feat: model-aware pricing with per-model cost breakdown

CHANGELOG.md

@@ -9,6 +9,19 @@ versioning follows [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
 ## [Unreleased]
 
+### Added
+
+- **Model-aware pricing** (`claudeStatus.pricing.useModelPricing`, default `true`) —
+  cost is now computed per JSONL entry using built-in rates for the entry's
+  `message.model` (Opus 4.5+ $5/$25, legacy Opus ≤4.1 $15/$75, Sonnet $3/$15,
+  Haiku $1/$5, Fable/Mythos 5 $10/$50 per 1M tokens, with matching cache rates).
+  Previously every entry was priced at the flat Sonnet-based `claudeStatus.pricing.*`
+  rates, overestimating Haiku usage 3× and underestimating Opus usage up to 5×.
+  Unrecognized models — and all entries when the setting is disabled — fall back
+  to the manual `claudeStatus.pricing.*` rates.
+- **Per-model cost breakdown in dashboard** — the token breakdown section now lists
+  the 5h-window cost per model (e.g. `sonnet-4-5: $0.42`, `haiku-4-5: $0.03`).
+
 ### Performance
 
 - **Skip stale JSONL files by mtime** — the 60-second status bar poll and the

README.ja.md

@@ -224,6 +224,7 @@ npm run package       # → vscode-claude-status-*.vsix
 | `claudeStatus.heatmap.days` | `30 \| 60 \| 90` | `90` | 使用ヒートマップに表示する日数 |
 | `claudeStatus.credentials.path` | `string \| null` | `null` | カスタム認証情報ファイルパス |
 | `claudeStatus.claudeProvider` | `"auto"` \| `"claude-ai"` \| `"aws-bedrock"` \| `"api-key"` | `"auto"` | プロバイダータイプ（自動検出または明示的指定） |
+| `claudeStatus.pricing.useModelPricing` | `boolean` | `true` | モデル別の組み込み料金（Opus/Sonnet/Haiku）で正確なコストを計算。下記の手動料金はフォールバック |
 | `claudeStatus.pricing.inputPerMillion` | `number` | `3.00` | 入力トークン 1M あたりの USD 単価 |
 | `claudeStatus.pricing.outputPerMillion` | `number` | `15.00` | 出力トークン 1M あたりの USD 単価 |
 | `claudeStatus.pricing.cacheReadPerMillion` | `number` | `0.30` | キャッシュ読み取りトークン 1M あたりの USD 単価 |

README.md

@@ -244,6 +244,7 @@ All settings are under the `claudeStatus` namespace in VS Code Settings.
 | `claudeStatus.heatmap.days` | `30 \| 60 \| 90` | `90` | Days shown in usage heatmap |
 | `claudeStatus.credentials.path` | `string \| null` | `null` | Custom credentials file path |
 | `claudeStatus.claudeProvider` | `"auto"` \| `"claude-ai"` \| `"aws-bedrock"` \| `"api-key"` | `"auto"` | Provider type (auto-detect or explicit) |
+| `claudeStatus.pricing.useModelPricing` | `boolean` | `true` | Use built-in per-model rates (Opus/Sonnet/Haiku) for accurate costs; manual rates below are the fallback |
 | `claudeStatus.pricing.inputPerMillion` | `number` | `3.00` | USD per 1M input tokens |
 | `claudeStatus.pricing.outputPerMillion` | `number` | `15.00` | USD per 1M output tokens |
 | `claudeStatus.pricing.cacheReadPerMillion` | `number` | `0.30` | USD per 1M cache-read tokens |

README.zh.md

@@ -219,6 +219,7 @@ npm run package       # → vscode-claude-status-*.vsix
 | `claudeStatus.heatmap.days` | `30 \| 60 \| 90` | `90` | 使用热力图显示的天数 |
 | `claudeStatus.credentials.path` | `string \| null` | `null` | 自定义凭证文件路径 |
 | `claudeStatus.claudeProvider` | `"auto"` \| `"claude-ai"` \| `"aws-bedrock"` \| `"api-key"` | `"auto"` | 提供商类型（自动检测或显式指定） |
+| `claudeStatus.pricing.useModelPricing` | `boolean` | `true` | 使用内置按模型费率（Opus/Sonnet/Haiku）以获得准确成本；下方手动费率为回退值 |
 | `claudeStatus.pricing.inputPerMillion` | `number` | `3.00` | 每百万输入令牌的 USD 单价 |
 | `claudeStatus.pricing.outputPerMillion` | `number` | `15.00` | 每百万输出令牌的 USD 单价 |
 | `claudeStatus.pricing.cacheReadPerMillion` | `number` | `0.30` | 每百万缓存读取令牌的 USD 单价 |

docs/DATA.md

@@ -76,8 +76,31 @@ Always keep a margin on the cutoff — never filter with the exact window start.
 
 ### Cost Calculation
 
+#### Model-aware pricing (default)
+
+Each JSONL entry carries `message.model`. When `claudeStatus.pricing.useModelPricing`
+is `true` (default), the cost of each entry is computed with the built-in rate for
+its model family (`resolveModelPricing()` in `jsonlReader.ts`). Rules are matched
+in order — first match wins:
+
+| Match (case-insensitive) | Model family | Input | Output | Cache read | Cache write |
+|--------------------------|--------------|-------|--------|------------|-------------|
+| `fable`, `mythos` | Fable 5 / Mythos 5 | $10.00 | $50.00 | $1.00 | $12.50 |
+| `claude-3-opus`, `opus-4-0`, `opus-4-1`, `opus-4-2025` | Legacy Opus (≤ 4.1) | $15.00 | $75.00 | $1.50 | $18.75 |
+| `opus` | Opus 4.5+ | $5.00 | $25.00 | $0.50 | $6.25 |
+| `haiku` | Haiku | $1.00 | $5.00 | $0.10 | $1.25 |
+| `sonnet` | Sonnet | $3.00 | $15.00 | $0.30 | $3.75 |
+
+Unrecognized models (e.g. `<synthetic>`) fall back to the manual rates below.
+Cost aggregation also produces `costByModel5h` (model id → USD in the 5h window),
+shown in the dashboard token breakdown.
+
+#### Manual rates (fallback / opt-out)
+
 Default rates are based on Claude Sonnet 4.x pricing.
-**All four rates are user-configurable** via `claudeStatus.pricing.*` settings.
+**All four rates are user-configurable** via `claudeStatus.pricing.*` settings and
+apply to **every** entry when `useModelPricing` is `false`, or to entries whose
+model is not recognized.
 
 | Token type | Setting key | Default (USD / 1M) |
 |------------|-------------|-------------------|

docs/SETTINGS.md

@@ -163,7 +163,13 @@ What the status bar and dashboard show, depending on your Claude plan and key se
     "description": "Claude provider type. Controls whether rate limit percentages or cost is shown in the status bar."
   },
 
-  // --- Token pricing (user-adjustable when Anthropic changes rates) ---
+  // --- Token pricing ---
+  "claudeStatus.pricing.useModelPricing": {
+    "type": "boolean",
+    "default": true,
+    "description": "Use built-in per-model rates (Opus/Sonnet/Haiku) based on each entry's model field for more accurate cost estimates. When disabled, or when a model is not recognized, the manual claudeStatus.pricing.* rates are used for every entry."
+  },
+  // --- Manual token pricing (fallback for unknown models; all entries when useModelPricing is false) ---
   "claudeStatus.pricing.inputPerMillion": {
     "type": "number",
     "default": 3.00,

l10n/bundle.l10n.ja.json

@@ -68,6 +68,7 @@
   "Output": "出力",
   "Cache read": "キャッシュ読み取り",
   "Cache create": "キャッシュ作成",
+  "By model (5h)": "モデル別（5h）",
   "Good! Cache is saving cost.": "良好！キャッシュがコストを節約しています。",
   "Low cache reuse.": "キャッシュの再利用が少ないです。",
   "No cache reads in this window.": "このウィンドウ内にキャッシュ読み取りはありません。",

l10n/bundle.l10n.zh-cn.json

@@ -68,6 +68,7 @@
   "Output": "输出",
   "Cache read": "缓存读取",
   "Cache create": "缓存创建",
+  "By model (5h)": "按模型（5h）",
   "Good! Cache is saving cost.": "很好！缓存正在节省费用。",
   "Low cache reuse.": "缓存复用率低。",
   "No cache reads in this window.": "此窗口内无缓存读取。",

package.json

@@ -195,6 +195,11 @@
           "default": "auto",
           "description": "%config.claudeProvider.desc%"
         },
+        "claudeStatus.pricing.useModelPricing": {
+          "type": "boolean",
+          "default": true,
+          "description": "%config.pricing.useModelPricing.desc%"
+        },
         "claudeStatus.pricing.inputPerMillion": {
           "type": "number",
           "default": 3.0,

package.nls.ja.json

@@ -26,6 +26,7 @@
   "config.claudeProvider.claudeAi": "Claude.ai サブスクリプション — レート制限の利用率を表示します（5h ウィンドウ、利用可能な場合は 7d ウィンドウ）。",
   "config.claudeProvider.awsBedrock": "AWS Bedrock — レート制限なし、トークンコストのみ表示。",
   "config.claudeProvider.apiKey": "Anthropic API キー — レート制限なし、トークンコストのみ表示。",
+  "config.pricing.useModelPricing.desc": "各エントリの model フィールドに基づき、モデル別の組み込み料金（Opus/Sonnet/Haiku）を使用してより正確なコストを推定します。無効の場合、またはモデルを認識できない場合は、手動設定の claudeStatus.pricing.* 料金がすべてのエントリに使用されます。",
   "config.pricing.inputPerMillion.desc": "入力トークン 1M あたりのコスト（USD）。デフォルト: $3.00（Claude Sonnet 4.x）。Anthropic が料金を変更した場合は更新してください。",
   "config.pricing.outputPerMillion.desc": "出力トークン 1M あたりのコスト（USD）。デフォルト: $15.00（Claude Sonnet 4.x）。Anthropic が料金を変更した場合は更新してください。",
   "config.pricing.cacheReadPerMillion.desc": "キャッシュ読み取りトークン 1M あたりのコスト（USD）。デフォルト: $0.30（Claude Sonnet 4.x）。Anthropic が料金を変更した場合は更新してください。",

package.nls.json

@@ -26,6 +26,7 @@
   "config.claudeProvider.claudeAi": "Claude.ai subscription — shows rate limit utilization (5h window, 7d window if available).",
   "config.claudeProvider.awsBedrock": "AWS Bedrock — no rate limits, shows token cost only.",
   "config.claudeProvider.apiKey": "Anthropic API key — no rate limits, shows token cost only.",
+  "config.pricing.useModelPricing.desc": "Use built-in per-model rates (Opus/Sonnet/Haiku) based on each entry's model field for more accurate cost estimates. When disabled, or when a model is not recognized, the manual claudeStatus.pricing.* rates are used for every entry.",
   "config.pricing.inputPerMillion.desc": "Cost per 1M input tokens in USD. Default: $3.00 (Claude Sonnet 4.x). Update this if Anthropic changes pricing.",
   "config.pricing.outputPerMillion.desc": "Cost per 1M output tokens in USD. Default: $15.00 (Claude Sonnet 4.x). Update this if Anthropic changes pricing.",
   "config.pricing.cacheReadPerMillion.desc": "Cost per 1M cache-read tokens in USD. Default: $0.30 (Claude Sonnet 4.x). Update this if Anthropic changes pricing.",

package.nls.zh-cn.json

@@ -26,6 +26,7 @@
   "config.claudeProvider.claudeAi": "Claude.ai 订阅 — 显示速率限制利用率（5h 窗口，如可用则显示 7d 窗口）。",
   "config.claudeProvider.awsBedrock": "AWS Bedrock — 无速率限制，仅显示令牌费用。",
   "config.claudeProvider.apiKey": "Anthropic API 密钥 — 无速率限制，仅显示令牌费用。",
+  "config.pricing.useModelPricing.desc": "根据每条记录的 model 字段使用内置的按模型费率（Opus/Sonnet/Haiku），以获得更准确的成本估算。禁用时或无法识别模型时，所有记录都将使用手动设置的 claudeStatus.pricing.* 费率。",
   "config.pricing.inputPerMillion.desc": "每百万输入令牌的费用（USD）。默认：$3.00（Claude Sonnet 4.x）。Anthropic 调整定价后请更新此值。",
   "config.pricing.outputPerMillion.desc": "每百万输出令牌的费用（USD）。默认：$15.00（Claude Sonnet 4.x）。Anthropic 调整定价后请更新此值。",
   "config.pricing.cacheReadPerMillion.desc": "每百万缓存读取令牌的费用（USD）。默认：$0.30（Claude Sonnet 4.x）。Anthropic 调整定价后请更新此值。",

src/config.ts

@@ -1,6 +1,6 @@
 import * as vscode from 'vscode';
 import type { ClaudeProvider } from './data/apiClient';
-import type { TokenPricing } from './data/jsonlReader';
+import type { CostOptions, TokenPricing } from './data/jsonlReader';
 
 export class ExtensionConfig {
   private get cfg() {
@@ -81,6 +81,17 @@ export class ExtensionConfig {
     };
   }
 
+  get useModelPricing(): boolean {
+    return this.cfg.get('pricing.useModelPricing', true);
+  }
+
+  get costOptions(): CostOptions {
+    return {
+      pricing: this.tokenPricing,
+      useModelPricing: this.useModelPricing,
+    };
+  }
+
   async setDisplayMode(mode: 'percent' | 'cost'): Promise<void> {
     await this.cfg.update('displayMode', mode, vscode.ConfigurationTarget.Global);
   }

src/data/dataManager.ts

@@ -27,6 +27,7 @@ export interface ClaudeUsageData {
   tokensOut5h: number
   tokensCacheRead5h: number
   tokensCacheCreate5h: number
+  costByModel5h: Record<string, number>
 
   // Rate limit metadata
   has7dLimit: boolean      // false for plans without a 7d window or non-Claude.ai providers
@@ -64,7 +65,7 @@ export class DataManager {
   }
 
   async getUsageData(forceRefresh = false): Promise<ClaudeUsageData> {
-    const [localUsage, cache] = await Promise.all([readAllUsage(config.tokenPricing), readCache()]);
+    const [localUsage, cache] = await Promise.all([readAllUsage(config.costOptions), readCache()]);
 
     // Determine provider type (user config or auto-detection)
     const configuredProvider = config.claudeProvider;
@@ -154,7 +155,7 @@ export class DataManager {
 
   async refreshProjectCosts(): Promise<void> {
     try {
-      this.lastProjectCosts = await getAllProjectCosts(config.tokenPricing);
+      this.lastProjectCosts = await getAllProjectCosts(config.costOptions);
     } catch {
       this.lastProjectCosts = [];
     }
@@ -226,6 +227,7 @@ export class DataManager {
         this.lastData.cost5h,
         this.lastData.costDay,
         config.dailyBudget,
+        config.costOptions,
       );
       this.lastPrediction = prediction;
       return prediction;
@@ -244,7 +246,7 @@ export class DataManager {
       return this.lastHeatmapData;
     }
     try {
-      const data = await computeHeatmapData(config.heatmapDays);
+      const data = await computeHeatmapData(config.heatmapDays, config.costOptions);
       this.lastHeatmapData = data;
       this.heatmapComputedAt = now;
       return data;

src/data/jsonlReader.ts

@@ -21,6 +21,7 @@ interface JsonlEntry {
   requestId?: string
   message?: {
     id?: string
+    model?: string
     usage?: TokenUsage
   }
 }
@@ -33,6 +34,7 @@ export interface AggregatedUsage {
   tokensOut5h: number
   tokensCacheRead5h: number
   tokensCacheCreate5h: number
+  costByModel5h: Record<string, number>  // model id → cost in the 5h window
 }
 
 export interface TokenPricing {
@@ -49,6 +51,51 @@ export const DEFAULT_PRICING: TokenPricing = {
   cacheCreatePerMillion: 3.75,
 };
 
+// Built-in per-model rates (USD per 1M tokens), matched against the JSONL
+// entry's message.model. Anthropic's published pricing: cache read = 0.1x
+// input, 5-minute cache write = 1.25x input. Order matters — first match wins,
+// so legacy patterns must precede their generic family pattern.
+const MODEL_PRICING_RULES: ReadonlyArray<{ pattern: RegExp; pricing: TokenPricing }> = [
+  // Fable 5 / Mythos 5: $10 / $50
+  { pattern: /fable|mythos/i, pricing: { inputPerMillion: 10.00, outputPerMillion: 50.00, cacheReadPerMillion: 1.00, cacheCreatePerMillion: 12.50 } },
+  // Legacy Opus (3.x, 4.0, 4.1): $15 / $75
+  { pattern: /claude-3-opus|opus-4-0|opus-4-1|opus-4-2025/i, pricing: { inputPerMillion: 15.00, outputPerMillion: 75.00, cacheReadPerMillion: 1.50, cacheCreatePerMillion: 18.75 } },
+  // Opus 4.5 and later: $5 / $25
+  { pattern: /opus/i, pricing: { inputPerMillion: 5.00, outputPerMillion: 25.00, cacheReadPerMillion: 0.50, cacheCreatePerMillion: 6.25 } },
+  // Haiku (4.5): $1 / $5
+  { pattern: /haiku/i, pricing: { inputPerMillion: 1.00, outputPerMillion: 5.00, cacheReadPerMillion: 0.10, cacheCreatePerMillion: 1.25 } },
+  // Sonnet (3.x–5): $3 / $15
+  { pattern: /sonnet/i, pricing: { inputPerMillion: 3.00, outputPerMillion: 15.00, cacheReadPerMillion: 0.30, cacheCreatePerMillion: 3.75 } },
+];
+
+// Returns the built-in rate for a recognized model, or the fallback (the
+// user-configured claudeStatus.pricing.* rates) when the model is unknown
+export function resolveModelPricing(model: string | undefined, fallback: TokenPricing): TokenPricing {
+  if (model) {
+    for (const rule of MODEL_PRICING_RULES) {
+      if (rule.pattern.test(model)) { return rule.pricing; }
+    }
+  }
+  return fallback;
+}
+
+export interface CostOptions {
+  pricing: TokenPricing      // manual rates; also the fallback for unknown models
+  useModelPricing: boolean   // when true, recognized models use built-in rates
+}
+
+export const DEFAULT_COST_OPTIONS: CostOptions = {
+  pricing: DEFAULT_PRICING,
+  useModelPricing: true,
+};
+
+export function entryCost(entry: UsageEntry, options: CostOptions = DEFAULT_COST_OPTIONS): number {
+  const pricing = options.useModelPricing
+    ? resolveModelPricing(entry.model, options.pricing)
+    : options.pricing;
+  return calculateCost(entry.usage, pricing);
+}
+
 export function calculateCost(usage: TokenUsage, pricing: TokenPricing = DEFAULT_PRICING): number {
   return (
     ((usage.input_tokens || 0) / 1_000_000) * pricing.inputPerMillion +
@@ -148,6 +195,7 @@ export interface UsageEntry {
   timestamp: number  // ms since epoch, guaranteed valid
   usage: TokenUsage  // all four fields normalized to numbers
   cwd?: string
+  model?: string     // e.g. "claude-sonnet-4-5-20250929"
 }
 
 export async function readUsageEntries(filePath: string): Promise<UsageEntry[]> {
@@ -167,12 +215,13 @@ export async function readUsageEntries(filePath: string): Promise<UsageEntry[]>
         cache_creation_input_tokens: rawUsage.cache_creation_input_tokens || 0,
       },
       cwd: typeof entry.cwd === 'string' ? entry.cwd : undefined,
+      model: typeof entry.message?.model === 'string' ? entry.message.model : undefined,
     });
   }
   return result;
 }
 
-export async function readAllUsage(pricing: TokenPricing = DEFAULT_PRICING): Promise<AggregatedUsage> {
+export async function readAllUsage(options: CostOptions = DEFAULT_COST_OPTIONS): Promise<AggregatedUsage> {
   const now = Date.now();
   const window5h = 5 * 3600 * 1000;
   const window7d = 7 * 24 * 3600 * 1000;
@@ -187,6 +236,7 @@ export async function readAllUsage(pricing: TokenPricing = DEFAULT_PRICING): Pro
     tokensOut5h: 0,
     tokensCacheRead5h: 0,
     tokensCacheCreate5h: 0,
+    costByModel5h: {},
   };
 
   // Skip files untouched for over 7 days (+1h margin) — they cannot contain
@@ -197,7 +247,7 @@ export async function readAllUsage(pricing: TokenPricing = DEFAULT_PRICING): Pro
     const entries = await readUsageEntries(file);
     for (const entry of entries) {
       const { timestamp: ts, usage } = entry;
-      const cost = calculateCost(usage, pricing);
+      const cost = entryCost(entry, options);
 
       const age = now - ts;
       if (age <= window7d) {
@@ -212,6 +262,9 @@ export async function readAllUsage(pricing: TokenPricing = DEFAULT_PRICING): Pro
         result.tokensOut5h += usage.output_tokens || 0;
         result.tokensCacheRead5h += usage.cache_read_input_tokens || 0;
         result.tokensCacheCreate5h += usage.cache_creation_input_tokens || 0;
+        if (entry.model) {
+          result.costByModel5h[entry.model] = (result.costByModel5h[entry.model] || 0) + cost;
+        }
       }
     }
   }