Merge remote-tracking branch 'origin/main' into codex/pool-key-bulk-management-20260714

2026-09-12 06:00:20 +08:00 · 2026-07-17 00:17:02 +08:00
parent ec1783a35c 0be380243b
commit 3b0d87b0fd
40 changed files with 5012 additions and 565 deletions
@@ -183,4 +183,98 @@ describe('resolveModelsDevTieredPricing', () => {
  it('does not synthesize pricing when the fetched cost is absent', () => {
    expect(resolveModelsDevTieredPricing('openai', 'gpt-5.6-sol', undefined)).toBeNull()
  })
+
+  it('keeps a models.dev fast cost as an explicit Priority catalog when bands differ', () => {
+    expect(resolveModelsDevTieredPricing('openai', 'gpt-5.6-sol', {
+      input: 5,
+      output: 30,
+      tiers: [{
+        input: 10,
+        output: 45,
+        tier: { type: 'context', size: 272_000 },
+      }],
+    }, {
+      fast: {
+        cost: { input: 10, output: 60 },
+        provider: { body: { service_tier: 'priority' } },
+      },
+    })).toEqual({
+      tiers: [
+        { up_to: 271_999, input_price_per_1m: 5, output_price_per_1m: 30 },
+        { up_to: null, input_price_per_1m: 10, output_price_per_1m: 45 },
+      ],
+      processing_tiers: {
+        priority: {
+          tiers: [{ up_to: null, input_price_per_1m: 10, output_price_per_1m: 60 }],
+        },
+      },
+    })
+  })
+
+  it('uses a multiplier only when every fast price has the same ratio', () => {
+    expect(resolveModelsDevTieredPricing('anthropic', 'claude-opus-4.8', {
+      input: 5,
+      output: 25,
+      cache_read: 0.5,
+      cache_write: 6.25,
+    }, {
+      fast: {
+        cost: { input: 10, output: 50, cache_read: 1, cache_write: 12.5 },
+        provider: { body: { speed: 'fast' } },
+      },
+    })?.processing_tiers).toEqual({
+      fast: { price_multiplier: 2 },
+    })
+
+    expect(resolveModelsDevTieredPricing('anthropic', 'claude-opus-4.7', {
+      input: 5,
+      output: 25,
+    }, {
+      fast: {
+        cost: { input: 30, output: 150 },
+        provider: { body: { speed: 'fast' } },
+      },
+    })?.processing_tiers).toEqual({
+      fast: { price_multiplier: 6 },
+    })
+  })
+
+  it('prefers Anthropic speed=fast when the mode body also carries a standard service tier', () => {
+    expect(resolveModelsDevTieredPricing('anthropic', 'claude-opus-fast', {
+      input: 5,
+      output: 25,
+    }, {
+      fast: {
+        cost: { input: 10, output: 50 },
+        provider: { body: { speed: ' FAST ', service_tier: 'default' } },
+      },
+    })?.processing_tiers).toEqual({
+      fast: { price_multiplier: 2 },
+    })
+  })
+
+  it('falls back to the mode key and keeps non-uniform prices explicit', () => {
+    expect(resolveModelsDevTieredPricing('vendor', 'model', {
+      input: 2,
+      output: 4,
+    }, {
+      flex: { cost: { input: 1, output: 3 } },
+    })?.processing_tiers).toEqual({
+      flex: {
+        tiers: [{ up_to: null, input_price_per_1m: 1, output_price_per_1m: 3 }],
+      },
+    })
+  })
+
+  it('does not reinterpret unrelated or special experimental modes as processing tiers', () => {
+    const modes = JSON.parse(
+      '{"pro":{"cost":{"input":2,"output":4}},"__proto__":{"cost":{"input":2,"output":4}}}',
+    )
+    const pricing = resolveModelsDevTieredPricing('vendor', 'model', {
+      input: 1,
+      output: 2,
+    }, modes)
+
+    expect(pricing?.processing_tiers).toBeUndefined()
+  })
 })
@@ -0,0 +1,77 @@
+import { beforeEach, describe, expect, it, vi } from 'vitest'
+
+const apiMocks = vi.hoisted(() => ({
+  get: vi.fn(),
+}))
+
+vi.mock('@/api/client', () => ({
+  default: { get: apiMocks.get },
+}))
+
+import { clearModelsDevCache, getModelsDevList } from '@/api/models-dev'
+
+beforeEach(() => {
+  clearModelsDevCache()
+  localStorage.clear()
+  apiMocks.get.mockReset()
+})
+
+describe('getModelsDevList', () => {
+  it('uses current modalities and experimental mode pricing while keeping legacy fallbacks', async () => {
+    apiMocks.get.mockResolvedValue({
+      data: {
+        openai: {
+          id: 'openai',
+          name: 'OpenAI',
+          official: true,
+          models: {
+            'gpt-test': {
+              id: 'gpt-test',
+              name: 'GPT Test',
+              input: ['text'],
+              output: ['text'],
+              modalities: {
+                input: ['text', 'image'],
+                output: ['text', 'image'],
+              },
+              cost: { input: 2, output: 4 },
+              experimental: {
+                modes: {
+                  fast: {
+                    cost: { input: 4, output: 8 },
+                    provider: { body: { service_tier: 'priority' } },
+                  },
+                },
+              },
+            },
+            legacy: {
+              id: 'legacy',
+              name: 'Legacy',
+              input: ['text', 'image'],
+              output: ['text'],
+              cost: { input: 1, output: 2 },
+            },
+          },
+        },
+      },
+    })
+
+    const models = await getModelsDevList()
+    const current = models.find(model => model.modelId === 'gpt-test')
+    const legacy = models.find(model => model.modelId === 'legacy')
+
+    expect(current).toMatchObject({
+      supportsVision: true,
+      inputModalities: ['text', 'image'],
+      outputModalities: ['text', 'image'],
+      tieredPricing: {
+        processing_tiers: { priority: { price_multiplier: 2 } },
+      },
+    })
+    expect(legacy).toMatchObject({
+      supportsVision: true,
+      inputModalities: ['text', 'image'],
+      outputModalities: ['text'],
+    })
+  })
+})
@@ -154,10 +154,34 @@ export interface RequestSchedulingFailure {
  no_upstream_attempt?: boolean | null
 }

+export interface RequestPricingTier {
+  up_to?: number | null
+  input_price_per_1m?: number | null
+  output_price_per_1m?: number | null
+  cache_creation_price_per_1m?: number | null
+  cache_read_price_per_1m?: number | null
+  cache_ttl_pricing?: Array<{
+    ttl_minutes?: number | null
+    cache_creation_price_per_1m?: number | null
+    cache_read_price_per_1m?: number | null
+  }> | null
+  [key: string]: unknown
+}
+
+export interface RequestSettlementTieredPricing {
+  tiers?: RequestPricingTier[] | null
+  [key: string]: unknown
+}
+
 export interface RequestSettlementPricingSnapshot {
  requested_processing_tier?: string | null
  actual_processing_tier?: string | null
  billing_processing_tier?: string | null
+  processing_tier_price_multiplier?: number | null
+  pricing_source?: string | null
+  tiered_pricing_source?: string | null
+  price_per_request_source?: string | null
+  tiered_pricing?: RequestSettlementTieredPricing | null
  [key: string]: unknown
 }

@@ -283,30 +307,8 @@ export interface RequestDetail {
    tier_index: number  // 命中的阶梯索引 (0-based)
    tier_count: number  // 阶梯总数
    source?: 'provider' | 'global'  // 定价来源: 提供商或全局
-    current_tier: {  // 当前命中的阶梯配置
-      up_to?: number | null
-      input_price_per_1m: number
-      output_price_per_1m: number
-      cache_creation_price_per_1m?: number
-      cache_read_price_per_1m?: number
-      cache_ttl_pricing?: Array<{
-        ttl_minutes: number
-        cache_creation_price_per_1m?: number
-        cache_read_price_per_1m?: number
-      }>
-    }
-    tiers: Array<{  // 完整阶梯配置列表
-      up_to?: number | null
-      input_price_per_1m: number
-      output_price_per_1m: number
-      cache_creation_price_per_1m?: number
-      cache_read_price_per_1m?: number
-      cache_ttl_pricing?: Array<{
-        ttl_minutes: number
-        cache_creation_price_per_1m?: number
-        cache_read_price_per_1m?: number
-      }>
-    }>
+    current_tier: RequestPricingTier  // 当前命中的阶梯配置
+    tiers: RequestPricingTier[]  // 完整阶梯配置列表
  } | null
  // 视频/图像/音频计费信息
  video_billing?: VideoBilling | null
@@ -35,6 +35,8 @@ export interface ImageOutputPriceRange {

 /** 按处理层级覆盖的费率配置。允许图像或未来计费字段独立扩展。 */
 export interface ProcessingTierPricingConfig {
+  /** 相对 Standard 目录的统一价格倍率。新写入应与显式目录二选一；读取混合配置时显式目录优先。 */
+  price_multiplier?: number
  tiers?: PricingTier[]
  image_output_prices?: Record<string, ImageOutputQualityPricing> | null
  image_output_price_default?: number | null
@@ -52,6 +54,19 @@ export interface TieredPricingConfig {
  [key: string]: unknown
 }

+/**
+ * Provider 价格覆盖可以只声明 processing_tiers，并继续从 GlobalModel
+ * 继承 Standard 目录，因此 tiers 在原始 Provider 配置中是可选的。
+ */
+export interface ProviderTieredPricingConfig {
+  tiers?: PricingTier[]
+  image_output_prices?: Record<string, ImageOutputQualityPricing> | null
+  image_output_price_default?: number | null
+  image_output_price_ranges?: ImageOutputPriceRange[] | null
+  processing_tiers?: Record<string, ProcessingTierPricingConfig> | null
+  [key: string]: unknown
+}
+
 export interface Model {
  id: string
  provider_id: string
@@ -61,7 +76,7 @@ export interface Model {
  config?: Record<string, unknown> | null  // 额外配置（如 billing/video 等）
  // 原始配置值（可能为空，为空时使用 GlobalModel 默认值）
  price_per_request?: number | null  // 按次计费价格
-  tiered_pricing?: TieredPricingConfig | null  // 阶梯计费配置
+  tiered_pricing?: ProviderTieredPricingConfig | null  // Provider 原始覆盖，可仅包含 processing_tiers
  supports_vision?: boolean | null
  supports_function_calling?: boolean | null
  supports_streaming?: boolean | null
@@ -69,7 +84,7 @@ export interface Model {
  supports_image_generation?: boolean | null
  supports_embedding?: boolean | null
  // 有效值（合并 Model 和 GlobalModel 默认值后的结果）
-  effective_tiered_pricing?: TieredPricingConfig | null  // 有效阶梯计费配置
+  effective_tiered_pricing?: ProviderTieredPricingConfig | null  // 当前响应可能是 Provider partial 覆盖
  effective_input_price?: number | null
  effective_output_price?: number | null
  effective_price_per_request?: number | null  // 有效按次计费价格
@@ -97,7 +112,7 @@ export interface ModelCreate {
  global_model_id: string  // 关联的 GlobalModel ID（必填）
  // 计费配置（可选，为空时使用 GlobalModel 默认值）
  price_per_request?: number  // 按次计费价格
-  tiered_pricing?: TieredPricingConfig  // 阶梯计费配置
+  tiered_pricing?: ProviderTieredPricingConfig  // Provider 阶梯计费覆盖
  // 能力配置（可选，为空时使用 GlobalModel 默认值）
  supports_vision?: boolean
  supports_function_calling?: boolean
@@ -113,7 +128,7 @@ export interface ModelUpdate {
  provider_model_mappings?: ProviderModelMapping[] | null  // 模型名称映射列表（带优先级）
  global_model_id?: string
  price_per_request?: number | null  // 按次计费价格（null 表示清空/使用默认值）
-  tiered_pricing?: TieredPricingConfig | null  // 阶梯计费配置
+  tiered_pricing?: ProviderTieredPricingConfig | null  // Provider 阶梯计费覆盖
  supports_vision?: boolean
  supports_function_calling?: boolean
  supports_streaming?: boolean
@@ -21,6 +21,14 @@ export interface ModelsDevCost extends ModelsDevTokenCost {
  tiers?: ModelsDevCostTier[]
 }

+const TOKEN_PRICE_FIELDS = [
+  'input_price_per_1m',
+  'output_price_per_1m',
+  'cache_creation_price_per_1m',
+  'cache_read_price_per_1m',
+] as const
+const PROCESSING_MODE_FALLBACK_KEYS = new Set(['fast', 'priority', 'flex', 'batch'])
+
 function isRecord(value: unknown): value is Record<string, unknown> {
  return typeof value === 'object' && value !== null && !Array.isArray(value)
 }
@@ -91,11 +99,88 @@ export function buildModelsDevTieredPricing(cost: unknown): TieredPricingConfig
  return { tiers }
 }

+function uniformPriceMultiplier(
+  standard: TieredPricingConfig,
+  processing: TieredPricingConfig,
+): number | null {
+  if (standard.tiers.length !== processing.tiers.length) return null
+
+  let candidate: number | null = null
+  for (const [index, standardTier] of standard.tiers.entries()) {
+    const processingTier = processing.tiers[index]
+    if (standardTier.up_to !== processingTier?.up_to) return null
+
+    for (const field of TOKEN_PRICE_FIELDS) {
+      const standardPrice = standardTier[field]
+      const processingPrice = processingTier[field]
+      if (standardPrice === undefined || processingPrice === undefined) {
+        if (standardPrice !== processingPrice) return null
+        continue
+      }
+      if (standardPrice === 0) {
+        if (processingPrice !== 0) return null
+        continue
+      }
+
+      const ratio = processingPrice / standardPrice
+      if (!Number.isFinite(ratio) || ratio < 0) return null
+      if (candidate === null) candidate = ratio
+      if (Math.abs(processingPrice - standardPrice * candidate) > 1e-9) return null
+    }
+  }
+  return candidate
+}
+
 export function resolveModelsDevTieredPricing(
  _providerId: string,
  _modelId: string,
  cost: unknown,
+  experimentalModes?: unknown,
 ): TieredPricingConfig | null {
  // Provider/model identities must never inject local prices over the fetched catalog.
-  return buildModelsDevTieredPricing(cost)
+  const standard = buildModelsDevTieredPricing(cost)
+  if (!standard || !isRecord(experimentalModes)) return standard
+
+  const processingTierEntries: Array<[string, NonNullable<TieredPricingConfig['processing_tiers']>[string]]> = []
+  const seenProcessingTiers = new Set<string>()
+  for (const [modeKey, rawMode] of Object.entries(experimentalModes)) {
+    if (!isRecord(rawMode)) continue
+    const modePricing = buildModelsDevTieredPricing(rawMode.cost)
+    if (!modePricing) continue
+
+    const provider = isRecord(rawMode.provider) ? rawMode.provider : null
+    const body = provider && isRecord(provider.body) ? provider.body : null
+    // Anthropic Fast is expressed with `speed=fast`. A provider body may also carry a
+    // `service_tier` fact (commonly `default`/`standard`), but runtime settlement deliberately
+    // gives Fast speed precedence, so catalog import must resolve the same processing-tier key.
+    const mappedProcessingTier = typeof body?.speed === 'string'
+      && body.speed.trim().toLowerCase() === 'fast'
+      ? body.speed
+      : typeof body?.service_tier === 'string'
+        ? body.service_tier
+        : null
+    const normalizedModeKey = modeKey.trim().toLowerCase()
+    const rawProcessingTier = mappedProcessingTier
+      ?? (PROCESSING_MODE_FALLBACK_KEYS.has(normalizedModeKey) ? normalizedModeKey : '')
+    const processingTier = rawProcessingTier.trim().toLowerCase()
+    if (
+      !processingTier
+      || processingTier.length > 64
+      || ['auto', 'default', 'standard'].includes(processingTier)
+      || seenProcessingTiers.has(processingTier)
+    ) {
+      continue
+    }
+
+    const multiplier = uniformPriceMultiplier(standard, modePricing)
+    seenProcessingTiers.add(processingTier)
+    processingTierEntries.push([processingTier, multiplier === null
+      ? { tiers: modePricing.tiers }
+      : { price_multiplier: multiplier }])
+  }
+  if (processingTierEntries.length === 0) return standard
+  return {
+    ...standard,
+    processing_tiers: Object.fromEntries(processingTierEntries),
+  }
 }
@@ -36,8 +36,21 @@ export interface ModelsDevModel {
  last_updated?: string
  input?: string[] // 输入模态: text, image, audio, video, pdf
  output?: string[] // 输出模态: text, image, audio
+  modalities?: {
+    input?: string[]
+    output?: string[]
+  }
  open_weights?: boolean
  cost?: ModelsDevCost
+  experimental?: {
+    modes?: Record<string, {
+      cost?: ModelsDevCost
+      provider?: {
+        body?: Record<string, unknown>
+        headers?: Record<string, string>
+      }
+    }>
+  }
  limit?: ModelsDevLimit
  deprecated?: boolean
 }
@@ -166,7 +179,14 @@ export async function getModelsDevList(officialOnly: boolean = true): Promise<Mo
      if (!provider.models) continue

      for (const [modelId, model] of Object.entries(provider.models)) {
-        const tieredPricing = resolveModelsDevTieredPricing(providerId, modelId, model.cost)
+        const inputModalities = model.modalities?.input ?? model.input
+        const outputModalities = model.modalities?.output ?? model.output
+        const tieredPricing = resolveModelsDevTieredPricing(
+          providerId,
+          modelId,
+          model.cost,
+          model.experimental?.modes,
+        )
        const basePricingTier = tieredPricing?.tiers[0]
        items.push({
          providerId,
@@ -179,7 +199,7 @@ export async function getModelsDevList(officialOnly: boolean = true): Promise<Mo
          tieredPricing: tieredPricing ?? undefined,
          contextLimit: model.limit?.context,
          outputLimit: model.limit?.output,
-          supportsVision: model.input?.includes('image'),
+          supportsVision: inputModalities?.includes('image'),
          supportsToolCall: model.tool_call,
          supportsReasoning: model.reasoning,
          supportsStructuredOutput: model.structured_output,
@@ -194,8 +214,8 @@ export async function getModelsDevList(officialOnly: boolean = true): Promise<Mo
          // display_metadata 相关字段
          knowledgeCutoff: model.knowledge,
          releaseDate: model.release_date,
-          inputModalities: model.input,
-          outputModalities: model.output,
+          inputModalities,
+          outputModalities,
        })
      }
    }