diff --git a/src/services/pricingService.js b/src/services/pricingService.js index 0eb3b2f6..cd04b5f2 100644 --- a/src/services/pricingService.js +++ b/src/services/pricingService.js @@ -528,11 +528,6 @@ class PricingService { } } - // Claude Fast Mode 目前仅适用于 Opus 4.6 系列 - isFastModeEligibleClaudeModel(modelName) { - return typeof modelName === 'string' && modelName.toLowerCase().includes('opus-4-6') - } - // 去掉模型名中的 [1m] 后缀，便于价格查找 stripLongContextSuffix(modelName) { if (typeof modelName !== 'string') { @@ -541,45 +536,6 @@ class PricingService { return modelName.replace(/\[1m\]/gi, '').trim() } - // 获取 Fast Mode 对应的价格条目（仅匹配 fast/ 前缀） - getFastModePricing(modelName) { - if (!this.pricingData || !modelName) { - return null - } - - const cleanedModelName = this.stripLongContextSuffix(modelName) - const exactCandidates = new Set([`fast/${cleanedModelName}`]) - - if (cleanedModelName.startsWith('fast/')) { - exactCandidates.add(cleanedModelName) - } - - for (const candidate of exactCandidates) { - if (this.pricingData[candidate]) { - logger.debug(`💰 Found exact fast pricing for ${modelName}: ${candidate}`) - return this.pricingData[candidate] - } - } - - const normalizedModel = cleanedModelName.toLowerCase().replace(/[_-]/g, '') - for (const [key, value] of Object.entries(this.pricingData)) { - if (!key.startsWith('fast/')) { - continue - } - const normalizedFastKey = key.slice('fast/'.length).toLowerCase().replace(/[_-]/g, '') - if ( - normalizedFastKey.includes(normalizedModel) || - normalizedModel.includes(normalizedFastKey) - ) { - logger.debug(`💰 Found fuzzy fast pricing for ${modelName}: ${key}`) - return value - } - } - - logger.debug(`💰 No fast pricing found for model: ${modelName}`) - return null - } - // 获取 1 小时缓存价格（优先使用 model_pricing.json 中的模型字段） getEphemeral1hPricing(modelName, pricing = null) { if ( @@ -606,7 +562,7 @@ class PricingService { // 检查是否是 Opus 系列 if (modelLower.includes('opus')) { - return 0.00003 // $30/MTok + return 0.00001 // $10/MTok } // 检查是否是 Sonnet 系列 @@ -616,7 +572,7 @@ class PricingService { // 检查是否是 Haiku 系列 if (modelLower.includes('haiku')) { - return 0.0000016 // $1.6/MTok + return 0.000002 // $2/MTok } // 默认返回 0（未知模型） @@ -647,15 +603,14 @@ class PricingService { const hasFastSpeedSignal = responseSpeed === this.claudeFeatureFlags.fastModeSpeed || requestSpeed === this.claudeFeatureFlags.fastModeSpeed - const isFastModeRequest = - hasFastModeBeta && - hasFastSpeedSignal && - this.isFastModeEligibleClaudeModel(normalizedModelName) + const isFastModeRequest = hasFastModeBeta && hasFastSpeedSignal const standardPricing = this.getModelPricing(modelName) - const fastPricing = isFastModeRequest ? this.getFastModePricing(normalizedModelName) : null - const pricing = fastPricing || standardPricing + const pricing = standardPricing const isLongContextModeEnabled = isLongContextModel || hasContext1mBeta + // Fast Mode 倍率：优先从 provider_specific_entry.fast 读取，默认 6 倍 + const fastMultiplier = isFastModeRequest ? pricing?.provider_specific_entry?.fast || 6 : 1 + // 当 [1m] 模型总输入超过 200K 时，进入 200K+ 计费逻辑 // 根据 Anthropic 官方文档：当总输入超过 200K 时，整个请求所有 token 类型都使用高档价格 if (isLongContextModeEnabled && totalInputTokens > 200000) { @@ -685,11 +640,13 @@ class PricingService { (typeof pricing?.litellm_provider === 'string' && pricing.litellm_provider.toLowerCase().includes('anthropic')) - if (isFastModeRequest && fastPricing) { - logger.info(`🚀 Fast mode pricing profile selected: fast/${normalizedModelName}`) - } else if (isFastModeRequest && !fastPricing) { + if (isFastModeRequest && fastMultiplier > 1) { + logger.info( + `🚀 Fast mode ${fastMultiplier}x multiplier applied for ${normalizedModelName} (from provider_specific_entry)` + ) + } else if (isFastModeRequest) { logger.warn( - `⚠️ Fast mode request detected but no fast pricing profile found for ${normalizedModelName}; fallback to standard profile` + `⚠️ Fast mode request detected but no fast pricing found for ${normalizedModelName}; fallback to standard profile` ) } @@ -700,7 +657,7 @@ class PricingService { // 确定实际使用的输入价格（普通或 200K+ 高档价格） // Claude 模型在 200K+ 场景下如果缺少官方字段，按 2 倍输入价兜底 - const actualInputPrice = useLongContextPricing + let actualInputPrice = useLongContextPricing ? hasInput200kPrice ? pricing.input_cost_per_token_above_200k_tokens : isClaudeModel @@ -712,12 +669,18 @@ class PricingService { const hasOutput200kPrice = pricing.output_cost_per_token_above_200k_tokens !== null && pricing.output_cost_per_token_above_200k_tokens !== undefined - const actualOutputPrice = useLongContextPricing + let actualOutputPrice = useLongContextPricing ? hasOutput200kPrice ? pricing.output_cost_per_token_above_200k_tokens : baseOutputPrice : baseOutputPrice + // 应用 Fast Mode 倍率（在 200K+ 价格之上叠加） + if (fastMultiplier > 1) { + actualInputPrice *= fastMultiplier + actualOutputPrice *= fastMultiplier + } + let actualCacheCreatePrice = 0 let actualCacheReadPrice = 0 let actualEphemeral1hPrice = 0 diff --git a/tests/pricingService.test.js b/tests/pricingService.test.js index 40dc0dd7..f0bc8687 100644 --- a/tests/pricingService.test.js +++ b/tests/pricingService.test.js @@ -39,49 +39,27 @@ jest.mock('fs', () => { describe('PricingService - 200K+ Long Context Pricing', () => { let pricingService const fs = require('fs') + const path = require('path') - // 模拟 claude-sonnet-4-20250514 的完整价格数据（来自 model_pricing.json） - const mockPricingData = { - 'claude-sonnet-4-20250514': { - input_cost_per_token: 0.000003, // $3/MTok - output_cost_per_token: 0.000015, // $15/MTok - cache_creation_input_token_cost: 0.00000375, // $3.75/MTok - cache_read_input_token_cost: 0.0000003, // $0.30/MTok - max_input_tokens: 1000000, - // 200K+ 高档价格 - input_cost_per_token_above_200k_tokens: 0.000006, // $6/MTok (2x) - output_cost_per_token_above_200k_tokens: 0.0000225, // $22.50/MTok (1.5x) - cache_creation_input_token_cost_above_200k_tokens: 0.0000075, // $7.50/MTok (2x) - cache_read_input_token_cost_above_200k_tokens: 0.0000006, // $0.60/MTok (2x) - // 1小时缓存价格 - cache_creation_input_token_cost_above_1hr: 0.0000075, - cache_creation_input_token_cost_above_1hr_above_200k_tokens: 0.000015 - }, - // 没有 above_200k 字段的模型 - 'claude-3-haiku-20240307': { - input_cost_per_token: 0.00000025, - output_cost_per_token: 0.00000125, - cache_creation_input_token_cost: 0.0000003, - cache_read_input_token_cost: 0.00000003 - }, - // Fast Mode 适配测试模型（Opus 4.6） - 'claude-opus-4-6': { - input_cost_per_token: 0.000005, - output_cost_per_token: 0.000025, - cache_creation_input_token_cost: 0.00000625, - cache_read_input_token_cost: 0.0000005, - input_cost_per_token_above_200k_tokens: 0.00001, - output_cost_per_token_above_200k_tokens: 0.0000375 - } - } + // 使用真实的 model_pricing.json 数据（优先 data/，fallback 到 resources/） + const realFs = jest.requireActual('fs') + const primaryPath = path.join(process.cwd(), 'data', 'model_pricing.json') + const fallbackPath = path.join( + process.cwd(), + 'resources', + 'model-pricing', + 'model_prices_and_context_window.json' + ) + const pricingFilePath = realFs.existsSync(primaryPath) ? primaryPath : fallbackPath + const pricingData = JSON.parse(realFs.readFileSync(pricingFilePath, 'utf8')) beforeEach(() => { // 清除缓存的模块 jest.resetModules() - // 配置 fs mock + // 配置 fs mock（防止 pricingService 初始化时的文件副作用） fs.existsSync.mockReturnValue(true) - fs.readFileSync.mockReturnValue(JSON.stringify(mockPricingData)) + fs.readFileSync.mockReturnValue(JSON.stringify(pricingData)) fs.statSync.mockReturnValue({ mtime: new Date(), mtimeMs: Date.now() }) fs.watchFile.mockImplementation(() => {}) fs.unwatchFile.mockImplementation(() => {}) @@ -89,8 +67,8 @@ describe('PricingService - 200K+ Long Context Pricing', () => { // 重新加载 pricingService pricingService = require('../src/services/pricingService') - // 直接设置价格数据（绕过初始化） - pricingService.pricingData = mockPricingData + // 直接设置真实价格数据（绕过网络初始化） + pricingService.pricingData = pricingData pricingService.lastUpdated = new Date() })