feat(model):claude模型5m/1h缓存收费

internal/logic/billing/pricing.go

@@ -45,8 +45,10 @@ type PricingResult struct {
 	Currency         string
 
 	// Cache 直接定价
-	CacheReadPrice     float64 // 缓存读取每 1M token 价格
-	CacheCreationPrice float64 // 缓存创建每 1M token 价格
+	CacheReadPrice       float64 // 缓存读取每 1M token 价格
+	CacheCreationPrice   float64 // 缓存创建每 1M token 价格（5m TTL 基础价）
+	CacheCreation5mPrice float64 // 5 分钟缓存创建价格（1.25× 基础价）
+	CacheCreation1hPrice float64 // 1 小时缓存创建价格（2× 基础价）
 
 	// 模型最大输出 token 数（随定价一起缓存，供预扣估算使用，
 	// 避免 EstimatePreDeductAmount 每请求单独查一次 mdl_models）。
@@ -252,21 +254,23 @@ func GetModelPrice(ctx context.Context, tenantID int64, modelName string) (*Pric
 	modelMultiplier := 1.0
 
 	result := &PricingResult{
-		InputPrice:         inputPrice,
-		OutputPrice:        outputPrice,
-		BaseInputPrice:     baseInputPrice,
-		BaseOutputPrice:    baseOutputPrice,
-		BillingMode:        billingMode,
-		BillingSource:      billingSource,
-		PerRequestPrice:    perRequestPrice,
-		DiscountRatio:      discountRatio,
-		TenantMultiplier:   tenantMultiplier,
-		ModelMultiplier:    modelMultiplier,
-		Currency:           "USD",
-		CacheReadPrice:     cacheReadPrice,
-		CacheCreationPrice: cacheCreationPrice,
-		MaxOutputTokens:    model.MaxOutputTokens,
-		CustomTiers:        customTiers,
+		InputPrice:           inputPrice,
+		OutputPrice:          outputPrice,
+		BaseInputPrice:       baseInputPrice,
+		BaseOutputPrice:      baseOutputPrice,
+		BillingMode:          billingMode,
+		BillingSource:        billingSource,
+		PerRequestPrice:      perRequestPrice,
+		DiscountRatio:        discountRatio,
+		TenantMultiplier:     tenantMultiplier,
+		ModelMultiplier:      modelMultiplier,
+		Currency:             "USD",
+		CacheReadPrice:       cacheReadPrice,
+		CacheCreationPrice:   cacheCreationPrice,
+		CacheCreation5mPrice: cacheCreationPrice,       // 5m TTL = cache_creation_price 基础价（对应官方 1.25× 输入价）
+		CacheCreation1hPrice: cacheCreationPrice * 1.6, // 1h TTL = 1.6× 基础价（因为 2.0÷1.25=1.6，对应官方 2× 输入价）
+		MaxOutputTokens:      model.MaxOutputTokens,
+		CustomTiers:          customTiers,
 	}
 
 	modelPriceCache.Set(ctx, cacheKey, result)
@@ -287,27 +291,27 @@ func CalculateCost(ctx context.Context, tenantID int64, modelName string, inputT
 // computeCost 纯计算：根据定价结果和 token 用量计算费用明细。
 // 提取为独立函数以便单元测试，不依赖数据库或缓存。
 func computeCost(pricing *PricingResult, inputTokens, outputTokens int, usage *rcommon.Usage) *CostBreakdown {
-	inputTokens, outputTokens, cacheReadTokens, cacheCreationTokens := resolveTokenCounts(pricing, inputTokens, outputTokens, usage)
+	baseInputTokens, outputTokens, cacheReadTokens, cacheCreation5mTokens, cacheCreation1hTokens := resolveTokenCounts(pricing, inputTokens, outputTokens, usage)
 
 	// 按次计费：直接返回单价
 	if pricing.BillingMode == "per_request" {
 		return &CostBreakdown{
 			BaseCost:            pricing.PerRequestPrice,
 			TotalCost:           pricing.PerRequestPrice,
-			InputTokens:         inputTokens,
+			InputTokens:         baseInputTokens,
 			OutputTokens:        outputTokens,
 			BillingMode:         pricing.BillingMode,
 			PerRequestPrice:     pricing.PerRequestPrice,
 			DiscountRatio:       pricing.DiscountRatio,
 			TenantMultiplier:    pricing.TenantMultiplier,
 			Currency:            pricing.Currency,
-			CacheCreationTokens: cacheCreationTokens,
+			CacheCreationTokens: cacheCreation5mTokens + cacheCreation1hTokens,
 			CacheReadTokens:     cacheReadTokens,
 		}
 	}
 
 	// 基础输入费用（已改为 decimal）
-	baseInputCostD := computeInputCost(pricing, inputTokens)
+	baseInputCostD := computeInputCost(pricing, baseInputTokens)
 
 	// 输出费用（已改为 decimal）
 	outputCostD := computeOutputCost(pricing, outputTokens)
@@ -318,7 +322,11 @@ func computeCost(pricing *PricingResult, inputTokens, outputTokens int, usage *r
 	mul := NewFromFloat(pricing.TenantMultiplier)
 
 	cacheReadCostD := decimal.NewFromInt(int64(cacheReadTokens)).Div(million).Mul(NewFromFloat(pricing.CacheReadPrice))
-	cacheCreationCostD := decimal.NewFromInt(int64(cacheCreationTokens)).Div(million).Mul(NewFromFloat(pricing.CacheCreationPrice))
+
+	// 缓存创建按 TTL 分别计价：5m 按 1.25×，1h 按 2×
+	cacheCreation5mCostD := decimal.NewFromInt(int64(cacheCreation5mTokens)).Div(million).Mul(NewFromFloat(pricing.CacheCreation5mPrice))
+	cacheCreation1hCostD := decimal.NewFromInt(int64(cacheCreation1hTokens)).Div(million).Mul(NewFromFloat(pricing.CacheCreation1hPrice))
+	cacheCreationCostD := cacheCreation5mCostD.Add(cacheCreation1hCostD)
 
 	// 总费用 = (基础输入 + 输出 + cache各项) × 租户倍率
 	subtotalD := baseInputCostD.Add(outputCostD).Add(cacheReadCostD).Add(cacheCreationCostD)
@@ -329,14 +337,14 @@ func computeCost(pricing *PricingResult, inputTokens, outputTokens int, usage *r
 		InputCost:           InexactFloat64(RoundMoney(baseInputCostD.Mul(mul))),
 		OutputCost:          InexactFloat64(RoundMoney(outputCostD.Mul(mul))),
 		TotalCost:           InexactFloat64(RoundMoney(totalCostD)),
-		InputTokens:         inputTokens,
+		InputTokens:         baseInputTokens,
 		OutputTokens:        outputTokens,
 		BillingMode:         pricing.BillingMode,
 		PerRequestPrice:     pricing.PerRequestPrice,
 		DiscountRatio:       pricing.DiscountRatio,
 		TenantMultiplier:    pricing.TenantMultiplier,
 		Currency:            pricing.Currency,
-		CacheCreationTokens: cacheCreationTokens,
+		CacheCreationTokens: cacheCreation5mTokens + cacheCreation1hTokens,
 		CacheReadTokens:     cacheReadTokens,
 		CacheCreationCost:   InexactFloat64(RoundMoney(cacheCreationCostD.Mul(mul))),
 		CacheReadCost:       InexactFloat64(RoundMoney(cacheReadCostD.Mul(mul))),
@@ -345,17 +353,25 @@ func computeCost(pricing *PricingResult, inputTokens, outputTokens int, usage *r
 
 // resolveTokenCounts 根据 usage 信息解析最终的 token 计数。
 // 处理 cacheIncludedInPrompt 逻辑：如果 PromptTokens 包含 cache tokens，则扣减以避免重复计费。
-func resolveTokenCounts(pricing *PricingResult, inputTokens, outputTokens int, usage *rcommon.Usage) (baseInput, output, cacheRead, cacheCreation int) {
+// 返回：baseInput, output, cacheRead, cacheCreation5m, cacheCreation1h
+func resolveTokenCounts(pricing *PricingResult, inputTokens, outputTokens int, usage *rcommon.Usage) (baseInput, output, cacheRead, cacheCreation5m, cacheCreation1h int) {
 	baseInput = inputTokens
 	output = outputTokens
 
 	if usage != nil {
 		if usage.PromptTokensDetails != nil {
 			cacheRead = usage.PromptTokensDetails.CachedTokens
-			cacheCreation = usage.PromptTokensDetails.CachedCreationTokens
+			// 优先使用细分的 5m/1h token（Claude 新协议）
+			if usage.PromptTokensDetails.CachedCreation5mTokens > 0 || usage.PromptTokensDetails.CachedCreation1hTokens > 0 {
+				cacheCreation5m = usage.PromptTokensDetails.CachedCreation5mTokens
+				cacheCreation1h = usage.PromptTokensDetails.CachedCreation1hTokens
+			} else if usage.PromptTokensDetails.CachedCreationTokens > 0 {
+				// 旧协议无 TTL 细分：全部按 5m 兜底（保守计费）
+				cacheCreation5m = usage.PromptTokensDetails.CachedCreationTokens
+			}
 		}
 		if usage.CacheIncludedInPrompt {
-			baseInput = inputTokens - cacheRead - cacheCreation
+			baseInput = inputTokens - cacheRead - cacheCreation5m - cacheCreation1h
 			if baseInput < 0 {
 				baseInput = 0
 			}

internal/logic/billing/pricing_test.go

@@ -556,13 +556,15 @@ func TestComputeCost_ZeroTokens(t *testing.T) {
 
 func TestComputeCost_WithCacheTokens(t *testing.T) {
 	pricing := &PricingResult{
-		InputPrice:         10.0,
-		OutputPrice:        30.0,
-		CacheReadPrice:     1.0, // $1/1M cache read
-		CacheCreationPrice: 2.5, // $2.5/1M cache creation
-		BillingMode:        "token",
-		TenantMultiplier:   1.0,
-		Currency:           "USD",
+		InputPrice:           10.0,
+		OutputPrice:          30.0,
+		CacheReadPrice:       1.0, // $1/1M cache read
+		CacheCreationPrice:   2.5, // $2.5/1M cache creation（5m 基础价）
+		CacheCreation5mPrice: 2.5, // 5m TTL 价格
+		CacheCreation1hPrice: 4.0, // 1h TTL 价格（2.5 × 1.6）
+		BillingMode:          "token",
+		TenantMultiplier:     1.0,
+		Currency:             "USD",
 	}
 
 	usage := &rcommon.Usage{
@@ -599,13 +601,15 @@ func TestComputeCost_WithCacheTokens(t *testing.T) {
 
 func TestComputeCost_CacheNotIncludedInPrompt(t *testing.T) {
 	pricing := &PricingResult{
-		InputPrice:         10.0,
-		OutputPrice:        30.0,
-		CacheReadPrice:     1.0,
-		CacheCreationPrice: 2.5,
-		BillingMode:        "token",
-		TenantMultiplier:   1.0,
-		Currency:           "USD",
+		InputPrice:           10.0,
+		OutputPrice:          30.0,
+		CacheReadPrice:       1.0,
+		CacheCreationPrice:   2.5,
+		CacheCreation5mPrice: 2.5,
+		CacheCreation1hPrice: 4.0,
+		BillingMode:          "token",
+		TenantMultiplier:     1.0,
+		Currency:             "USD",
 	}
 
 	usage := &rcommon.Usage{
@@ -660,9 +664,9 @@ func TestComputeCost_CacheWithTenantMultiplier(t *testing.T) {
 // ─── resolveTokenCounts ─────────────────────────────────────────────
 
 func TestResolveTokenCounts_NoUsage(t *testing.T) {
-	baseIn, out, cr, cc := resolveTokenCounts(&PricingResult{}, 1000, 500, nil)
-	if baseIn != 1000 || out != 500 || cr != 0 || cc != 0 {
-		t.Errorf("expected (1000,500,0,0), got (%d,%d,%d,%d)", baseIn, out, cr, cc)
+	baseIn, out, cr, cc5m, cc1h := resolveTokenCounts(&PricingResult{}, 1000, 500, nil)
+	if baseIn != 1000 || out != 500 || cr != 0 || cc5m != 0 || cc1h != 0 {
+		t.Errorf("expected (1000,500,0,0,0), got (%d,%d,%d,%d,%d)", baseIn, out, cr, cc5m, cc1h)
 	}
 }
 
@@ -675,12 +679,31 @@ func TestResolveTokenCounts_CacheIncluded(t *testing.T) {
 		},
 		CacheIncludedInPrompt: true,
 	}
-	baseIn, _, cr, cc := resolveTokenCounts(&PricingResult{}, 10_000, 500, usage)
+	baseIn, _, cr, cc5m, cc1h := resolveTokenCounts(&PricingResult{}, 10_000, 500, usage)
 	if baseIn != 5_000 {
 		t.Errorf("baseInput = %d, want 5000 (10000 - 3000 - 2000)", baseIn)
 	}
-	if cr != 3_000 || cc != 2_000 {
-		t.Errorf("cache: read=%d creation=%d, want 3000, 2000", cr, cc)
+	if cr != 3_000 || cc5m != 2_000 || cc1h != 0 {
+		t.Errorf("cache: read=%d creation5m=%d creation1h=%d, want 3000, 2000, 0", cr, cc5m, cc1h)
+	}
+}
+
+func TestResolveTokenCounts_CacheIncluded_5m1hSplit(t *testing.T) {
+	usage := &rcommon.Usage{
+		PromptTokens: 10_000,
+		PromptTokensDetails: &rcommon.TokenDetails{
+			CachedTokens:           3_000,
+			CachedCreation5mTokens: 1_200,
+			CachedCreation1hTokens: 800,
+		},
+		CacheIncludedInPrompt: true,
+	}
+	baseIn, _, cr, cc5m, cc1h := resolveTokenCounts(&PricingResult{}, 10_000, 500, usage)
+	if baseIn != 5_000 {
+		t.Errorf("baseInput = %d, want 5000 (10000 - 3000 - 1200 - 800)", baseIn)
+	}
+	if cr != 3_000 || cc5m != 1_200 || cc1h != 800 {
+		t.Errorf("cache: read=%d creation5m=%d creation1h=%d, want 3000, 1200, 800", cr, cc5m, cc1h)
 	}
 }
 
@@ -693,7 +716,7 @@ func TestResolveTokenCounts_CacheExceedsInput(t *testing.T) {
 		},
 		CacheIncludedInPrompt: true,
 	}
-	baseIn, _, _, _ := resolveTokenCounts(&PricingResult{}, 100, 50, usage)
+	baseIn, _, _, _, _ := resolveTokenCounts(&PricingResult{}, 100, 50, usage)
 	// 100 - 200 - 50 = -150 → clamped to 0
 	if baseIn != 0 {
 		t.Errorf("baseInput = %d, want 0 (clamped)", baseIn)
@@ -708,7 +731,7 @@ func TestResolveTokenCounts_CacheNotIncluded(t *testing.T) {
 		},
 		CacheIncludedInPrompt: false,
 	}
-	baseIn, _, cr, _ := resolveTokenCounts(&PricingResult{}, 10_000, 500, usage)
+	baseIn, _, cr, _, _ := resolveTokenCounts(&PricingResult{}, 10_000, 500, usage)
 	// No deduction when cache not included
 	if baseIn != 10_000 {
 		t.Errorf("baseInput = %d, want 10000 (no deduction)", baseIn)
@@ -725,8 +748,8 @@ func TestResolveTokenCounts_NilTokenDetails(t *testing.T) {
 		PromptTokensDetails:   nil,
 		CacheIncludedInPrompt: true,
 	}
-	baseIn, out, cr, cc := resolveTokenCounts(&PricingResult{}, 1000, 500, usage)
-	if baseIn != 1000 || out != 500 || cr != 0 || cc != 0 {
-		t.Errorf("expected (1000,500,0,0), got (%d,%d,%d,%d)", baseIn, out, cr, cc)
+	baseIn, out, cr, cc5m, cc1h := resolveTokenCounts(&PricingResult{}, 1000, 500, usage)
+	if baseIn != 1000 || out != 500 || cr != 0 || cc5m != 0 || cc1h != 0 {
+		t.Errorf("expected (1000,500,0,0,0), got (%d,%d,%d,%d,%d)", baseIn, out, cr, cc5m, cc1h)
 	}
 }

relay/handler/relay_handler.go

@@ -693,7 +693,7 @@ func RelayHandler(ctx context.Context, body []byte, path string, headers http.He
 						billing.IncrApiKeyQuotaUsed(settleCtx, rc.ApiKeyID, settleResult.ActualCost)
 					}
 				}
-				recordFailedUsage(provider, rc, selection, v.modelName, v.relayMode, v.isStream, err)
+				recordFailedUsageWithTokens(provider, rc, selection, v.modelName, v.relayMode, v.isStream, err, streamUsage)
 				finalizeTrace(trace, rc, hop, false, attempt, selection, err.Error(), info.LatencyMs())
 				return usage, v.billingResult, err
 			}
@@ -1002,6 +1002,37 @@ func recordFailedUsage(provider common.DataProvider, rc *RelayContext, selection
 	provider.RecordUsage(context.Background(), record)
 }
 
+// recordFailedUsageWithTokens 记录失败用量（含 token 明细，用于流中断等已有部分 usage 的场景）。
+// 与 recordFailedUsage 的区别：此函数会填充 token 字段，避免报表中流中断记录的 token 全为 0。
+func recordFailedUsageWithTokens(provider common.DataProvider, rc *RelayContext, selection *common.ChannelSelection, modelName string, relayMode constant.RelayMode, isStream bool, err error, usage *common.Usage) {
+	record := &common.UsageRecord{
+		TenantID:       rc.TenantID,
+		UserID:         rc.UserID,
+		ApiKeyID:       rc.ApiKeyID,
+		ProjectID:      rc.ProjectID,
+		ModelName:      modelName,
+		RequestedModel: modelName,
+		RelayMode:      int(relayMode),
+		IsStream:       isStream,
+		Success:        false,
+		RequestID:      rc.RequestID,
+		Status:         "error",
+		ErrorMessage:   err.Error(),
+	}
+	if selection != nil {
+		record.ChannelID = selection.ChannelID
+		record.ChannelName = selection.ChannelName
+		record.ChannelType = selection.ChannelType
+		record.UpstreamModel = selection.UpstreamModelName
+	}
+	if usage != nil {
+		record.PromptTokens = usage.PromptTokens
+		record.CompletionTokens = usage.CompletionTokens
+		record.TotalTokens = usage.TotalTokens
+	}
+	provider.RecordUsage(context.Background(), record)
+}
+
 // recordChannelError 记录渠道错误事件到 chn_error_events（异步，不阻塞请求）
 func recordChannelError(rc *RelayContext, selection *common.ChannelSelection, modelName string, attempt int, isFinal bool, err error, latencyMs float64) {
 	if commonlogic.DefaultChannelErrorWriter == nil {