI && apiKey.Group.Platform != service.PlatformComposite {
 		return "", nil, "", false
 	}
-	if effectiveAPIKeyPlatform(c, apiKey) != service.PlatformOpenAI {
+	effectivePlatform := effectiveAPIKeyPlatform(c, apiKey)
+	if effectivePlatform != service.PlatformAnthropic && effectivePlatform != service.PlatformOpenAI {
 		return "", nil, "", false
 	}
-	return apiKey.Group.MaxReasoningEffort, apiKey.Group.ReasoningEffortMappings, apiKey.Group.MaxReasoningEffortOverLimit, true
+	maxEffort, mappings := apiKey.Group.MaxReasoningEffort, apiKey.Group.ReasoningEffortMappings
+	if effectivePlatform == service.PlatformAnthropic {
+		maxEffort, mappings = anthropicCompatibleReasoningEffortPolicy(maxEffort, mappings)
+	}
+	return maxEffort, mappings, apiKey.Group.MaxReasoningEffortOverLimit, true
+}
+
+func anthropicReasoningEffortPolicyForRequest(c *gin.Context, apiKey *service.APIKey) (string, []service.ReasoningEffortMapping, string, bool) {
+	if apiKey == nil || apiKey.Group == nil {
+		return "", nil, "", false
+	}
+	if apiKey.Group.Platform != service.PlatformAnthropic && apiKey.Group.Platform != service.PlatformComposite {
+		return "", nil, "", false
+	}
+	if effectiveAPIKeyPlatform(c, apiKey) != service.PlatformAnthropic {
+		return "", nil, "", false
+	}
+	maxEffort, mappings := anthropicCompatibleReasoningEffortPolicy(apiKey.Group.MaxReasoningEffort, apiKey.Group.ReasoningEffortMappings)
+	return maxEffort, mappings, apiKey.Group.MaxReasoningEffortOverLimit, true
+}
+
+func anthropicCompatibleReasoningEffortPolicy(maxEffort string, mappings []service.ReasoningEffortMapping) (string, []service.ReasoningEffortMapping) {
+	if service.NormalizeMaxReasoningEffort(maxEffort) == "minimal" {
+		maxEffort = "low"
+	}
+	normalizedMappings := append([]service.ReasoningEffortMapping(nil), mappings...)
+	for i := range normalizedMappings {
+		if service.NormalizeMaxReasoningEffort(normalizedMappings[i].To) == "minimal" {
+			normalizedMappings[i].To = "low"
+		}
+	}
+	return maxEffort, normalizedMappings
 }
 
 func bindRequestedReasoningEffort(c *gin.Context, body []byte, model string) {
@@ -118,6 +150,14 @@ func respondOpenAIReasoningEffortPolicyError(c *gin.Context, err error, write fu
 	write(c, http.StatusForbidden, "permission_error", err.Error())
 }
 
+func applyAnthropicReasoningEffortPolicyForRequest(c *gin.Context, apiKey *service.APIKey, body []byte) ([]byte, bool, error) {
+	maxEffort, mappings, overLimit, ok := anthropicReasoningEffortPolicyForRequest(c, apiKey)
+	if !ok {
+		return body, false, nil
+	}
+	return service.ApplyReasoningEffortPolicy(body, maxEffort, mappings, overLimit)
+}
+
 func bindOpenAIReasoningEffortPolicyForMessagesRequest(c *gin.Context, apiKey *service.APIKey, body []byte) {
 	if c == nil || c.Request == nil {
 		return

backend/internal/handler/composite_platform_test.go

@@ -8,6 +8,7 @@ import (
 	"github.com/Wei-Shaw/sub2api/internal/service"
 	"github.com/gin-gonic/gin"
 	"github.com/stretchr/testify/require"
+	"github.com/tidwall/gjson"
 )
 
 func TestCompositeTargetPlatformAllowedResolvesKnownAllowedModel(t *testing.T) {
@@ -160,6 +161,75 @@ func TestOpenAIReasoningEffortPolicyForCompositeTarget(t *testing.T) {
 	require.Equal(t, body, got)
 }
 
+func TestOpenAIShapedReasoningEffortPolicyForAnthropicGroup(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	c, _ := gin.CreateTestContext(httptest.NewRecorder())
+	c.Request = httptest.NewRequest("POST", "/v1/responses", nil)
+	apiKey := &service.APIKey{Group: &service.Group{
+		Platform:           service.PlatformAnthropic,
+		MaxReasoningEffort: "xhigh",
+	}}
+
+	got, changed, err := applyOpenAIReasoningEffortPolicyForRequest(c, apiKey, []byte(`{"reasoning":{"effort":"max"}}`))
+	require.NoError(t, err)
+	require.True(t, changed)
+	require.Equal(t, "xhigh", gjson.GetBytes(got, "reasoning.effort").String())
+}
+
+func TestAnthropicReasoningEffortPolicyForConcreteAndCompositeTargets(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	body := []byte(`{"model":"claude-fable-5-1","output_config":{"effort":"max"}}`)
+	policy := func(platform string) *service.APIKey {
+		return &service.APIKey{Group: &service.Group{
+			Platform:           platform,
+			MaxReasoningEffort: "xhigh",
+			ReasoningEffortMappings: []service.ReasoningEffortMapping{
+				{From: "max", To: "xhigh"},
+			},
+		}}
+	}
+
+	concrete, _ := gin.CreateTestContext(httptest.NewRecorder())
+	concrete.Request = httptest.NewRequest("POST", "/v1/messages", nil)
+	got, changed, err := applyAnthropicReasoningEffortPolicyForRequest(concrete, policy(service.PlatformAnthropic), body)
+	require.NoError(t, err)
+	require.True(t, changed)
+	require.JSONEq(t, `{"model":"claude-fable-5-1","output_config":{"effort":"xhigh"}}`, string(got))
+
+	denyMax := policy(service.PlatformAnthropic)
+	denyMax.Group.ReasoningEffortMappings = nil
+	denyMax.Group.MaxReasoningEffortOverLimit = service.ReasoningEffortOverLimitDeny
+	_, changed, err = applyAnthropicReasoningEffortPolicyForRequest(concrete, denyMax, body)
+	require.Error(t, err)
+	require.False(t, changed)
+	var overLimit *service.ReasoningEffortOverLimitError
+	require.ErrorAs(t, err, &overLimit)
+	require.Equal(t, "max", overLimit.Requested)
+	require.Equal(t, "xhigh", overLimit.Max)
+
+	composite, _ := gin.CreateTestContext(httptest.NewRecorder())
+	composite.Request = httptest.NewRequest("POST", "/v1/messages", nil)
+	composite.Request = composite.Request.WithContext(service.WithResolvedTargetPlatform(composite.Request.Context(), service.PlatformAnthropic))
+	got, changed, err = applyAnthropicReasoningEffortPolicyForRequest(composite, policy(service.PlatformComposite), body)
+	require.NoError(t, err)
+	require.True(t, changed)
+	require.Equal(t, "xhigh", gjson.GetBytes(got, "output_config.effort").String())
+
+	openAI, _ := gin.CreateTestContext(httptest.NewRecorder())
+	openAI.Request = httptest.NewRequest("POST", "/v1/messages", nil)
+	openAI.Request = openAI.Request.WithContext(service.WithResolvedTargetPlatform(openAI.Request.Context(), service.PlatformOpenAI))
+	got, changed, err = applyAnthropicReasoningEffortPolicyForRequest(openAI, policy(service.PlatformComposite), body)
+	require.NoError(t, err)
+	require.False(t, changed)
+	require.Equal(t, body, got)
+}
+
+func TestAnthropicCompositePolicyClampsUnsupportedMinimalToLow(t *testing.T) {
+	maxEffort, mappings := anthropicCompatibleReasoningEffortPolicy("minimal", []service.ReasoningEffortMapping{{From: "max", To: "minimal"}})
+	require.Equal(t, "low", maxEffort)
+	require.Equal(t, []service.ReasoningEffortMapping{{From: "max", To: "low"}}, mappings)
+}
+
 func TestClientRequestedModelUsesCompositePublicModel(t *testing.T) {
 	gin.SetMode(gin.TestMode)
 	c, _ := gin.CreateTestContext(httptest.NewRecorder())

backend/internal/handler/dto/types.go

@@ -151,11 +151,11 @@ type Group struct {
 
 	// RPMLimit 分组级每分钟请求数上限（0 = 不限制），设置后覆盖用户级 rpm_limit。
 	RPMLimit int `json:"rpm_limit"`
-	// MaxReasoningEffort OpenAI/Codex 请求的推理强度上限，空字符串表示不限制。
+	// MaxReasoningEffort Anthropic/OpenAI 请求的推理强度上限，空字符串表示不限制。
 	MaxReasoningEffort string `json:"max_reasoning_effort"`
 	// MaxReasoningEffortOverLimit 超过上限时的访问控制：downgrade（默认）或 deny。
 	MaxReasoningEffortOverLimit string `json:"max_reasoning_effort_over_limit"`
-	// ReasoningEffortMappings OpenAI/Codex 推理强度映射，可按模型精确名、前缀或后缀限定。
+	// ReasoningEffortMappings Anthropic/OpenAI 推理强度映射，可按模型精确名、前缀或后缀限定。
 	ReasoningEffortMappings []domain.ReasoningEffortMapping `json:"reasoning_effort_mappings"`
 
 	CreatedAt time.Time `json:"created_at"`

backend/internal/handler/gateway_handler.go

@@ -171,6 +171,19 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 	reqStream := parsedReq.Stream
 	bindRequestedReasoningEffort(c, body, reqModel)
 	ensureCompositeTargetPlatform(c, apiKey, reqModel)
+	if policyBody, changed, err := applyAnthropicReasoningEffortPolicyForRequest(c, apiKey, body); err != nil {
+		respondOpenAIReasoningEffortPolicyError(c, err, h.errorResponse)
+		return
+	} else if changed {
+		if err := parsedReq.ReplaceBody(policyBody); err != nil {
+			reqLog.Warn("gateway.reasoning_effort_policy_parse_failed", zap.Error(err))
+			h.errorResponse(c, http.StatusBadRequest, "invalid_request_error", "Failed to apply reasoning effort policy")
+			return
+		}
+		body = parsedReq.Body.Bytes()
+		reqModel = parsedReq.Model
+		reqStream = parsedReq.Stream
+	}
 	reqLog = reqLog.With(zap.String("model", reqModel), zap.Bool("stream", reqStream))
 
 	// 解析渠道级模型映射