fix(pricing): GPT-5.3 Codex resolves to $0.00 — filter github_copilot + inject overrides

README.ja.md

@@ -115,7 +115,7 @@ AI支援開発の時代において、**トークンは新しいエネルギー*
   - リアルタイムフィルタリングとソート
   - ゼロフリッカーレンダリング（ネイティブZigエンジン）
 - **マルチプラットフォームサポート** - OpenCode、Claude Code、Codex CLI、Cursor IDE、Gemini CLI、Amp、Droid、OpenClaw、Pi全体の使用量追跡
-- **リアルタイム価格** - 1時間ディスクキャッシュ付きでLiteLLMから現在の価格を取得
+- **リアルタイム価格** - 1時間ディスクキャッシュ付きでLiteLLMから現在の価格を取得；OpenRouter自動フォールバックと新規モデル向けCursor価格サポート
 - **詳細な内訳** - 入力、出力、キャッシュ読み書き、推論トークン追跡
 - **ネイティブRustコア** - 10倍高速な処理のため、すべての解析と集計をRustで実行
 - **Web可視化** - 2Dと3Dビューのインタラクティブ貢献グラフ
@@ -295,7 +295,8 @@ tokscale pricing "claude-3-5-sonnet" --provider litellm
 3. **ティアサフィックス除去** - 品質ティアを削除（`gpt-5.2-xhigh` → `gpt-5.2`）
 4. **バージョン正規化** - バージョン形式を処理（`claude-3-5-sonnet` ↔ `claude-3.5-sonnet`）
 5. **プロバイダープレフィックスマッチング** - 一般的なプレフィックスを試行（`anthropic/`、`openai/`など）
-6. **ファジーマッチング** - 部分モデル名の単語境界マッチング
+6. **Cursorモデル価格** - LiteLLM/OpenRouterにまだ存在しないモデルのハードコード価格（例：`gpt-5.3-codex`）
+7. **ファジーマッチング** - 部分モデル名の単語境界マッチング
 
 **プロバイダー優先順位：**
 
@@ -911,6 +912,10 @@ model_changeイベントとアシスタントメッセージを含むセッシ
 
 Tokscaleは[LiteLLMの価格データベース](https://github.com/BerriAI/litellm/blob/main/model_prices_and_context_window.json)からリアルタイム価格を取得します。
 
+**ダイナミックフォールバック**: LiteLLMにまだ存在しないモデル（例：最近リリースされたモデル）は、[OpenRouterのエンドポイントAPI](https://openrouter.ai/docs/api/api-reference/endpoints/list-endpoints)から自動的に価格を取得します。
+
+**Cursorモデル価格**: LiteLLMとOpenRouterの両方にまだ存在しない最新モデル（例：`gpt-5.3-codex`）は、[Cursorモデルドキュメント](https://cursor.com/en-US/docs/models)から取得したハードコード価格を使用します。これらのオーバーライドはすべてのアップストリームソースの後、ファジーマッチングの前にチェックされるため、実際のアップストリーム価格が利用可能になると自動的に優先されます。
+
 **キャッシュ**: 価格データは1時間TTLでディスクにキャッシュされ、高速な起動を確保します：
 - LiteLLMキャッシュ: `~/.cache/tokscale/pricing-litellm.json`
 - OpenRouterキャッシュ: `~/.cache/tokscale/pricing-openrouter.json`（増分式、使用したモデルのみキャッシュ）

README.ko.md

@@ -115,7 +115,7 @@ AI 지원 개발 시대에 **토큰은 새로운 에너지**입니다. 토큰은
   - 실시간 필터링 및 정렬
   - 깜빡임 없는 렌더링 (네이티브 Zig 엔진)
 - **멀티 플랫폼 지원** - OpenCode, Claude Code, Codex CLI, Cursor IDE, Gemini CLI, Amp, Droid, OpenClaw, Pi 사용량 통합 추적
-- **실시간 가격 반영** - LiteLLM에서 최신 가격을 가져와(디스크 캐시 1시간) 비용 계산
+- **실시간 가격 반영** - LiteLLM에서 최신 가격을 가져와(디스크 캐시 1시간) 비용 계산; OpenRouter 자동 폴백 및 신규 모델용 Cursor 가격 지원
 - **상세 분석** - 입력, 출력, 캐시 읽기/쓰기, 추론 토큰까지 추적
 - **네이티브 Rust 코어** - 모든 파싱과 집계를 Rust로 처리해 최대 10배 빠른 성능
 - **웹 시각화** - 2D 및 3D 뷰의 인터랙티브 기여 그래프
@@ -294,7 +294,8 @@ tokscale pricing "claude-3-5-sonnet" --provider litellm
 3. **티어 접미사 제거** - 품질 티어 제거 (`gpt-5.2-xhigh` → `gpt-5.2`)
 4. **버전 정규화** - 버전 형식 처리 (`claude-3-5-sonnet` ↔ `claude-3.5-sonnet`)
 5. **프로바이더 접두사 매칭** - 일반 접두사 시도 (`anthropic/`, `openai/` 등)
-6. **퍼지 매칭** - 부분 모델 이름에 대한 단어 경계 매칭
+6. **Cursor 모델 가격** - LiteLLM/OpenRouter에 아직 없는 모델의 하드코딩 가격 (예: `gpt-5.3-codex`)
+7. **퍼지 매칭** - 부분 모델 이름에 대한 단어 경계 매칭
 
 **프로바이더 우선순위:**
 
@@ -910,6 +911,10 @@ model_change 이벤트와 어시스턴트 메시지가 포함된 세션 JSONL 
 
 Tokscale은 [LiteLLM의 가격 데이터베이스](https://github.com/BerriAI/litellm/blob/main/model_prices_and_context_window.json)에서 실시간 가격을 가져옵니다.
 
+**동적 폴백**: LiteLLM에 아직 없는 모델(예: 최근 출시된 모델)은 [OpenRouter의 엔드포인트 API](https://openrouter.ai/docs/api/api-reference/endpoints/list-endpoints)에서 자동으로 가격을 가져옵니다.
+
+**Cursor 모델 가격**: LiteLLM과 OpenRouter 모두에 없는 최신 모델(예: `gpt-5.3-codex`)은 [Cursor 모델 문서](https://cursor.com/en-US/docs/models)에서 가져온 하드코딩 가격을 사용합니다. 이 오버라이드는 모든 업스트림 소스 다음에, 퍼지 매칭 이전에 확인되므로 실제 업스트림 가격이 사용 가능해지면 자동으로 양보합니다.
+
 **캐싱**: 가격 데이터는 1시간 TTL로 디스크에 캐시되어 빠른 시작을 보장합니다:
 - LiteLLM 캐시: `~/.cache/tokscale/pricing-litellm.json`
 - OpenRouter 캐시: `~/.cache/tokscale/pricing-openrouter.json` (증분 방식, 사용한 모델만 캐시)

README.md

@@ -116,7 +116,7 @@ In the age of AI-assisted development, **tokens are the new energy**. They power
   - Real-time filtering and sorting
   - Zero flicker rendering (native Zig engine)
 - **Multi-platform support** - Track usage across OpenCode, Claude Code, Codex CLI, Cursor IDE, Gemini CLI, Amp, Droid, OpenClaw, and Pi
-- **Real-time pricing** - Fetches current pricing from LiteLLM with 1-hour disk cache; automatic OpenRouter fallback for new models
+- **Real-time pricing** - Fetches current pricing from LiteLLM with 1-hour disk cache; automatic OpenRouter fallback and Cursor model pricing for newly released models
 - **Detailed breakdowns** - Input, output, cache read/write, and reasoning token tracking
 - **Native Rust core** - All parsing and aggregation done in Rust for 10x faster processing
 - **Web visualization** - Interactive contribution graph with 2D and 3D views
@@ -311,7 +311,8 @@ The pricing lookup uses a multi-step resolution strategy:
 3. **Tier Suffix Stripping** - Removes quality tiers (`gpt-5.2-xhigh` → `gpt-5.2`)
 4. **Version Normalization** - Handles version formats (`claude-3-5-sonnet` ↔ `claude-3.5-sonnet`)
 5. **Provider Prefix Matching** - Tries common prefixes (`anthropic/`, `openai/`, etc.)
-6. **Fuzzy Matching** - Word-boundary matching for partial model names
+6. **Cursor Model Pricing** - Hardcoded pricing for models not yet in LiteLLM/OpenRouter (e.g., `gpt-5.3-codex`)
+7. **Fuzzy Matching** - Word-boundary matching for partial model names
 
 **Provider Preference:**
 
@@ -948,6 +949,8 @@ Tokscale fetches real-time pricing from [LiteLLM's pricing database](https://git
 
 **Dynamic Fallback**: For models not yet available in LiteLLM (e.g., recently released models), Tokscale automatically fetches pricing from [OpenRouter's endpoints API](https://openrouter.ai/docs/api/api-reference/endpoints/list-endpoints). This ensures you get accurate pricing from the model's author provider (e.g., Z.AI for glm-4.7) without waiting for LiteLLM updates.
 
+**Cursor Model Pricing**: For very recently released models not yet in either LiteLLM or OpenRouter (e.g., `gpt-5.3-codex`), Tokscale includes hardcoded pricing sourced from [Cursor's model docs](https://cursor.com/en-US/docs/models). These overrides are checked after all upstream sources but before fuzzy matching, so they automatically yield once real upstream pricing becomes available.
+
 **Caching**: Pricing data is cached to disk with 1-hour TTL for fast startup:
 - LiteLLM cache: `~/.cache/tokscale/pricing-litellm.json`
 - OpenRouter cache: `~/.cache/tokscale/pricing-openrouter.json` (incremental, caches only models you've used)

README.zh-cn.md

@@ -115,7 +115,7 @@
   - 实时筛选和排序
   - 零闪烁渲染（原生 Zig 引擎）
 - **多平台支持** - 跟踪 OpenCode、Claude Code、Codex CLI、Cursor IDE、Gemini CLI、Amp、Droid、OpenClaw 和 Pi 的使用情况
-- **实时定价** - 从 LiteLLM 获取当前价格，带 1 小时磁盘缓存
+- **实时定价** - 从 LiteLLM 获取当前价格，带 1 小时磁盘缓存；OpenRouter 自动回退和新模型的 Cursor 定价支持
 - **详细分解** - 输入、输出、缓存读写和推理 Token 跟踪
 - **原生 Rust 核心** - 所有解析和聚合在 Rust 中完成，处理速度提升 10 倍
 - **Web 可视化** - 带 2D 和 3D 视图的交互式贡献图
@@ -295,7 +295,8 @@ tokscale pricing "claude-3-5-sonnet" --provider litellm
 3. **层级后缀剥离** - 移除质量层级（`gpt-5.2-xhigh` → `gpt-5.2`）
 4. **版本标准化** - 处理版本格式（`claude-3-5-sonnet` ↔ `claude-3.5-sonnet`）
 5. **提供商前缀匹配** - 尝试常见前缀（`anthropic/`、`openai/` 等）
-6. **模糊匹配** - 部分模型名称的词边界匹配
+6. **Cursor 模型定价** - LiteLLM/OpenRouter 中尚未收录的模型的硬编码定价（例如：`gpt-5.3-codex`）
+7. **模糊匹配** - 部分模型名称的词边界匹配
 
 **提供商优先级：**
 
@@ -911,6 +912,10 @@ Cursor 数据使用您的会话令牌从 Cursor API 获取并本地缓存。运
 
 Tokscale 从 [LiteLLM 的价格数据库](https://github.com/BerriAI/litellm/blob/main/model_prices_and_context_window.json)获取实时价格。
 
+**动态回退**：对于 LiteLLM 中尚未收录的模型（例如最近发布的模型），Tokscale 会自动从 [OpenRouter 的端点 API](https://openrouter.ai/docs/api/api-reference/endpoints/list-endpoints) 获取定价。
+
+**Cursor 模型定价**：对于 LiteLLM 和 OpenRouter 中都尚未收录的最新模型（例如 `gpt-5.3-codex`），Tokscale 使用从 [Cursor 模型文档](https://cursor.com/en-US/docs/models)获取的硬编码定价。这些覆盖在所有上游来源之后、模糊匹配之前检查，因此当真正的上游定价可用时会自动让步。
+
 **缓存**：价格数据以 1 小时 TTL 缓存到磁盘，确保快速启动：
 - LiteLLM 缓存：`~/.cache/tokscale/pricing-litellm.json`
 - OpenRouter 缓存：`~/.cache/tokscale/pricing-openrouter.json`（增量式，仅缓存您使用过的模型）

packages/cli/src/cli.ts

@@ -1628,7 +1628,8 @@ async function handlePricingCommand(modelId: string, options: { json?: boolean;
         },
       }, null, 2));
     } else {
-      const sourceLabel = result.source.toLowerCase() === "litellm" ? pc.blue("LiteLLM") : pc.magenta("OpenRouter");
+      const sourceLower = result.source.toLowerCase();
+      const sourceLabel = sourceLower === "litellm" ? pc.blue("LiteLLM") : sourceLower === "cursor" ? pc.yellow("Cursor") : pc.magenta("OpenRouter");
       const inputCost = result.pricing.input_cost_per_token ?? 0;
       const outputCost = result.pricing.output_cost_per_token ?? 0;
       const cacheReadCost = result.pricing.cache_read_input_token_cost;

packages/core/index.js

@@ -536,17 +536,13 @@ if (!nativeBinding || process.env.NAPI_RS_FORCE_WASI) {
       wasiBindingError = err
     }
   }
-  if (!nativeBinding || process.env.NAPI_RS_FORCE_WASI) {
+  if (!nativeBinding) {
     try {
       wasiBinding = require('@tokscale/core-wasm32-wasi')
       nativeBinding = wasiBinding
     } catch (err) {
       if (process.env.NAPI_RS_FORCE_WASI) {
-        if (!wasiBindingError) {
-          wasiBindingError = err
-        } else {
-          wasiBindingError.cause = err
-        }
+        wasiBindingError.cause = err
         loadErrors.push(err)
       }
     }

packages/core/src/pricing/lookup.rs

@@ -69,11 +69,13 @@ struct CachedResult {
 pub struct PricingLookup {
     litellm: HashMap<String, ModelPricing>,
     openrouter: HashMap<String, ModelPricing>,
+    cursor: HashMap<String, ModelPricing>,
     litellm_keys: Vec<String>,
     openrouter_keys: Vec<String>,
     litellm_lower: HashMap<String, String>,
     openrouter_lower: HashMap<String, String>,
     openrouter_model_part: HashMap<String, String>,
+    cursor_lower: HashMap<String, String>,
     lookup_cache: RwLock<HashMap<String, Option<CachedResult>>>,
 }
 
@@ -87,6 +89,7 @@ impl PricingLookup {
     pub fn new(
         litellm: HashMap<String, ModelPricing>,
         openrouter: HashMap<String, ModelPricing>,
+        cursor: HashMap<String, ModelPricing>,
     ) -> Self {
         let mut litellm_keys: Vec<String> = litellm.keys().cloned().collect();
         litellm_keys.sort_by(|a, b| b.len().cmp(&a.len()));
@@ -111,14 +114,21 @@ impl PricingLookup {
             }
         }
 
+        let mut cursor_lower = HashMap::with_capacity(cursor.len());
+        for key in cursor.keys() {
+            cursor_lower.insert(key.to_lowercase(), key.clone());
+        }
+
         Self {
             litellm,
             openrouter,
+            cursor,
             litellm_keys,
             openrouter_keys,
             litellm_lower,
             openrouter_lower,
             openrouter_model_part,
+            cursor_lower,
             lookup_cache: RwLock::new(HashMap::with_capacity(64)),
         }
     }
@@ -230,6 +240,15 @@ impl PricingLookup {
             }
         }
 
+        if let Some(result) = self.exact_match_cursor(model_id) {
+            return Some(result);
+        }
+        if let Some(version_normalized) = normalize_version_separator(model_id) {
+            if let Some(result) = self.exact_match_cursor(&version_normalized) {
+                return Some(result);
+            }
+        }
+
         if !is_fuzzy_eligible(model_id) {
             return None;
         }
@@ -353,6 +372,28 @@ impl PricingLookup {
         None
     }
 
+    fn exact_match_cursor(&self, model_id: &str) -> Option<LookupResult> {
+        if let Some(key) = self.cursor_lower.get(model_id) {
+            return Some(LookupResult {
+                pricing: self.cursor.get(key).unwrap().clone(),
+                source: "Cursor".into(),
+                matched_key: key.clone(),
+            });
+        }
+        if let Some(model_part) = model_id.split('/').last() {
+            if model_part != model_id {
+                if let Some(key) = self.cursor_lower.get(model_part) {
+                    return Some(LookupResult {
+                        pricing: self.cursor.get(key).unwrap().clone(),
+                        source: "Cursor".into(),
+                        matched_key: key.clone(),
+                    });
+                }
+            }
+        }
+        None
+    }
+
     fn prefix_match_litellm(&self, model_id: &str) -> Option<LookupResult> {
         for prefix in PROVIDER_PREFIXES {
             let key = format!("{}{}", prefix, model_id);
@@ -451,19 +492,35 @@ impl PricingLookup {
             None => return 0.0,
         };
 
-        let p = &result.pricing;
-        let safe_price =
-            |opt: Option<f64>| opt.filter(|v| v.is_finite() && *v >= 0.0).unwrap_or(0.0);
-
-        let input_cost = input as f64 * safe_price(p.input_cost_per_token);
-        let output_cost = (output + reasoning) as f64 * safe_price(p.output_cost_per_token);
-        let cache_read_cost = cache_read as f64 * safe_price(p.cache_read_input_token_cost);
-        let cache_write_cost = cache_write as f64 * safe_price(p.cache_creation_input_token_cost);
-
-        input_cost + output_cost + cache_read_cost + cache_write_cost
+        compute_cost(
+            &result.pricing,
+            input,
+            output,
+            cache_read,
+            cache_write,
+            reasoning,
+        )
     }
 }
 
+pub fn compute_cost(
+    pricing: &ModelPricing,
+    input: i64,
+    output: i64,
+    cache_read: i64,
+    cache_write: i64,
+    reasoning: i64,
+) -> f64 {
+    let safe_price = |opt: Option<f64>| opt.filter(|v| v.is_finite() && *v >= 0.0).unwrap_or(0.0);
+
+    let input_cost = input as f64 * safe_price(pricing.input_cost_per_token);
+    let output_cost = (output + reasoning) as f64 * safe_price(pricing.output_cost_per_token);
+    let cache_read_cost = cache_read as f64 * safe_price(pricing.cache_read_input_token_cost);
+    let cache_write_cost = cache_write as f64 * safe_price(pricing.cache_creation_input_token_cost);
+
+    input_cost + output_cost + cache_read_cost + cache_write_cost
+}
+
 fn extract_model_family(model_id: &str) -> String {
     let lower = model_id.to_lowercase();
 
@@ -1069,7 +1126,7 @@ mod tests {
     }
 
     fn create_lookup() -> PricingLookup {
-        PricingLookup::new(mock_litellm(), mock_openrouter())
+        PricingLookup::new(mock_litellm(), mock_openrouter(), HashMap::new())
     }
 
     // =========================================================================
@@ -1468,7 +1525,7 @@ mod tests {
         );
         // Note: gpt-5-codex is NOT in the pricing data
 
-        let lookup = PricingLookup::new(litellm, HashMap::new());
+        let lookup = PricingLookup::new(litellm, HashMap::new(), HashMap::new());
 
         // Looking up gpt-5-codex should fall back to gpt-5
         let result = lookup.lookup("gpt-5-codex").unwrap();
@@ -1495,7 +1552,7 @@ mod tests {
             },
         );
 
-        let lookup = PricingLookup::new(litellm, HashMap::new());
+        let lookup = PricingLookup::new(litellm, HashMap::new(), HashMap::new());
 
         // gpt-5-codex-high should strip -high first, then fall back from gpt-5-codex to gpt-5
         let result = lookup.lookup("gpt-5-codex-high").unwrap();
@@ -1531,7 +1588,7 @@ mod tests {
             },
         );
 
-        let lookup = PricingLookup::new(litellm, HashMap::new());
+        let lookup = PricingLookup::new(litellm, HashMap::new(), HashMap::new());
 
         // Should use the exact match, not fall back
         let result = lookup.lookup("gpt-5-codex").unwrap();
@@ -1635,7 +1692,7 @@ mod tests {
             },
         );
 
-        let lookup = PricingLookup::new(litellm, HashMap::new());
+        let lookup = PricingLookup::new(litellm, HashMap::new(), HashMap::new());
         let result = lookup.lookup("grok-code").unwrap();
 
         // Must prefer xai (original provider) over azure_ai (reseller)