@sudobility/shapeshyft_engine 1.0.2 → 1.0.4

package/dist/config/providers.d.ts

@@ -6,16 +6,17 @@
  *
  * ## Supported Providers
  *
- * Catalog verified against provider documentation on 2026-08-18.
+ * Catalog verified against provider documentation on 2026-08-18; pricing
+ * re-verified against each provider's pricing page on 2026-09-14.
  *
  * - **openai**: GPT-5.6 (Sol/Terra/Luna), GPT-5.5, GPT-5.4, GPT-5, GPT-4.1, GPT-4o
  * - **anthropic**: Claude Fable 5, Opus 5, Sonnet 5, Haiku 4.5 (+ Opus 4.8/4.7/4.6, Sonnet 4.6)
  * - **gemini**: Gemini 3.7/3.6/3.5/3.1 and 2.5, Nano Banana image models, Veo 3.1 video
- * - **mistral**: Mistral Large 3, Medium 3.5, Small 4, Ministral 3, Codestral, OCR
+ * - **mistral**: Mistral Large 3, Medium 3.5, Small 4, Ministral 3, Codestral
  * - **cohere**: Command A+, Command A (reasoning/vision/translate), Command R
  * - **groq**: GPT-OSS, Qwen3.6, MiniMax M2.7, Compound, Whisper
  * - **xai**: Grok 4.6, 4.5, 4.3, Grok 4.20 variants, Grok Build
- * - **deepseek**: DeepSeek V4 Pro and V4 Flash
+ * - **deepseek**: DeepSeek V4.1 Flash and V4 Pro
  * - **perplexity**: Sonar models with live web search grounding
  * - **lm_studio**: Local LLM server (LM Studio or any OpenAI-compatible endpoint)
  *
@@ -30,6 +31,10 @@
  * `llama-3.1-8b-instant` on 2026-08-16, and both were listed here until then.
  *
  * Pricing is in **cents per 1M tokens** (image per image, audio/video per minute).
+ * `estimateUsageCost` (`lib/cost-estimation.ts`) prices a call from these rates
+ * plus the optional refinements on `ModelPricing`: cached and audio input,
+ * long-context tiers, peak hours, and per-request and per-search fees. A rate a
+ * provider does not publish is left out rather than guessed.
  *
  * ## LM Studio Model Identifiers
  *
@@ -96,13 +101,28 @@ export declare function getModelsForProvider(providerId: LlmProvider): string[];
  * @returns Model capabilities
  */
 export declare function getModelCapabilities(model: string): ModelCapabilities;
+export interface ModelPricingLookup {
+    /** The provider serving the model. `lm_studio` is always free. */
+    provider?: LlmProvider;
+    /** The model the endpoint asked for, tried when the reported one is unknown. */
+    configuredModel?: string | null;
+}
+/**
+ * Pricing for a model, or undefined when the catalog does not know it.
+ *
+ * Tries the model as reported, then without a snapshot suffix, then the same
+ * for `configuredModel`.
+ */
+export declare function findModelPricing(model: string, options?: ModelPricingLookup): ModelPricing | undefined;
 /**
  * Get pricing for a specific model (cents per 1M tokens).
- * Returns DEFAULT_MODEL_PRICING for unknown models.
- * @param model - The model identifier
+ * Returns DEFAULT_MODEL_PRICING for unknown models, and zero for any model on a
+ * self-hosted (`lm_studio`) server.
+ * @param model - The model identifier, as reported by the provider
+ * @param options - Provider and configured model, for a more reliable match
  * @returns Model pricing
  */
-export declare function getModelPricing(model: string): ModelPricing;
+export declare function getModelPricing(model: string, options?: ModelPricingLookup): ModelPricing;
 /**
  * Get the provider for a given model name.
  * Searches through PROVIDER_MODELS to find which provider owns this model.

package/dist/config/providers.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"providers.d.ts","sourceRoot":"","sources":["../../src/config/providers.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAwEG;AAEH,OAAO,KAAK,EACV,WAAW,EACX,iBAAiB,EACjB,YAAY,EACZ,cAAc,EACf,MAAM,mBAAmB,CAAC;AAM3B,eAAO,MAAM,SAAS,EAAE,cAAc,EAoFrC,CAAC;AAMF,eAAO,MAAM,eAAe,EAAE,MAAM,CAAC,WAAW,EAAE,MAAM,EAAE,CA2JzD,CAAC;AAMF,eAAO,MAAM,kBAAkB,EAAE,MAAM,CAAC,MAAM,EAAE,iBAAiB,CAs8BhE,CAAC;AAMF,eAAO,MAAM,aAAa,EAAE,MAAM,CAAC,MAAM,EAAE,YAAY,CA6MtD,CAAC;AAGF,eAAO,MAAM,qBAAqB,EAAE,YAA0C,CAAC;AAM/E;;;;GAIG;AACH,wBAAgB,eAAe,CAC7B,UAAU,EAAE,WAAW,GACtB,cAAc,GAAG,SAAS,CAE5B;AAED;;;;GAIG;AACH,wBAAgB,oBAAoB,CAAC,UAAU,EAAE,WAAW,GAAG,MAAM,EAAE,CAEtE;AAED;;;;;GAKG;AACH,wBAAgB,oBAAoB,CAAC,KAAK,EAAE,MAAM,GAAG,iBAAiB,CAErE;AAED;;;;;GAKG;AACH,wBAAgB,eAAe,CAAC,KAAK,EAAE,MAAM,GAAG,YAAY,CAK3D;AAED;;;GAGG;AACH,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,MAAM,GAAG,WAAW,CAQ9D"}
+{"version":3,"file":"providers.d.ts","sourceRoot":"","sources":["../../src/config/providers.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA6EG;AAEH,OAAO,KAAK,EACV,WAAW,EACX,iBAAiB,EACjB,YAAY,EACZ,cAAc,EACf,MAAM,mBAAmB,CAAC;AAM3B,eAAO,MAAM,SAAS,EAAE,cAAc,EAoFrC,CAAC;AAMF,eAAO,MAAM,eAAe,EAAE,MAAM,CAAC,WAAW,EAAE,MAAM,EAAE,CAgKzD,CAAC;AAMF,eAAO,MAAM,kBAAkB,EAAE,MAAM,CAAC,MAAM,EAAE,iBAAiB,CAq8BhE,CAAC;AAeF,eAAO,MAAM,aAAa,EAAE,MAAM,CAAC,MAAM,EAAE,YAAY,CAwVtD,CAAC;AAGF,eAAO,MAAM,qBAAqB,EAAE,YAA0C,CAAC;AAM/E;;;;GAIG;AACH,wBAAgB,eAAe,CAC7B,UAAU,EAAE,WAAW,GACtB,cAAc,GAAG,SAAS,CAE5B;AAED;;;;GAIG;AACH,wBAAgB,oBAAoB,CAAC,UAAU,EAAE,WAAW,GAAG,MAAM,EAAE,CAEtE;AAED;;;;;GAKG;AACH,wBAAgB,oBAAoB,CAAC,KAAK,EAAE,MAAM,GAAG,iBAAiB,CAErE;AAYD,MAAM,WAAW,kBAAkB;IACjC,kEAAkE;IAClE,QAAQ,CAAC,EAAE,WAAW,CAAC;IACvB,gFAAgF;IAChF,eAAe,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;CACjC;AASD;;;;;GAKG;AACH,wBAAgB,gBAAgB,CAC9B,KAAK,EAAE,MAAM,EACb,OAAO,GAAE,kBAAuB,GAC/B,YAAY,GAAG,SAAS,CAG1B;AAED;;;;;;;GAOG;AACH,wBAAgB,eAAe,CAC7B,KAAK,EAAE,MAAM,EACb,OAAO,GAAE,kBAAuB,GAC/B,YAAY,CAEd;AAED;;;GAGG;AACH,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,MAAM,GAAG,WAAW,CAQ9D"}

package/dist/config/providers.js

@@ -6,16 +6,17 @@
  *
  * ## Supported Providers
  *
- * Catalog verified against provider documentation on 2026-08-18.
+ * Catalog verified against provider documentation on 2026-08-18; pricing
+ * re-verified against each provider's pricing page on 2026-09-14.
  *
  * - **openai**: GPT-5.6 (Sol/Terra/Luna), GPT-5.5, GPT-5.4, GPT-5, GPT-4.1, GPT-4o
  * - **anthropic**: Claude Fable 5, Opus 5, Sonnet 5, Haiku 4.5 (+ Opus 4.8/4.7/4.6, Sonnet 4.6)
  * - **gemini**: Gemini 3.7/3.6/3.5/3.1 and 2.5, Nano Banana image models, Veo 3.1 video
- * - **mistral**: Mistral Large 3, Medium 3.5, Small 4, Ministral 3, Codestral, OCR
+ * - **mistral**: Mistral Large 3, Medium 3.5, Small 4, Ministral 3, Codestral
  * - **cohere**: Command A+, Command A (reasoning/vision/translate), Command R
  * - **groq**: GPT-OSS, Qwen3.6, MiniMax M2.7, Compound, Whisper
  * - **xai**: Grok 4.6, 4.5, 4.3, Grok 4.20 variants, Grok Build
- * - **deepseek**: DeepSeek V4 Pro and V4 Flash
+ * - **deepseek**: DeepSeek V4.1 Flash and V4 Pro
  * - **perplexity**: Sonar models with live web search grounding
  * - **lm_studio**: Local LLM server (LM Studio or any OpenAI-compatible endpoint)
  *
@@ -30,6 +31,10 @@
  * `llama-3.1-8b-instant` on 2026-08-16, and both were listed here until then.
  *
  * Pricing is in **cents per 1M tokens** (image per image, audio/video per minute).
+ * `estimateUsageCost` (`lib/cost-estimation.ts`) prices a call from these rates
+ * plus the optional refinements on `ModelPricing`: cached and audio input,
+ * long-context tiers, peak hours, and per-request and per-search fees. A rate a
+ * provider does not publish is left out rather than guessed.
  *
  * ## LM Studio Model Identifiers
  *
@@ -134,9 +139,9 @@ export const PROVIDERS = [
     {
         id: "deepseek",
         name: "DeepSeek",
-        description: "DeepSeek V4 Pro and V4 Flash models",
+        description: "DeepSeek V4.1 Flash and V4 Pro models",
         allowsCustomModel: false,
-        defaultModel: "deepseek-v4-flash",
+        defaultModel: "deepseek-flash",
         requiresEndpointUrl: false,
     },
     {
@@ -230,7 +235,7 @@ export const PROVIDER_MODELS = {
         "ministral-3b-2512",
         "codestral-latest",
         "codestral-2508",
-        "mistral-ocr-latest",
+        // No OCR model: Mistral serves OCR only at /v1/ocr, never chat completions.
     ],
     cohere: [
         "command-a-plus-05-2026",
@@ -262,7 +267,12 @@ export const PROVIDER_MODELS = {
         "grok-4.20-multi-agent-0309",
         "grok-build-0.1",
     ],
-    deepseek: ["deepseek-v4-pro", "deepseek-v4-flash"],
+    deepseek: [
+        "deepseek-flash", // V4.1 Flash (2026-09-10)
+        "deepseek-v4-pro",
+        // Retired V4 Flash ID; DeepSeek currently serves it with V4.1 Flash
+        "deepseek-v4-flash",
+    ],
     perplexity: [
         "sonar",
         "sonar-pro",
@@ -844,15 +854,6 @@ export const MODEL_CAPABILITIES = {
         audioOutput: false,
         videoOutput: false,
     },
-    "mistral-ocr-latest": {
-        visionInput: true,
-        audioInput: false,
-        videoInput: false,
-        imageOutput: false,
-        audioOutput: false,
-        videoOutput: false,
-        mediaFormats: { imageFormats: ["url", "base64"] },
-    },
     // Cohere — https://docs.cohere.com/docs/models
     "command-a-plus-05-2026": {
         visionInput: true,
@@ -1065,6 +1066,14 @@ export const MODEL_CAPABILITIES = {
         mediaFormats: { imageFormats: ["url", "base64"] },
     },
     // DeepSeek — https://api-docs.deepseek.com/quick_start/pricing (standard, cache miss)
+    "deepseek-flash": {
+        visionInput: false,
+        audioInput: false,
+        videoInput: false,
+        imageOutput: false,
+        audioOutput: false,
+        videoOutput: false,
+    },
     "deepseek-v4-pro": {
         visionInput: false,
         audioInput: false,
@@ -1279,24 +1288,87 @@ export const MODEL_CAPABILITIES = {
 // =============================================================================
 // Model Pricing (cents per 1M tokens)
 // =============================================================================
+/** DeepSeek bills 2x during these weekday UTC hours. */
+const DEEPSEEK_PEAK = {
+    multiplier: 2,
+    windowsUtc: [
+        { days: [1, 2, 3, 4, 5], startHour: 1, endHour: 4 },
+        { days: [1, 2, 3, 4, 5], startHour: 6, endHour: 10 },
+    ],
+};
 export const MODEL_PRICING = {
-    // OpenAI — https://developers.openai.com/api/docs/models
-    "gpt-5.6-sol": { input: 500, output: 3000 },
-    "gpt-5.6-terra": { input: 200, output: 1200 },
-    "gpt-5.6-luna": { input: 20, output: 120 },
-    "gpt-5.5": { input: 500, output: 3000 },
-    "gpt-5.4": { input: 250, output: 1500 },
-    "gpt-5.4-mini": { input: 75, output: 450 },
-    "gpt-5.4-nano": { input: 20, output: 125 },
-    "gpt-5": { input: 125, output: 1000 },
-    "gpt-5-mini": { input: 25, output: 200 },
-    "gpt-5-nano": { input: 5, output: 40 },
-    "gpt-4.1": { input: 200, output: 800 },
-    "gpt-4.1-mini": { input: 40, output: 160 },
-    "gpt-4.1-nano": { input: 10, output: 40 },
-    "gpt-4o": { input: 250, output: 1000 },
-    "gpt-4o-mini": { input: 15, output: 60 },
-    // Anthropic — https://platform.claude.com/docs/en/about-claude/models/overview
+    // OpenAI — https://developers.openai.com/api/docs/pricing
+    // Cached input is 0.1x. GPT-5.6 also bills cache writes at 1.25x input.
+    // web_search_preview: $10 / 1K calls on reasoning models (GPT-5 family),
+    // $25 / 1K on GPT-4.1 and GPT-4o.
+    // Sol is on promotion "at least through November 21, 2026"; list is 500 / 3000.
+    "gpt-5.6-sol": {
+        input: 400,
+        output: 2000,
+        cachedInput: 40,
+        cacheWriteInput: 500,
+        searchCall: 1,
+        longContext: {
+            minPromptTokens: 272_001,
+            input: 800,
+            output: 3000,
+            cachedInput: 80,
+        },
+    },
+    "gpt-5.6-terra": {
+        input: 200,
+        output: 1200,
+        cachedInput: 20,
+        cacheWriteInput: 250,
+        searchCall: 1,
+        longContext: {
+            minPromptTokens: 272_001,
+            input: 400,
+            output: 1800,
+            cachedInput: 40,
+        },
+    },
+    "gpt-5.6-luna": {
+        input: 20,
+        output: 120,
+        cachedInput: 2,
+        cacheWriteInput: 25,
+        searchCall: 1,
+        longContext: {
+            minPromptTokens: 272_001,
+            input: 40,
+            output: 180,
+            cachedInput: 4,
+        },
+    },
+    // 5.5 / 5.4 long context: "2x input and 1.5x output for the full session"
+    "gpt-5.5": {
+        input: 500,
+        output: 3000,
+        cachedInput: 50,
+        searchCall: 1,
+        longContext: { minPromptTokens: 272_001, input: 1000, output: 4500 },
+    },
+    "gpt-5.4": {
+        input: 250,
+        output: 1500,
+        cachedInput: 25,
+        searchCall: 1,
+        longContext: { minPromptTokens: 272_001, input: 500, output: 2250 },
+    },
+    "gpt-5.4-mini": { input: 75, output: 450, cachedInput: 7.5, searchCall: 1 },
+    "gpt-5.4-nano": { input: 20, output: 125, cachedInput: 2, searchCall: 1 },
+    "gpt-5": { input: 125, output: 1000, cachedInput: 12.5, searchCall: 1 },
+    "gpt-5-mini": { input: 25, output: 200, cachedInput: 2.5, searchCall: 1 },
+    "gpt-5-nano": { input: 5, output: 40, cachedInput: 0.5, searchCall: 1 },
+    "gpt-4.1": { input: 200, output: 800, cachedInput: 50, searchCall: 2.5 },
+    "gpt-4.1-mini": { input: 40, output: 160, cachedInput: 10, searchCall: 2.5 },
+    "gpt-4.1-nano": { input: 10, output: 40, cachedInput: 2.5, searchCall: 2.5 }, // Shuts down 2026-10-23
+    "gpt-4o": { input: 250, output: 1000, cachedInput: 125, searchCall: 2.5 },
+    "gpt-4o-mini": { input: 15, output: 60, cachedInput: 7.5, searchCall: 2.5 },
+    // Anthropic — https://platform.claude.com/docs/en/about-claude/pricing
+    // 4.6 and later: full 1M context at standard rates, no long-context surcharge.
+    // No cachedInput: the adapter sends no cache_control, so nothing is cached.
     "claude-fable-5": { input: 1000, output: 5000 },
     "claude-opus-5": { input: 500, output: 2500 },
     "claude-sonnet-5": { input: 200, output: 1000 },
@@ -1306,111 +1378,89 @@ export const MODEL_PRICING = {
     "claude-opus-4-6": { input: 500, output: 2500 },
     "claude-sonnet-4-6": { input: 300, output: 1500 },
     "claude-opus-4-5-20251101": { input: 500, output: 2500 },
-    "claude-sonnet-4-5-20250929": { input: 300, output: 1500 },
-    // Google Gemini — https://ai.google.dev/gemini-api/docs/models
+    "claude-sonnet-4-5-20250929": { input: 300, output: 1500 }, // Retires 2026-09-29
+    // Google Gemini — https://ai.google.dev/gemini-api/docs/pricing
+    // Images, video and (unless audioTokenInput says otherwise) audio are billed
+    // as input tokens at `input`; output includes thinking tokens.
     "gemini-3.7-flash": {
         input: 75,
         output: 375,
-        imageInput: 39,
-        audioInput: 10,
-        videoInput: 39,
+        cachedInput: 7.5,
+        priceChanges: [
+            { from: "2027-01-01", input: 150, output: 750, cachedInput: 15 },
+        ],
     },
     "gemini-3.6-flash": {
         input: 75,
         output: 375,
-        imageInput: 39,
-        audioInput: 10,
-        videoInput: 39,
-    },
-    "gemini-3.5-flash": {
-        input: 150,
-        output: 900,
-        imageInput: 39,
-        audioInput: 10,
-        videoInput: 39,
-    },
-    "gemini-3.5-flash-lite": {
-        input: 30,
-        output: 250,
-        imageInput: 30,
-        audioInput: 50,
-        videoInput: 30,
+        cachedInput: 7.5,
+        priceChanges: [
+            { from: "2027-01-01", input: 150, output: 750, cachedInput: 15 },
+        ],
     },
+    "gemini-3.5-flash": { input: 150, output: 900, cachedInput: 15 },
+    "gemini-3.5-flash-lite": { input: 30, output: 250, cachedInput: 3 },
     "gemini-3.1-flash-lite": {
         input: 25,
         output: 150,
-        imageInput: 25,
-        audioInput: 50,
-        videoInput: 25,
-    },
+        cachedInput: 2.5,
+        audioTokenInput: 50,
+    }, // Shuts down 2027-05-07
     "gemini-3.1-pro-preview": {
         input: 200,
         output: 1200,
-        imageInput: 32.9,
-        audioInput: 10,
-        videoInput: 32.9,
+        cachedInput: 20,
+        longContext: {
+            minPromptTokens: 200_001,
+            input: 400,
+            output: 1800,
+            cachedInput: 40,
+        },
     },
     "gemini-3-flash-preview": {
-        input: 75,
-        output: 375,
-        imageInput: 39,
-        audioInput: 10,
-        videoInput: 39,
-    },
-    "gemini-3.1-flash-image": {
-        input: 75,
-        output: 375,
-        imageInput: 39,
-        imageOutput: 3.9,
-    }, // Nano Banana 2
-    "gemini-3.1-flash-lite-image": {
-        input: 25,
-        output: 150,
-        imageInput: 25,
-        imageOutput: 2.4,
-    }, // Nano Banana 2 Lite
-    "gemini-3-pro-image": {
-        input: 125,
-        output: 1000,
-        imageInput: 32.9,
-        imageOutput: 12,
-    }, // Nano Banana Pro
+        input: 50,
+        output: 300,
+        cachedInput: 5,
+        audioTokenInput: 100,
+    },
+    "gemini-3.1-flash-image": { input: 50, output: 300, imageOutput: 6.7 }, // Nano Banana 2; per image at 1K (4.5 at 0.5K, 10.1 at 2K, 15.1 at 4K)
+    "gemini-3.1-flash-lite-image": { input: 25, output: 150, imageOutput: 3.36 }, // Nano Banana 2 Lite; per image at 1K
+    "gemini-3-pro-image": { input: 200, output: 1200, imageOutput: 13.4 }, // Nano Banana Pro; per image at 1K/2K (24 at 4K)
     "gemini-2.5-pro": {
         input: 125,
         output: 1000,
-        imageInput: 32.9,
-        audioInput: 10,
-        videoInput: 32.9,
+        cachedInput: 12.5,
+        longContext: {
+            minPromptTokens: 200_001,
+            input: 250,
+            output: 1500,
+            cachedInput: 25,
+        },
     },
     "gemini-2.5-flash": {
         input: 30,
         output: 250,
-        imageInput: 30,
-        audioInput: 100,
-        videoInput: 30,
+        cachedInput: 3,
+        audioTokenInput: 100,
     },
     "gemini-2.5-flash-lite": {
         input: 10,
         output: 40,
-        imageInput: 10,
-        audioInput: 30,
-        videoInput: 10,
-    },
-    "gemini-2.5-flash-image": {
-        input: 30,
-        output: 250,
-        imageInput: 30,
-        imageOutput: 3.9,
+        cachedInput: 1,
+        audioTokenInput: 30,
     },
+    "gemini-2.5-flash-image": { input: 30, output: 250, imageOutput: 3.9 }, // Shuts down 2026-10-02
     "gemini-2.5-flash-native-audio-preview-12-2025": {
-        input: 30,
-        output: 250,
-        audioInput: 100,
-        audioOutput: 200,
-    },
-    "veo-3.1-generate-preview": { input: 0, output: 0, videoOutput: 4000 }, // priced per second of video
-    "veo-3.1-lite-generate-preview": { input: 0, output: 0, videoOutput: 1500 }, // priced per second of video
-    // Mistral — https://docs.mistral.ai/getting-started/models/models_overview/
+        input: 50,
+        output: 200,
+        audioTokenInput: 300,
+    }, // Audio output is $12 / 1M tokens
+    // Veo is billed per second of video; stored per minute as ModelPricing expects.
+    "veo-3.1-generate-preview": { input: 0, output: 0, videoOutput: 2400 }, // $0.40/s at 720p/1080p, $0.60/s at 4K
+    "veo-3.1-lite-generate-preview": { input: 0, output: 0, videoOutput: 300 }, // $0.05/s at 720p, $0.08/s at 1080p
+    // Mistral — https://mistral.ai/pricing/api
+    // Cached input is "up to 90%" cheaper; the exact rate is unpublished, so
+    // cached tokens are priced at the full input rate.
     "mistral-large-latest": { input: 50, output: 150 },
     "mistral-large-2512": { input: 50, output: 150 },
     "mistral-medium-latest": { input: 150, output: 750 },
@@ -1421,10 +1471,12 @@ export const MODEL_PRICING = {
     "ministral-3b-2512": { input: 10, output: 10 },
     "codestral-latest": { input: 30, output: 90 },
     "codestral-2508": { input: 30, output: 90 },
-    "mistral-ocr-latest": { input: 100, output: 300 }, // document OCR
     // Cohere — https://docs.cohere.com/docs/models
-    "command-a-plus-05-2026": { input: 250, output: 1000 },
     "command-a-03-2025": { input: 250, output: 1000 },
+    // No published token price for A+, Reasoning, Vision, or Translate: free up
+    // to rate limits, production use via Model Vault or sales. Command A's rate
+    // is assumed.
+    "command-a-plus-05-2026": { input: 250, output: 1000 },
     "command-a-reasoning-08-2025": { input: 250, output: 1000 },
     "command-a-vision-07-2025": { input: 250, output: 1000 },
     "command-a-translate-08-2025": { input: 250, output: 1000 },
@@ -1432,31 +1484,135 @@ export const MODEL_PRICING = {
     "command-r-plus-08-2024": { input: 250, output: 1000 },
     "command-r-08-2024": { input: 15, output: 60 },
     // Groq — https://console.groq.com/docs/models
-    "openai/gpt-oss-120b": { input: 15, output: 75 },
-    "openai/gpt-oss-20b": { input: 10, output: 50 },
-    "qwen/qwen3.6-27b": { input: 29, output: 59 },
+    "openai/gpt-oss-120b": { input: 15, output: 60, cachedInput: 7.5 },
+    "openai/gpt-oss-20b": { input: 7.5, output: 30, cachedInput: 3.75 },
+    "qwen/qwen3.6-27b": { input: 60, output: 300 },
+    // Enterprise only, "Contact Sales"; no public price. Last public rate kept.
     "minimaxai/minimax-m2.7": { input: 29, output: 115 },
-    "groq/compound": { input: 15, output: 75 },
-    "groq/compound-mini": { input: 10, output: 50 },
-    "whisper-large-v3": { input: 11, output: 0 }, // per hour of audio
-    "whisper-large-v3-turbo": { input: 4, output: 0 }, // per hour of audio
+    // Compound is priced from its per-model and per-tool breakdown
+    // (`compoundCostCents`); these rates apply only when a response lacks one.
+    "groq/compound": { input: 15, output: 60 },
+    "groq/compound-mini": { input: 15, output: 60 },
+    // Whisper, per minute of audio ($0.111 and $0.04 per hour), 10 s minimum
+    "whisper-large-v3": { input: 0, output: 0, audioInput: 0.185 },
+    "whisper-large-v3-turbo": { input: 0, output: 0, audioInput: 0.04 / 0.6 },
     // xAI — https://docs.x.ai/docs/models
-    "grok-4.6": { input: 200, output: 600 },
-    "grok-4.5": { input: 200, output: 600 },
-    "grok-4.3": { input: 125, output: 250 },
-    "grok-4.20-0309-reasoning": { input: 125, output: 250 },
-    "grok-4.20-0309-non-reasoning": { input: 125, output: 250 },
-    "grok-4.20-multi-agent-0309": { input: 125, output: 250 },
-    "grok-build-0.1": { input: 100, output: 200 },
-    // DeepSeek — https://api-docs.deepseek.com/quick_start/pricing (standard, cache miss)
-    "deepseek-v4-pro": { input: 66, output: 198 },
-    "deepseek-v4-flash": { input: 22, output: 66 },
-    // Perplexity — https://docs.perplexity.ai (per-request search fees are billed separately)
-    sonar: { input: 100, output: 100 },
-    "sonar-pro": { input: 300, output: 1500 },
-    "sonar-pro-search": { input: 300, output: 1500 },
-    "sonar-reasoning-pro": { input: 200, output: 800 },
-    "sonar-deep-research": { input: 200, output: 800 },
+    // Responses carry the billed amount (`cost_in_usd_ticks`), which is used when
+    // present. Prompts of 200K tokens or more bill every token at 2x.
+    "grok-4.6": {
+        input: 200,
+        output: 600,
+        cachedInput: 50,
+        longContext: {
+            minPromptTokens: 200_000,
+            input: 400,
+            output: 1200,
+            cachedInput: 100,
+        },
+    },
+    "grok-4.5": {
+        input: 200,
+        output: 600,
+        cachedInput: 30,
+        longContext: {
+            minPromptTokens: 200_000,
+            input: 400,
+            output: 1200,
+            cachedInput: 60,
+        },
+    },
+    "grok-4.3": {
+        input: 125,
+        output: 250,
+        cachedInput: 20,
+        longContext: {
+            minPromptTokens: 200_000,
+            input: 250,
+            output: 500,
+            cachedInput: 40,
+        },
+    },
+    "grok-4.20-0309-reasoning": {
+        input: 125,
+        output: 250,
+        cachedInput: 20,
+        longContext: {
+            minPromptTokens: 200_000,
+            input: 250,
+            output: 500,
+            cachedInput: 40,
+        },
+    },
+    "grok-4.20-0309-non-reasoning": {
+        input: 125,
+        output: 250,
+        cachedInput: 20,
+        longContext: {
+            minPromptTokens: 200_000,
+            input: 250,
+            output: 500,
+            cachedInput: 40,
+        },
+    },
+    "grok-4.20-multi-agent-0309": {
+        input: 125,
+        output: 250,
+        cachedInput: 20,
+        longContext: {
+            minPromptTokens: 200_000,
+            input: 250,
+            output: 500,
+            cachedInput: 40,
+        },
+    },
+    "grok-build-0.1": {
+        input: 100,
+        output: 200,
+        cachedInput: 20,
+        longContext: {
+            minPromptTokens: 200_000,
+            input: 200,
+            output: 400,
+            cachedInput: 40,
+        },
+    },
+    // DeepSeek — https://api-docs.deepseek.com/quick_start/pricing
+    // Off-peak rates; peak hours (01:00-04:00 and 06:00-10:00 UTC, Mon-Fri) bill
+    // at 2x. V4.1 Flash repriced 2026-09-10; `deepseek-v4-flash` now runs on it.
+    "deepseek-flash": {
+        input: 15,
+        output: 60,
+        cachedInput: 0.3,
+        peak: DEEPSEEK_PEAK,
+    },
+    "deepseek-v4-pro": {
+        input: 66,
+        output: 198,
+        cachedInput: 2.2,
+        peak: DEEPSEEK_PEAK,
+    },
+    "deepseek-v4-flash": {
+        input: 15,
+        output: 60,
+        cachedInput: 0.3,
+        peak: DEEPSEEK_PEAK,
+    },
+    // Perplexity — https://docs.perplexity.ai/docs/getting-started/pricing
+    // Responses carry the billed amount (`usage.cost.total_cost`), which is used
+    // when present. The request fee below is the default (low) search context.
+    // Sonar is supported until 2026-09-27, then replaced by the Agent API.
+    sonar: { input: 100, output: 100, requestFee: 0.5 },
+    "sonar-pro": { input: 300, output: 1500, requestFee: 0.6 },
+    // Not a real model ID: Pro Search is sonar-pro with search_type "pro"
+    "sonar-pro-search": { input: 300, output: 1500, requestFee: 1.4 },
+    "sonar-reasoning-pro": { input: 200, output: 800, requestFee: 0.6 },
+    "sonar-deep-research": {
+        input: 200,
+        output: 800,
+        citationTokens: 200,
+        reasoningTokens: 300,
+        searchCall: 0.5,
+    },
     // LM Studio / custom OpenAI-compatible servers — self-hosted, no per-token cost
     "qwen3-8b": { input: 0, output: 0 },
     "qwen3-14b": { input: 0, output: 0 },
@@ -1506,18 +1662,40 @@ export function getModelsForProvider(providerId) {
 export function getModelCapabilities(model) {
     return MODEL_CAPABILITIES[model] ?? {};
 }
+/** Self-hosted models cost nothing per token, whatever they are called. */
+const FREE_PRICING = { input: 0, output: 0 };
+/**
+ * Snapshot suffixes providers append to a model they report back:
+ * `gpt-4.1-mini-2025-04-14` (OpenAI) and `claude-haiku-4-5-20251001`
+ * (Anthropic, for an alias).
+ */
+const SNAPSHOT_SUFFIX = /-(\d{4}-\d{2}-\d{2}|\d{8})$/;
+function catalogPricing(model) {
+    if (!model)
+        return undefined;
+    return (MODEL_PRICING[model] ?? MODEL_PRICING[model.replace(SNAPSHOT_SUFFIX, "")]);
+}
+/**
+ * Pricing for a model, or undefined when the catalog does not know it.
+ *
+ * Tries the model as reported, then without a snapshot suffix, then the same
+ * for `configuredModel`.
+ */
+export function findModelPricing(model, options = {}) {
+    if (options.provider === "lm_studio")
+        return FREE_PRICING;
+    return catalogPricing(model) ?? catalogPricing(options.configuredModel);
+}
 /**
  * Get pricing for a specific model (cents per 1M tokens).
- * Returns DEFAULT_MODEL_PRICING for unknown models.
- * @param model - The model identifier
+ * Returns DEFAULT_MODEL_PRICING for unknown models, and zero for any model on a
+ * self-hosted (`lm_studio`) server.
+ * @param model - The model identifier, as reported by the provider
+ * @param options - Provider and configured model, for a more reliable match
  * @returns Model pricing
  */
-export function getModelPricing(model) {
-    // Try exact match first, then strip date suffix (e.g., "gpt-4.1-mini-2025-04-14" -> "gpt-4.1-mini")
-    if (MODEL_PRICING[model])
-        return MODEL_PRICING[model];
-    const baseModel = model.replace(/-\d{4}-\d{2}-\d{2}$/, "");
-    return MODEL_PRICING[baseModel] ?? DEFAULT_MODEL_PRICING;
+export function getModelPricing(model, options = {}) {
+    return findModelPricing(model, options) ?? DEFAULT_MODEL_PRICING;
 }
 /**
  * Get the provider for a given model name.