From f0ee427f0ad2f7eab6ab8f095ba05041af3a48f8 Mon Sep 17 00:00:00 2001 From: feeeei Date: Sun, 11 Oct 2026 11:22:43 +0800 Subject: [PATCH] =?UTF-8?q?feat(gemini):=20=E6=98=BE=E5=BC=8F=E7=BC=93?= =?UTF-8?q?=E5=AD=98=E5=AD=98=E5=82=A8=E4=BB=B7=E5=86=85=E7=BD=AE=E5=AE=98?= =?UTF-8?q?=E6=96=B9=E4=BB=B7=E5=B9=B6=E6=8C=89=20Pro=20/=20Flash=20?= =?UTF-8?q?=E7=B3=BB=E5=88=97=E5=85=9C=E5=BA=95?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit - 新增 Gemini 显式缓存存储单价表（Standard 档，USD / token / 小时）： Pro 全系 $4.50、Flash 与 Flash-Lite 全系 $1.00；3.6 / 3.7 / 3.8 Flash 为优惠价 $0.50（有效期至 2026-12-31，2027-01-01 起恢复 $1.00）， AI Studio 与 Vertex 统一按优惠价计 - 型号表未命中的 Gemini 模型按系列兜底：Pro 取 $4.50，Flash / Flash-Lite 取标准价 $1.00；Flash-Lite 变体不落到同版本 Flash 的优惠价 - 取值优先级：价格表（override_file 等）配置值 > 内置型号价 > 内置系列价； 两者都没有的模型仍拒绝创建缓存 --- backend/internal/service/billing_service.go | 25 +++--- .../service/gemini_cache_storage_pricing.go | 72 +++++++++++++++++ .../gemini_cache_storage_pricing_test.go | 77 +++++++++++++++++++ backend/internal/service/pricing_service.go | 3 +- deploy/config.example.yaml | 10 ++- 5 files changed, 169 insertions(+), 18 deletions(-) create mode 100644 backend/internal/service/gemini_cache_storage_pricing.go create mode 100644 backend/internal/service/gemini_cache_storage_pricing_test.go diff --git a/backend/internal/service/billing_service.go b/backend/internal/service/billing_service.go index 6dfb9a0399..84ee77960d 100644 --- a/backend/internal/service/billing_service.go +++ b/backend/internal/service/billing_service.go @@ -1325,22 +1325,21 @@ func (s *BillingService) GetModelPricing(model string) (*ModelPricing, error) { return s.getModelPricingAt(model, timezone.Now()) } +// CacheStoragePricePerTokenHour 返回模型的显式上下文缓存存储单价（每 token 每小时，USD）。 +// 优先级：价格表（override_file 等）按确定性识别配置的值 > 内置型号价 > 内置 Pro/Flash 系列兜底价； +// 均未命中返回 0。 +func (s *BillingService) CacheStoragePricePerTokenHour(model string) float64 { + if s != nil && s.pricingService != nil { + if pricing := s.pricingService.GetIdentifiedModelPricing(model); pricing != nil && pricing.CacheStorageCostPerTokenPerHour > 0 { + return pricing.CacheStorageCostPerTokenPerHour + } + } + return builtinGeminiCacheStoragePrice(model) +} + // getModelPricingAt 是 GetModelPricing 的带计费时点内部变体：pricingAt 显式 // 驱动 DeepSeek pro→Flash 切换判定（切换点前 Pro 价、之后 Flash 价），使 // 展示/估算路径可与历史补账同刻复算，测试也能用固定时点钉住断言。 -// CacheStoragePricePerTokenHour 返回模型的显式上下文缓存存储单价（每 token 每小时，USD）。 -// 只做确定性识别，不按系列兜底；价格表未配置时返回 0。 -func (s *BillingService) CacheStoragePricePerTokenHour(model string) float64 { - if s == nil || s.pricingService == nil { - return 0 - } - pricing := s.pricingService.GetIdentifiedModelPricing(model) - if pricing == nil || pricing.CacheStorageCostPerTokenPerHour <= 0 { - return 0 - } - return pricing.CacheStorageCostPerTokenPerHour -} - func (s *BillingService) getModelPricingAt(model string, pricingAt time.Time) (*ModelPricing, error) { // 标准化模型名称（转小写） model = strings.ToLower(model) diff --git a/backend/internal/service/gemini_cache_storage_pricing.go b/backend/internal/service/gemini_cache_storage_pricing.go new file mode 100644 index 0000000000..c08ed2f683 --- /dev/null +++ b/backend/internal/service/gemini_cache_storage_pricing.go @@ -0,0 +1,72 @@ +package service + +import ( + "slices" + "strings" +) + +// Gemini 显式上下文缓存存储的内置官方单价（USD / token / 小时，Standard 档）。 +// 存储价与上下文长度无关（≤200K 与 >200K 同价）；Priority 档的更高存储价不在此列。 +// 来源：https://ai.google.dev/gemini-api/docs/pricing +// 与 https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing +const ( + geminiProCacheStoragePrice = 4.5e-6 // $4.50 / 1M tokens / 小时 + geminiFlashCacheStoragePrice = 1e-6 // $1.00 / 1M tokens / 小时，Flash-Lite 同价 + + // 3.6/3.7/3.8 Flash 的优惠价 $0.50 / 1M tokens / 小时，有效期至 2026-12-31， + // 2027-01-01 起恢复为 geminiFlashCacheStoragePrice。AI Studio 价目页直接标注该优惠价； + // Vertex 价目页存储价标 $1.00，优惠以「按该模型净消费返还 50% 抵扣金」发放， + // 官方未明确存储费是否在返还范围内，两类账号统一按优惠价计。 + geminiFlashPromoCacheStoragePrice = 0.5e-6 +) + +// geminiCacheStorageModelPrices 是已发布型号的存储单价。模型名等于前缀，或以「前缀-」开头 +// （-preview、-customtools、-001 等变体）即命中；更具体的前缀排在前面。 +var geminiCacheStorageModelPrices = []struct { + prefix string + price float64 +}{ + {"gemini-3.8-flash-cyber", geminiFlashCacheStoragePrice}, + {"gemini-3.8-flash", geminiFlashPromoCacheStoragePrice}, + {"gemini-3.7-flash", geminiFlashPromoCacheStoragePrice}, + {"gemini-3.6-flash", geminiFlashPromoCacheStoragePrice}, + {"gemini-3.5-flash-lite", geminiFlashCacheStoragePrice}, + {"gemini-3.5-flash", geminiFlashCacheStoragePrice}, + {"gemini-3.1-flash-lite", geminiFlashCacheStoragePrice}, + {"gemini-3.1-pro", geminiProCacheStoragePrice}, + {"gemini-3-flash", geminiFlashCacheStoragePrice}, + {"gemini-3-pro", geminiProCacheStoragePrice}, + {"gemini-2.5-flash-lite", geminiFlashCacheStoragePrice}, + {"gemini-2.5-flash", geminiFlashCacheStoragePrice}, + {"gemini-2.5-pro", geminiProCacheStoragePrice}, + {"gemini-2.0-flash-lite", geminiFlashCacheStoragePrice}, + {"gemini-2.0-flash", geminiFlashCacheStoragePrice}, +} + +// builtinGeminiCacheStoragePrice 返回内置的显式缓存存储单价：先按已发布型号匹配，未命中时按系列兜底， +// Pro 系列取 geminiProCacheStoragePrice，Flash 与 Flash-Lite 系列取 geminiFlashCacheStoragePrice（标准价， +// 不套用优惠价）。其他 Gemini 模型与非 Gemini 模型返回 0。 +func builtinGeminiCacheStoragePrice(model string) float64 { + name := lastSegment(strings.ToLower(strings.TrimSpace(model))) + if !strings.HasPrefix(name, "gemini-") { + return 0 + } + parts := strings.Split(name, "-") + lite := slices.Contains(parts, "lite") + for _, entry := range geminiCacheStorageModelPrices { + // Flash-Lite 变体只匹配 Flash-Lite 条目：gemini-3.8-flash-lite 不能落到 3.8 Flash 的优惠价。 + if lite != strings.HasSuffix(entry.prefix, "-lite") { + continue + } + if name == entry.prefix || strings.HasPrefix(name, entry.prefix+"-") { + return entry.price + } + } + switch { + case slices.Contains(parts, "pro"): + return geminiProCacheStoragePrice + case slices.Contains(parts, "flash"): + return geminiFlashCacheStoragePrice + } + return 0 +} diff --git a/backend/internal/service/gemini_cache_storage_pricing_test.go b/backend/internal/service/gemini_cache_storage_pricing_test.go new file mode 100644 index 0000000000..846f7b56ec --- /dev/null +++ b/backend/internal/service/gemini_cache_storage_pricing_test.go @@ -0,0 +1,77 @@ +//go:build unit + +package service + +import ( + "testing" + + "github.com/Wei-Shaw/sub2api/internal/config" + "github.com/stretchr/testify/require" +) + +func TestBuiltinGeminiCacheStoragePrice(t *testing.T) { + for _, tc := range []struct { + model string + want float64 + }{ + // 已发布型号 + {"gemini-3.8-flash", 0.5e-6}, + {"gemini-3.7-flash", 0.5e-6}, + {"gemini-3.6-flash", 0.5e-6}, + {"gemini-3.8-flash-cyber", 1e-6}, + {"gemini-3.5-flash", 1e-6}, + {"gemini-3.5-flash-lite", 1e-6}, + {"gemini-3.1-flash-lite", 1e-6}, + {"gemini-3.1-pro-preview", 4.5e-6}, + {"gemini-3.1-pro-preview-customtools", 4.5e-6}, + {"gemini-3-flash-preview", 1e-6}, + {"gemini-3-pro-preview", 4.5e-6}, + {"gemini-2.5-pro", 4.5e-6}, + {"gemini-2.5-flash", 1e-6}, + {"gemini-2.5-flash-lite", 1e-6}, + {"gemini-2.0-flash-001", 1e-6}, + {"gemini-2.0-flash-lite-001", 1e-6}, + // 资源名与大小写 + {"models/gemini-3.8-flash", 0.5e-6}, + {"projects/p/locations/global/publishers/google/models/gemini-2.5-pro", 4.5e-6}, + {" Gemini-3.8-Flash ", 0.5e-6}, + // 系列兜底：未发布型号与别名按标准价 + {"gemini-3.9-flash", 1e-6}, + {"gemini-3.9-flash-lite", 1e-6}, + {"gemini-3.8-flash-lite", 1e-6}, + {"gemini-3.9-pro-preview", 4.5e-6}, + {"gemini-flash-latest", 1e-6}, + {"gemini-flash-lite-latest", 1e-6}, + {"gemini-pro-latest", 4.5e-6}, + // 无存储价 + {"gemini-embedding-001", 0}, + {"gemini-unknown-model", 0}, + {"claude-sonnet-5-5", 0}, + {"", 0}, + } { + require.InDelta(t, tc.want, builtinGeminiCacheStoragePrice(tc.model), 1e-15, tc.model) + } +} + +func TestCacheStoragePricePerTokenHourPriority(t *testing.T) { + pricing := &PricingService{} + data, err := pricing.parsePricingData([]byte(`{ + "gemini-3.8-flash":{"litellm_provider":"gemini","input_cost_per_token":0.00000075,"output_cost_per_token":0.00000375, + "cache_storage_cost_per_token_per_hour":0.000002}, + "gemini-3.9-flash":{"litellm_provider":"gemini","input_cost_per_token":0.0000015,"output_cost_per_token":0.0000075, + "cache_storage_cost_per_token_per_hour":0.0000007}, + "gemini-2.5-pro":{"litellm_provider":"gemini","input_cost_per_token":0.00000125,"output_cost_per_token":0.00001}}`)) + require.NoError(t, err) + pricing.pricingData = data + billing := NewBillingService(&config.Config{}, pricing) + + require.InDelta(t, 2e-6, billing.CacheStoragePricePerTokenHour("gemini-3.8-flash"), 1e-15, "价格表配置值优先于内置型号价") + require.InDelta(t, 0.7e-6, billing.CacheStoragePricePerTokenHour("gemini-3.9-flash"), 1e-15, "价格表配置值优先于内置系列兜底价") + require.InDelta(t, 4.5e-6, billing.CacheStoragePricePerTokenHour("gemini-2.5-pro"), 1e-15, "价格表条目缺存储价时取内置型号价") + require.InDelta(t, 0.5e-6, billing.CacheStoragePricePerTokenHour("gemini-3.7-flash"), 1e-15, "价格表无条目时取内置型号价") + require.InDelta(t, 4.5e-6, billing.CacheStoragePricePerTokenHour("gemini-3.9-pro"), 1e-15, "内置型号未命中时取系列兜底价") + require.Zero(t, billing.CacheStoragePricePerTokenHour("customer-alias")) + + noCatalog := NewBillingService(&config.Config{}, nil) + require.InDelta(t, 0.5e-6, noCatalog.CacheStoragePricePerTokenHour("gemini-3.8-flash"), 1e-15, "无价格服务时取内置价") +} diff --git a/backend/internal/service/pricing_service.go b/backend/internal/service/pricing_service.go index 45b530fd6d..679ee78226 100644 --- a/backend/internal/service/pricing_service.go +++ b/backend/internal/service/pricing_service.go @@ -223,7 +223,8 @@ type LiteLLMModelPricing struct { OutputCostPerImageToken float64 `json:"output_cost_per_image_token"` // 图片输出 token 价格 InputCostPerImageToken float64 `json:"input_cost_per_image_token"` // 图片输入 token 价格（如 gpt-image-2 图片编辑） CacheReadInputImageTokenCost float64 `json:"cache_read_input_image_token_cost"` - // CacheStorageCostPerTokenPerHour 显式上下文缓存存储单价（每 token 每小时，USD），仅由定价覆盖文件提供。 + // CacheStorageCostPerTokenPerHour 显式上下文缓存存储单价（每 token 每小时，USD），一般由定价覆盖文件提供， + // 配置后优先于内置价（见 BillingService.CacheStoragePricePerTokenHour）。 CacheStorageCostPerTokenPerHour float64 `json:"cache_storage_cost_per_token_per_hour"` // TokenPricingAbsent 表示源数据中 input/output token 价格均缺失（仅有图片价）。 diff --git a/deploy/config.example.yaml b/deploy/config.example.yaml index 174d5a4437..6b8b049838 100644 --- a/deploy/config.example.yaml +++ b/deploy/config.example.yaml @@ -1105,11 +1105,13 @@ pricing: # 覆盖补丁文件（可选，优先级最高）：条目按字段浅合并覆盖目录/回退数据， # 字段值为 null 表示删除该字段。模型名须与目录键完全一致。 # 修改后在下一次哈希检查时按内容哈希自动重载，无需重启。 - # Gemini explicit context caching requires a storage price per model, e.g. + # Gemini explicit context caching storage price (USD per token per hour) comes + # from the field cache_storage_cost_per_token_per_hour here, e.g. # {"gemini-3.8-flash": {"cache_storage_cost_per_token_per_hour": 0.0000005}}; - # models without it reject POST /v1beta/cachedContents. - # Gemini 显式缓存需按模型配置存储单价（每 token 每小时，USD）字段 - # cache_storage_cost_per_token_per_hour；未配置的模型拒绝创建缓存。 + # without it, built-in official prices apply (released models first, then the + # Pro / Flash series rate). Models with no price at all reject POST /v1beta/cachedContents. + # Gemini 显式缓存存储单价（每 token 每小时，USD）优先取这里的 cache_storage_cost_per_token_per_hour 字段， + # 未配置时使用内置官方价（先按已发布型号，再按 Pro / Flash 系列兜底）；两者都没有的模型拒绝创建缓存。 # 目录中不存在的模型条目还须同时提供 input_cost_per_token / output_cost_per_token。 # override_file: "./data/model_pricing_overrides.json" # Update interval in hours