feat: 区分 Anthropic 5m/1h 缓存创建 token 的差异化计费

Anthropic API 的 cache_creation 对象区分了 ephemeral_5m 和 ephemeral_1h 两种缓存创建 token，1h 单价远高于 5m（如 claude-3-5-haiku: 5m=$1/MTok, 1h=$6/MTok）。此前系统统一按 5m 单价计费，导致计费偏低。后端： - pricing_service: 加载 LiteLLM 的 cache_creation_input_token_cost_above_1hr - billing_service: GetModelPricing 启用分类计费（安全守卫 1h>5m）， CalculateCost 按 5m/1h 分别计费，无明细时回退到 5m 单价 - gateway_service: parseSSEUsage/handleNonStreamingResponse 用 gjson 提取嵌套 cache_creation 对象的 ephemeral_5m/1h_input_tokens - antigravity_gateway_service: extractSSEUsage/extractClaudeUsage 同步提取 - usage_log: 修复 GORM column tag 确保写入正确的数据库列 - 新增迁移 054: 删除 GORM 自动生成的重复列前端： - 使用记录 tooltip 展示 5m/1h 缓存创建明细（带彩色 badge 区分） - 表格单元格缓存写入数值旁显示 1h 标识
2026-02-14 18:15:35 +08:00
parent 2857fa2ef7
commit a817cafe3d
10 changed files with 174 additions and 42 deletions
--- a/backend/internal/service/antigravity_gateway_service.go
+++ b/backend/internal/service/antigravity_gateway_service.go
@@ -4111,6 +4111,15 @@ func (s *AntigravityGatewayService) extractSSEUsage(line string, usage *ClaudeUs
 	if v, ok := u["cache_creation_input_tokens"].(float64); ok && int(v) > 0 {
 		usage.CacheCreationInputTokens = int(v)
 	}
+	// 解析嵌套的 cache_creation 对象中的 5m/1h 明细
+	if cc, ok := u["cache_creation"].(map[string]any); ok {
+		if v, ok := cc["ephemeral_5m_input_tokens"].(float64); ok {
+			usage.CacheCreation5mTokens = int(v)
+		}
+		if v, ok := cc["ephemeral_1h_input_tokens"].(float64); ok {
+			usage.CacheCreation1hTokens = int(v)
+		}
+	}
 }

 // extractClaudeUsage 从非流式 Claude 响应提取 usage
@@ -4133,6 +4142,15 @@ func (s *AntigravityGatewayService) extractClaudeUsage(body []byte) *ClaudeUsage
 		if v, ok := u["cache_creation_input_tokens"].(float64); ok {
 			usage.CacheCreationInputTokens = int(v)
 		}
+		// 解析嵌套的 cache_creation 对象中的 5m/1h 明细
+		if cc, ok := u["cache_creation"].(map[string]any); ok {
+			if v, ok := cc["ephemeral_5m_input_tokens"].(float64); ok {
+				usage.CacheCreation5mTokens = int(v)
+			}
+			if v, ok := cc["ephemeral_1h_input_tokens"].(float64); ok {
+				usage.CacheCreation1hTokens = int(v)
+			}
+		}
 	}
 	return usage
 }
--- a/backend/internal/service/billing_service.go
+++ b/backend/internal/service/billing_service.go
@@ -31,8 +31,8 @@ type ModelPricing struct {
 	OutputPricePerToken        float64 // 每token输出价格 (USD)
 	CacheCreationPricePerToken float64 // 缓存创建每token价格 (USD)
 	CacheReadPricePerToken     float64 // 缓存读取每token价格 (USD)
-	CacheCreation5mPrice       float64 // 5分钟缓存创建价格（每百万token）- 仅用于硬编码回退
-	CacheCreation1hPrice       float64 // 1小时缓存创建价格（每百万token）- 仅用于硬编码回退
+	CacheCreation5mPrice       float64 // 5分钟缓存创建每token价格 (USD)
+	CacheCreation1hPrice       float64 // 1小时缓存创建每token价格 (USD)
 	SupportsCacheBreakdown     bool    // 是否支持详细的缓存分类
 }

@@ -172,12 +172,20 @@ func (s *BillingService) GetModelPricing(model string) (*ModelPricing, error) {
 	if s.pricingService != nil {
 		litellmPricing := s.pricingService.GetModelPricing(model)
 		if litellmPricing != nil {
+			// 启用 5m/1h 分类计费的条件：
+			// 1. 存在 1h 价格
+			// 2. 1h 价格 > 5m 价格（防止 LiteLLM 数据错误导致少收费）
+			price5m := litellmPricing.CacheCreationInputTokenCost
+			price1h := litellmPricing.CacheCreationInputTokenCostAbove1hr
+			enableBreakdown := price1h > 0 && price1h > price5m
 			return &ModelPricing{
 				InputPricePerToken:         litellmPricing.InputCostPerToken,
 				OutputPricePerToken:        litellmPricing.OutputCostPerToken,
 				CacheCreationPricePerToken: litellmPricing.CacheCreationInputTokenCost,
 				CacheReadPricePerToken:     litellmPricing.CacheReadInputTokenCost,
-				SupportsCacheBreakdown:     false,
+				CacheCreation5mPrice:       price5m,
+				CacheCreation1hPrice:       price1h,
+				SupportsCacheBreakdown:     enableBreakdown,
 			}, nil
 		}
 	}
@@ -209,9 +217,14 @@ func (s *BillingService) CalculateCost(model string, tokens UsageTokens, rateMul

 	// 计算缓存费用
 	if pricing.SupportsCacheBreakdown && (pricing.CacheCreation5mPrice > 0 || pricing.CacheCreation1hPrice > 0) {
-		// 支持详细缓存分类的模型（5分钟/1小时缓存）
-		breakdown.CacheCreationCost = float64(tokens.CacheCreation5mTokens)/1_000_000*pricing.CacheCreation5mPrice +
-			float64(tokens.CacheCreation1hTokens)/1_000_000*pricing.CacheCreation1hPrice
+		// 支持详细缓存分类的模型（5分钟/1小时缓存，价格为 per-token）
+		if tokens.CacheCreation5mTokens == 0 && tokens.CacheCreation1hTokens == 0 && tokens.CacheCreationTokens > 0 {
+			// API 未返回 ephemeral 明细，回退到全部按 5m 单价计费
+			breakdown.CacheCreationCost = float64(tokens.CacheCreationTokens) * pricing.CacheCreation5mPrice
+		} else {
+			breakdown.CacheCreationCost = float64(tokens.CacheCreation5mTokens)*pricing.CacheCreation5mPrice +
+				float64(tokens.CacheCreation1hTokens)*pricing.CacheCreation1hPrice
+		}
 	} else {
 		// 标准缓存创建价格（per-token）
 		breakdown.CacheCreationCost = float64(tokens.CacheCreationTokens) * pricing.CacheCreationPricePerToken
@@ -280,10 +293,12 @@ func (s *BillingService) CalculateCostWithLongContext(model string, tokens Usage

 	// 范围内部分：正常计费
 	inRangeTokens := UsageTokens{
-		InputTokens:         inRangeInputTokens,
-		OutputTokens:        tokens.OutputTokens, // 输出只算一次
-		CacheCreationTokens: tokens.CacheCreationTokens,
-		CacheReadTokens:     inRangeCacheTokens,
+		InputTokens:           inRangeInputTokens,
+		OutputTokens:          tokens.OutputTokens, // 输出只算一次
+		CacheCreationTokens:   tokens.CacheCreationTokens,
+		CacheReadTokens:       inRangeCacheTokens,
+		CacheCreation5mTokens: tokens.CacheCreation5mTokens,
+		CacheCreation1hTokens: tokens.CacheCreation1hTokens,
 	}
 	inRangeCost, err := s.CalculateCost(model, inRangeTokens, rateMultiplier)
 	if err != nil {
--- a/backend/internal/service/gateway_service.go
+++ b/backend/internal/service/gateway_service.go
@@ -349,6 +349,8 @@ type ClaudeUsage struct {
 	OutputTokens             int `json:"output_tokens"`
 	CacheCreationInputTokens int `json:"cache_creation_input_tokens"`
 	CacheReadInputTokens     int `json:"cache_read_input_tokens"`
+	CacheCreation5mTokens    int // 5分钟缓存创建token（来自嵌套 cache_creation 对象）
+	CacheCreation1hTokens    int // 1小时缓存创建token（来自嵌套 cache_creation 对象）
 }

 // ForwardResult 转发结果
@@ -4401,6 +4403,14 @@ func (s *GatewayService) parseSSEUsage(data string, usage *ClaudeUsage) {
 		usage.InputTokens = msgStart.Message.Usage.InputTokens
 		usage.CacheCreationInputTokens = msgStart.Message.Usage.CacheCreationInputTokens
 		usage.CacheReadInputTokens = msgStart.Message.Usage.CacheReadInputTokens
+
+		// 解析嵌套的 cache_creation 对象中的 5m/1h 明细
+		cc5m := gjson.Get(data, "message.usage.cache_creation.ephemeral_5m_input_tokens")
+		cc1h := gjson.Get(data, "message.usage.cache_creation.ephemeral_1h_input_tokens")
+		if cc5m.Exists() || cc1h.Exists() {
+			usage.CacheCreation5mTokens = int(cc5m.Int())
+			usage.CacheCreation1hTokens = int(cc1h.Int())
+		}
 	}

 	// 解析message_delta获取tokens（兼容GLM等把所有usage放在delta中的API）
@@ -4429,6 +4439,14 @@ func (s *GatewayService) parseSSEUsage(data string, usage *ClaudeUsage) {
 		if msgDelta.Usage.CacheReadInputTokens > 0 {
 			usage.CacheReadInputTokens = msgDelta.Usage.CacheReadInputTokens
 		}
+
+		// 解析嵌套的 cache_creation 对象中的 5m/1h 明细
+		cc5m := gjson.Get(data, "usage.cache_creation.ephemeral_5m_input_tokens")
+		cc1h := gjson.Get(data, "usage.cache_creation.ephemeral_1h_input_tokens")
+		if cc5m.Exists() || cc1h.Exists() {
+			usage.CacheCreation5mTokens = int(cc5m.Int())
+			usage.CacheCreation1hTokens = int(cc1h.Int())
+		}
 	}
 }

@@ -4449,6 +4467,14 @@ func (s *GatewayService) handleNonStreamingResponse(ctx context.Context, resp *h
 		return nil, fmt.Errorf("parse response: %w", err)
 	}

+	// 解析嵌套的 cache_creation 对象中的 5m/1h 明细
+	cc5m := gjson.GetBytes(body, "usage.cache_creation.ephemeral_5m_input_tokens")
+	cc1h := gjson.GetBytes(body, "usage.cache_creation.ephemeral_1h_input_tokens")
+	if cc5m.Exists() || cc1h.Exists() {
+		response.Usage.CacheCreation5mTokens = int(cc5m.Int())
+		response.Usage.CacheCreation1hTokens = int(cc1h.Int())
+	}
+
 	// 兼容 Kimi cached_tokens → cache_read_input_tokens
 	if response.Usage.CacheReadInputTokens == 0 {
 		cachedTokens := gjson.GetBytes(body, "usage.cached_tokens").Int()
@@ -4566,10 +4592,12 @@ func (s *GatewayService) RecordUsage(ctx context.Context, input *RecordUsageInpu
 	} else {
 		// Token 计费
 		tokens := UsageTokens{
-			InputTokens:         result.Usage.InputTokens,
-			OutputTokens:        result.Usage.OutputTokens,
-			CacheCreationTokens: result.Usage.CacheCreationInputTokens,
-			CacheReadTokens:     result.Usage.CacheReadInputTokens,
+			InputTokens:           result.Usage.InputTokens,
+			OutputTokens:          result.Usage.OutputTokens,
+			CacheCreationTokens:   result.Usage.CacheCreationInputTokens,
+			CacheReadTokens:       result.Usage.CacheReadInputTokens,
+			CacheCreation5mTokens: result.Usage.CacheCreation5mTokens,
+			CacheCreation1hTokens: result.Usage.CacheCreation1hTokens,
 		}
 		var err error
 		cost, err = s.billingService.CalculateCost(result.Model, tokens, multiplier)
@@ -4603,6 +4631,8 @@ func (s *GatewayService) RecordUsage(ctx context.Context, input *RecordUsageInpu
 		OutputTokens:          result.Usage.OutputTokens,
 		CacheCreationTokens:   result.Usage.CacheCreationInputTokens,
 		CacheReadTokens:       result.Usage.CacheReadInputTokens,
+		CacheCreation5mTokens: result.Usage.CacheCreation5mTokens,
+		CacheCreation1hTokens: result.Usage.CacheCreation1hTokens,
 		InputCost:             cost.InputCost,
 		OutputCost:            cost.OutputCost,
 		CacheCreationCost:     cost.CacheCreationCost,
@@ -4747,10 +4777,12 @@ func (s *GatewayService) RecordUsageWithLongContext(ctx context.Context, input *
 	} else {
 		// Token 计费（使用长上下文计费方法）
 		tokens := UsageTokens{
-			InputTokens:         result.Usage.InputTokens,
-			OutputTokens:        result.Usage.OutputTokens,
-			CacheCreationTokens: result.Usage.CacheCreationInputTokens,
-			CacheReadTokens:     result.Usage.CacheReadInputTokens,
+			InputTokens:           result.Usage.InputTokens,
+			OutputTokens:          result.Usage.OutputTokens,
+			CacheCreationTokens:   result.Usage.CacheCreationInputTokens,
+			CacheReadTokens:       result.Usage.CacheReadInputTokens,
+			CacheCreation5mTokens: result.Usage.CacheCreation5mTokens,
+			CacheCreation1hTokens: result.Usage.CacheCreation1hTokens,
 		}
 		var err error
 		cost, err = s.billingService.CalculateCostWithLongContext(result.Model, tokens, multiplier, input.LongContextThreshold, input.LongContextMultiplier)
@@ -4784,6 +4816,8 @@ func (s *GatewayService) RecordUsageWithLongContext(ctx context.Context, input *
 		OutputTokens:          result.Usage.OutputTokens,
 		CacheCreationTokens:   result.Usage.CacheCreationInputTokens,
 		CacheReadTokens:       result.Usage.CacheReadInputTokens,
+		CacheCreation5mTokens: result.Usage.CacheCreation5mTokens,
+		CacheCreation1hTokens: result.Usage.CacheCreation1hTokens,
 		InputCost:             cost.InputCost,
 		OutputCost:            cost.OutputCost,
 		CacheCreationCost:     cost.CacheCreationCost,
--- a/backend/internal/service/pricing_service.go
+++ b/backend/internal/service/pricing_service.go
@@ -27,14 +27,15 @@ var (
 // LiteLLMModelPricing LiteLLM价格数据结构
 // 只保留我们需要的字段，使用指针来处理可能缺失的值
 type LiteLLMModelPricing struct {
-	InputCostPerToken           float64 `json:"input_cost_per_token"`
-	OutputCostPerToken          float64 `json:"output_cost_per_token"`
-	CacheCreationInputTokenCost float64 `json:"cache_creation_input_token_cost"`
-	CacheReadInputTokenCost     float64 `json:"cache_read_input_token_cost"`
-	LiteLLMProvider             string  `json:"litellm_provider"`
-	Mode                        string  `json:"mode"`
-	SupportsPromptCaching       bool    `json:"supports_prompt_caching"`
-	OutputCostPerImage          float64 `json:"output_cost_per_image"` // 图片生成模型每张图片价格
+	InputCostPerToken                   float64 `json:"input_cost_per_token"`
+	OutputCostPerToken                  float64 `json:"output_cost_per_token"`
+	CacheCreationInputTokenCost         float64 `json:"cache_creation_input_token_cost"`
+	CacheCreationInputTokenCostAbove1hr float64 `json:"cache_creation_input_token_cost_above_1hr"`
+	CacheReadInputTokenCost             float64 `json:"cache_read_input_token_cost"`
+	LiteLLMProvider                     string  `json:"litellm_provider"`
+	Mode                                string  `json:"mode"`
+	SupportsPromptCaching               bool    `json:"supports_prompt_caching"`
+	OutputCostPerImage                  float64 `json:"output_cost_per_image"` // 图片生成模型每张图片价格
 }

 // PricingRemoteClient 远程价格数据获取接口
@@ -45,14 +46,15 @@ type PricingRemoteClient interface {

 // LiteLLMRawEntry 用于解析原始JSON数据
 type LiteLLMRawEntry struct {
-	InputCostPerToken           *float64 `json:"input_cost_per_token"`
-	OutputCostPerToken          *float64 `json:"output_cost_per_token"`
-	CacheCreationInputTokenCost *float64 `json:"cache_creation_input_token_cost"`
-	CacheReadInputTokenCost     *float64 `json:"cache_read_input_token_cost"`
-	LiteLLMProvider             string   `json:"litellm_provider"`
-	Mode                        string   `json:"mode"`
-	SupportsPromptCaching       bool     `json:"supports_prompt_caching"`
-	OutputCostPerImage          *float64 `json:"output_cost_per_image"`
+	InputCostPerToken                   *float64 `json:"input_cost_per_token"`
+	OutputCostPerToken                  *float64 `json:"output_cost_per_token"`
+	CacheCreationInputTokenCost         *float64 `json:"cache_creation_input_token_cost"`
+	CacheCreationInputTokenCostAbove1hr *float64 `json:"cache_creation_input_token_cost_above_1hr"`
+	CacheReadInputTokenCost             *float64 `json:"cache_read_input_token_cost"`
+	LiteLLMProvider                     string   `json:"litellm_provider"`
+	Mode                                string   `json:"mode"`
+	SupportsPromptCaching               bool     `json:"supports_prompt_caching"`
+	OutputCostPerImage                  *float64 `json:"output_cost_per_image"`
 }

 // PricingService 动态价格服务
@@ -318,6 +320,9 @@ func (s *PricingService) parsePricingData(body []byte) (map[string]*LiteLLMModel
 		if entry.CacheCreationInputTokenCost != nil {
 			pricing.CacheCreationInputTokenCost = *entry.CacheCreationInputTokenCost
 		}
+		if entry.CacheCreationInputTokenCostAbove1hr != nil {
+			pricing.CacheCreationInputTokenCostAbove1hr = *entry.CacheCreationInputTokenCostAbove1hr
+		}
 		if entry.CacheReadInputTokenCost != nil {
 			pricing.CacheReadInputTokenCost = *entry.CacheReadInputTokenCost
 		}
--- a/backend/internal/service/usage_log.go
+++ b/backend/internal/service/usage_log.go
@@ -26,8 +26,8 @@ type UsageLog struct {
 	CacheCreationTokens int
 	CacheReadTokens     int

-	CacheCreation5mTokens int
-	CacheCreation1hTokens int
+	CacheCreation5mTokens int `gorm:"column:cache_creation_5m_tokens"`
+	CacheCreation1hTokens int `gorm:"column:cache_creation_1h_tokens"`

 	InputCost         float64
 	OutputCost        float64
--- a/backend/migrations/054_drop_legacy_cache_columns.sql
+++ b/backend/migrations/054_drop_legacy_cache_columns.sql
@@ -0,0 +1,14 @@
+-- Drop legacy cache token columns that lack the underscore separator.
+-- These were created by GORM's automatic snake_case conversion:
+--   CacheCreation5mTokens → cache_creation5m_tokens  (incorrect)
+--   CacheCreation1hTokens → cache_creation1h_tokens  (incorrect)
+--
+-- The canonical columns are:
+--   cache_creation_5m_tokens  (defined in 001_init.sql)
+--   cache_creation_1h_tokens  (defined in 001_init.sql)
+--
+-- Migration 009 already copied data from legacy → canonical columns.
+-- This migration drops the legacy columns to avoid confusion.
+
+ALTER TABLE usage_logs DROP COLUMN IF EXISTS cache_creation5m_tokens;
+ALTER TABLE usage_logs DROP COLUMN IF EXISTS cache_creation1h_tokens;