✨ feat: 完善 Embed 并支持 GPT-5.6 定价

docs/pricing.md

@@ -40,6 +40,7 @@ packages/shared/src/pricing/
 | `currency: 'USD' \| 'CNY'` | 价格币种。Worker 端结算时按 `catalog.fx` 折算到 USD |
 | `input_per_million` / `output_per_million` | 基础单价 / 1M tokens |
 | `cached_input_per_million` | cache hit 价（Anthropic 叫 cache_read，Kimi 叫缓存命中） |
+| `cache_write_per_million` | 不区分缓存时长的通用 cache write 价（如 OpenAI GPT-5.6） |
 | `cache_write_5m_per_million` / `cache_write_1h_per_million` | Anthropic 风格 prompt caching write |
 | `tiers?: PricingTier[]` | 阶梯定价：按 input token 数命中不同档位（Qwen / Gemini 2.5 Pro / GLM 等） |
 | `effective_from` / `effective_to` | 价格生效区间（审计用） |
@@ -51,6 +52,7 @@ packages/shared/src/pricing/
 
 ```ts
 'claude-opus-4-7-20260201': 'claude-opus-4-7'  // 带日期后缀的版本号
+'gpt-5.6': 'gpt-5.6-sol'                       // 官方系列别名 → Sol
 'codex-auto-review': 'gpt-5.4'                  // 工具内部模型 → 实际推理模型
 ```
 
@@ -82,6 +84,8 @@ tiers: [
 ]
 ```
 
+Codex scanner 会先按单次请求命中阶梯并累计 `costUSD`，同时携带 `pricingVersion`，确保同一天混合长短请求时仍按各自档位精确计费。CLI / Worker 仅在版本与当前 catalog 一致时采用该成本；旧版、版本不匹配或不含逐请求成本的聚合 breakdown 会用 `totalInputTokens / eventCount` 估算档位，并将 `costStatus` 标为 `estimated`。
+
 > **限制**：Gemini 2.5 Flash 等模型还按 **output** 长度分档；GLM-4.7 同样有 input × output 双维度。当前实现只按 input 命中，output 分档暂用保守取低档（标了 `notes`）。
 
 ## 贡献新模型 / 修订单价

packages/cli/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@aiusage/cli",
-  "version": "1.7.1",
+  "version": "1.7.2",
   "description": "Track and sync AI tool token usage across devices.",
   "license": "MIT",
   "private": false,

packages/cli/pricing-catalog.json

@@ -1,5 +1,5 @@
 {
-  "version": "2026-06-22-codex-tier-v1",
+  "version": "2026-07-10-gpt-5.6-v1",
   "fx": {
     "CNY": 7.2
   },
@@ -9,6 +9,7 @@
     "claude-opus-4-6-20250301": "claude-opus-4-6",
     "claude-haiku-4-5-20251001": "claude-haiku-4-5",
     "claude-sonnet-4.6": "claude-sonnet-4-6",
+    "gpt-5.6": "gpt-5.6-sol",
     "codex-auto-review": "gpt-5.4"
   },
   "providers": {
@@ -112,23 +113,134 @@
     "openai": {
       "codex": {
         "models": {
+          "gpt-5.6-sol": {
+            "currency": "USD",
+            "notes": "prompts over 272K input tokens bill the full request at long-context rates",
+            "input_per_million": 5,
+            "cached_input_per_million": 0.5,
+            "cache_write_per_million": 6.25,
+            "output_per_million": 30,
+            "tiers": [
+              {
+                "threshold": 272000,
+                "input_per_million": 5,
+                "cached_input_per_million": 0.5,
+                "cache_write_per_million": 6.25,
+                "output_per_million": 30
+              },
+              {
+                "input_per_million": 10,
+                "cached_input_per_million": 1,
+                "cache_write_per_million": 12.5,
+                "output_per_million": 45
+              }
+            ]
+          },
+          "gpt-5.6-terra": {
+            "currency": "USD",
+            "notes": "prompts over 272K input tokens bill the full request at long-context rates",
+            "input_per_million": 2.5,
+            "cached_input_per_million": 0.25,
+            "cache_write_per_million": 3.125,
+            "output_per_million": 15,
+            "tiers": [
+              {
+                "threshold": 272000,
+                "input_per_million": 2.5,
+                "cached_input_per_million": 0.25,
+                "cache_write_per_million": 3.125,
+                "output_per_million": 15
+              },
+              {
+                "input_per_million": 5,
+                "cached_input_per_million": 0.5,
+                "cache_write_per_million": 6.25,
+                "output_per_million": 22.5
+              }
+            ]
+          },
+          "gpt-5.6-luna": {
+            "currency": "USD",
+            "notes": "prompts over 272K input tokens bill the full request at long-context rates",
+            "input_per_million": 1,
+            "cached_input_per_million": 0.1,
+            "cache_write_per_million": 1.25,
+            "output_per_million": 6,
+            "tiers": [
+              {
+                "threshold": 272000,
+                "input_per_million": 1,
+                "cached_input_per_million": 0.1,
+                "cache_write_per_million": 1.25,
+                "output_per_million": 6
+              },
+              {
+                "input_per_million": 2,
+                "cached_input_per_million": 0.2,
+                "cache_write_per_million": 2.5,
+                "output_per_million": 9
+              }
+            ]
+          },
           "gpt-5.5": {
             "currency": "USD",
+            "notes": "prompts over 272K input tokens bill the full request at long-context rates",
             "input_per_million": 5,
             "cached_input_per_million": 0.5,
-            "output_per_million": 30
+            "output_per_million": 30,
+            "tiers": [
+              {
+                "threshold": 272000,
+                "input_per_million": 5,
+                "cached_input_per_million": 0.5,
+                "output_per_million": 30
+              },
+              {
+                "input_per_million": 10,
+                "cached_input_per_million": 1,
+                "output_per_million": 45
+              }
+            ]
           },
           "gpt-5.5-pro": {
             "currency": "USD",
+            "notes": "prompts over 272K input tokens bill the full request at long-context rates",
             "input_per_million": 30,
             "cached_input_per_million": null,
-            "output_per_million": 180
+            "output_per_million": 180,
+            "tiers": [
+              {
+                "threshold": 272000,
+                "input_per_million": 30,
+                "cached_input_per_million": null,
+                "output_per_million": 180
+              },
+              {
+                "input_per_million": 60,
+                "cached_input_per_million": null,
+                "output_per_million": 270
+              }
+            ]
           },
           "gpt-5.4": {
             "currency": "USD",
+            "notes": "prompts over 272K input tokens bill the full request at long-context rates",
             "input_per_million": 2.5,
             "cached_input_per_million": 0.25,
-            "output_per_million": 15
+            "output_per_million": 15,
+            "tiers": [
+              {
+                "threshold": 272000,
+                "input_per_million": 2.5,
+                "cached_input_per_million": 0.25,
+                "output_per_million": 15
+              },
+              {
+                "input_per_million": 5,
+                "cached_input_per_million": 0.5,
+                "output_per_million": 22.5
+              }
+            ]
           },
           "gpt-5.4-mini": {
             "currency": "USD",
@@ -144,9 +256,23 @@
           },
           "gpt-5.4-pro": {
             "currency": "USD",
+            "notes": "prompts over 272K input tokens bill the full request at long-context rates",
             "input_per_million": 30,
             "cached_input_per_million": null,
-            "output_per_million": 180
+            "output_per_million": 180,
+            "tiers": [
+              {
+                "threshold": 272000,
+                "input_per_million": 30,
+                "cached_input_per_million": null,
+                "output_per_million": 180
+              },
+              {
+                "input_per_million": 60,
+                "cached_input_per_million": null,
+                "output_per_million": 270
+              }
+            ]
           },
           "gpt-5.3-codex": {
             "currency": "USD",

packages/cli/src/__tests__/report.test.ts

@@ -95,6 +95,33 @@ describe('buildLocalReport', () => {
 });
 
 describe('calculateBreakdownCost', () => {
+  it('only trusts scanner cost when the selected catalog version matches', async () => {
+    const { calculateBreakdownCost } = await import('../report.js');
+    const { catalog } = await import('@aiusage/shared');
+    const breakdown = {
+      provider: 'openai' as const,
+      product: 'codex' as const,
+      channel: 'cli' as const,
+      model: 'gpt-5.6-sol',
+      project: '/tmp/project',
+      eventCount: 2,
+      inputTokens: 500_000,
+      cachedInputTokens: 0,
+      cacheWriteTokens: 0,
+      outputTokens: 20_000,
+      reasoningOutputTokens: 0,
+      costUSD: 5.25,
+      pricingVersion: catalog.version,
+    };
+
+    expect(calculateBreakdownCost(breakdown, new Set(), catalog)).toBe(5.25);
+
+    const warnings = new Set<string>();
+    const futureCatalog = { ...catalog, version: 'future-catalog' };
+    expect(calculateBreakdownCost(breakdown, warnings, futureCatalog)).toBe(3.1);
+    expect([...warnings]).toEqual(['gpt-5.6-sol 的阶梯价格已按每事件平均输入量估算。']);
+  });
+
   it('prices local Codex GPT-5.5 usage', async () => {
     const { calculateBreakdownCost } = await import('../report.js');
     const warnings = new Set<string>();
@@ -113,7 +140,7 @@ describe('calculateBreakdownCost', () => {
       reasoningOutputTokens: 0,
     }, warnings);
 
-    expect(cost).toBe(20.5);
+    expect(cost).toBe(33.5);
     expect([...warnings]).toEqual([]);
   });
 
@@ -135,8 +162,31 @@ describe('calculateBreakdownCost', () => {
       reasoningOutputTokens: 0,
     }, warnings);
 
-    expect(cost).toBe(17.75);
+    expect(cost).toBe(28);
     // codex-auto-review 是 catalog 里的显式 alias → gpt-5.4，按 exact 处理，不应有 warning
     expect([...warnings]).toEqual([]);
   });
+
+  it('estimates legacy aggregated GPT-5.6 usage from average per-event input', async () => {
+    const { calculateBreakdownCost } = await import('../report.js');
+    const warnings = new Set<string>();
+
+    const cost = calculateBreakdownCost({
+      provider: 'openai',
+      product: 'codex',
+      channel: 'cli',
+      model: 'gpt-5.6-sol',
+      project: '/tmp/project',
+      eventCount: 2,
+      inputTokens: 400_000,
+      cachedInputTokens: 0,
+      cacheWriteTokens: 0,
+      outputTokens: 20_000,
+      reasoningOutputTokens: 0,
+      costUSD: 0,
+    }, warnings);
+
+    expect(cost).toBe(2.6);
+    expect([...warnings]).toEqual(['gpt-5.6-sol 的阶梯价格已按每事件平均输入量估算。']);
+  });
 });

packages/cli/src/report.ts

@@ -2,7 +2,7 @@ import { readdir, readFile } from 'node:fs/promises';
 import { homedir } from 'node:os';
 import { basename, join } from 'node:path';
 import type { IngestBreakdown } from '@aiusage/shared';
-import { calculateCost, type PricingCatalog } from '@aiusage/shared';
+import { calculateCost, PRICING_VERSION, type PricingCatalog } from '@aiusage/shared';
 import { scanDates } from './scan.js';
 import { parseTs, dateKey } from './scanners/utils.js';
 import type { PricingInfo } from './pricing.js';
@@ -578,7 +578,10 @@ export function calculateBreakdownCost(
   warnings: Set<string>,
   pricingCatalog?: PricingCatalog,
 ): number {
-  if (breakdown.costUSD != null && breakdown.costUSD > 0) {
+  const effectivePricingVersion = pricingCatalog?.version ?? PRICING_VERSION;
+  const sourceCostMatchesCatalog =
+    breakdown.pricingVersion == null || breakdown.pricingVersion === effectivePricingVersion;
+  if (breakdown.costUSD != null && breakdown.costUSD > 0 && sourceCostMatchesCatalog) {
     return breakdown.costUSD;
   }
 
@@ -594,13 +597,18 @@ export function calculateBreakdownCost(
       cacheWrite1hTokens: breakdown.cacheWrite1hTokens,
       outputTokens: breakdown.outputTokens,
     },
-    pricingCatalog ? { catalog: pricingCatalog } : {},
+    {
+      ...(pricingCatalog ? { catalog: pricingCatalog } : {}),
+      requestCount: breakdown.eventCount,
+    },
   );
 
   if (result.costStatus === 'unavailable') {
     warnings.add(`${breakdown.provider}/${breakdown.product}/${breakdown.model} 暂无定价配置，已跳过成本估算。`);
   } else if (result.costStatus === 'estimated' && result.resolvedModel && result.resolvedModel !== breakdown.model) {
     warnings.add(`${breakdown.model} 已按 ${result.resolvedModel} 的公开单价估算。`);
+  } else if (result.costStatus === 'estimated' && result.matchedTierIndex !== undefined) {
+    warnings.add(`${breakdown.model} 的阶梯价格已按每事件平均输入量估算。`);
   }
 
   return result.estimatedCostUsd;

packages/cli/src/scanners/__tests__/codex.test.ts

@@ -124,7 +124,70 @@ describe('Fix 1: non-cached input cost formula', () => {
   });
 });
 
+describe('Codex tiered pricing', () => {
+  it('accumulates mixed short and long requests at their individual tiers', async () => {
+    const day = '2026-07-10';
+    const sessionDir = join(tmpDir, 'sessions', '2026', '07', '10');
+    const lines = [
+      ...tokenCountEvent(
+        `${day}T10:00:00.000Z`,
+        { input: 100_000, cached: 0, output: 10_000 },
+        { input: 100_000, cached: 0, output: 10_000 },
+        'gpt-5.6-sol',
+      ),
+      ...tokenCountEvent(
+        `${day}T10:05:00.000Z`,
+        { input: 400_000, cached: 0, output: 10_000 },
+        { input: 500_000, cached: 0, output: 20_000 },
+        'gpt-5.6-sol',
+      ),
+    ];
+    await writeSession(sessionDir, 'rollout-test.jsonl', lines);
+
+    const [result] = await scanCodex(day, tmpDir);
+    expect(result.eventCount).toBe(2);
+    expect(result.inputTokens).toBe(500_000);
+    expect(result.outputTokens).toBe(20_000);
+    expect(result.costUSD).toBeCloseTo(5.25, 4);
+    expect(result.pricingVersion).toMatch(/^2026-07-10/);
+  });
+});
+
 describe('Codex service tier', () => {
+  it('adds priority suffix for GPT-5.6 Codex usage', async () => {
+    const day = '2026-07-10';
+    await writeFile(join(tmpDir, 'config.toml'), 'service_tier = "priority"\n');
+    const sessionDir = join(tmpDir, 'sessions', '2026', '07', '10');
+    const events = tokenCountEvent(
+      `${day}T10:00:00.000Z`,
+      { input: 10000, cached: 8000, output: 500 },
+      { input: 10000, cached: 8000, output: 500 },
+      'gpt-5.6-sol',
+    );
+    await writeSession(sessionDir, 'rollout-test.jsonl', events);
+
+    const results = await scanCodex(day, tmpDir);
+    expect(results).toHaveLength(1);
+    expect(results[0].model).toBe('gpt-5.6-sol-priority');
+  });
+
+  it('does not add fast suffix to GPT-5.6 while Codex Fast is unsupported', async () => {
+    const day = '2026-07-10';
+    await writeFile(join(tmpDir, 'config.toml'), 'service_tier = "fast"\n');
+    const sessionDir = join(tmpDir, 'sessions', '2026', '07', '10');
+    const events = tokenCountEvent(
+      `${day}T10:00:00.000Z`,
+      { input: 10000, cached: 8000, output: 500 },
+      { input: 10000, cached: 8000, output: 500 },
+      'gpt-5.6-sol',
+    );
+    await writeSession(sessionDir, 'rollout-test.jsonl', events);
+
+    const results = await scanCodex(day, tmpDir);
+    expect(results).toHaveLength(1);
+    expect(results[0].model).toBe('gpt-5.6-sol');
+  });
+
   it('adds priority suffix for supported GPT-5.5 Codex usage', async () => {
     const day = '2026-06-22';
     await writeFile(join(tmpDir, 'config.toml'), 'service_tier = "priority"\n');