@tyvm/knowhow 0.0.104 → 0.0.106

package/src/clients/openai.ts

@@ -34,12 +34,14 @@ import {
   EmbeddingModels,
   Models,
   OpenAiReasoningModels,
+  OpenAiChatModels,
   OpenAiResponsesOnlyModels,
   OpenAiImageModels,
   OpenAiVideoModels,
   OpenAiTTSModels,
   OpenAiTranscriptionModels,
-  OpenAiEmbeddingModels,
+  OpenAiEmbeddingModelsList,
+  OpenAiRealtimeModels,
 } from "../types";
 import { ModelModality } from "./types";
 
@@ -64,7 +66,11 @@ export class GenericOpenAiClient implements GenericClient {
   reasoningEffort(
     messages: CompletionOptions["messages"]
   ): "low" | "medium" | "high" {
-    const effortMap = {
+    return this.detectReasoningEffort(messages);
+  }
+
+  detectReasoningEffort(messages: CompletionOptions["messages"]): "low" | "medium" | "high" {
+    const effortMap: Record<string, "low" | "medium" | "high"> = {
       ultrathink: "high",
       "think hard": "high",
       "reason hard": "high",
@@ -96,6 +102,30 @@ export class GenericOpenAiClient implements GenericClient {
     return "medium"; // Default to medium if no specific effort is mentioned
   }
 
+  resolveReasoningEffort(options: CompletionOptions): "low" | "medium" | "high" {
+    return options.reasoning_effort ?? this.detectReasoningEffort(options.messages);
+  }
+
+  /**
+   * Resolves the reasoning effort for a specific model, clamping to the model's
+   * supported levels if `reasoningLevels` is set in its pricing entry.
+   * If the requested level is not supported, picks the lowest supported level.
+   */
+  resolveReasoningEffortForModel(options: CompletionOptions): string {
+    const requested = options.reasoning_effort ?? this.detectReasoningEffort(options.messages);
+    const pricing = OpenAiTextPricing[options.model];
+    const supportedLevels = pricing?.reasoningLevels;
+    if (!supportedLevels || supportedLevels.length === 0) {
+      return requested;
+    }
+    // If the requested level is supported, use it
+    if (supportedLevels.includes(requested)) {
+      return requested;
+    }
+    // Otherwise use the first (lowest) supported level
+    return supportedLevels[0];
+  }
+
   async createChatCompletion(
     options: CompletionOptions
   ): Promise<CompletionResponse> {
@@ -122,8 +152,8 @@ export class GenericOpenAiClient implements GenericClient {
       max_tokens: options.max_tokens,
       ...(OpenAiReasoningModels.includes(options.model) && {
         max_tokens: undefined,
-        max_completion_tokens: Math.max(options.max_tokens, 100),
-        reasoning_effort: this.reasoningEffort(options.messages),
+        max_completion_tokens: Math.max(options.max_tokens ?? 0, 16_000),
+        reasoning_effort: this.resolveReasoningEffort(options),
       }),
 
       ...(options.tools && {
@@ -254,7 +284,7 @@ export class GenericOpenAiClient implements GenericClient {
       // Don't limit max_output_tokens for Responses API - codex truncates tool call arguments when limited
       ...(OpenAiReasoningModels.includes(options.model) && {
         max_output_tokens: Math.max(options.max_tokens || 0, 16000),
-        reasoning: { effort: this.reasoningEffort(options.messages) },
+        reasoning: { effort: this.resolveReasoningEffortForModel(options) },
       }),
       ...(tools?.length && {
         tools,
@@ -349,14 +379,14 @@ export class GenericOpenAiClient implements GenericClient {
       ("prompt_tokens_details" in usage &&
         usage.prompt_tokens_details?.cached_tokens) ||
       0;
-    const cachedInputCost = (cachedInputTokens * pricing.cached_input) / 1e6;
+    const cachedInputCost = (cachedInputTokens * (pricing.cached_input ?? 0)) / 1e6;
 
     const inputTokens = usage.prompt_tokens;
-    const inputCost = ((inputTokens - cachedInputCost) * pricing.input) / 1e6;
+    const inputCost = ((inputTokens - cachedInputTokens) * (pricing.input ?? 0)) / 1e6;
 
     const outputTokens =
       ("completion_tokens" in usage && usage?.completion_tokens) || 0;
-    const outputCost = (outputTokens * pricing.output) / 1e6;
+    const outputCost = (outputTokens * (pricing.output ?? 0)) / 1e6;
 
     const total = cachedInputCost + inputCost + outputCost;
     return total;
@@ -365,8 +395,8 @@ export class GenericOpenAiClient implements GenericClient {
   async getModels(modality?: ModelModality): Promise<{ id: string }[]> {
     if (modality) {
       const map: Partial<Record<ModelModality, string[]>> = {
-        completion: Object.values(Models.openai),
-        embedding: OpenAiEmbeddingModels,
+        completion: [...new Set([...OpenAiChatModels, ...OpenAiResponsesOnlyModels])],
+        embedding: OpenAiEmbeddingModelsList,
         image: OpenAiImageModels,
         audio: [...OpenAiTTSModels, ...OpenAiTranscriptionModels],
         transcription: OpenAiTranscriptionModels,
@@ -406,7 +436,7 @@ export class GenericOpenAiClient implements GenericClient {
     }
 
     const response = await this.client.audio.transcriptions.create({
-      file: file,
+      file,
       model: options.model || "whisper-1",
       language: options.language,
       prompt: options.prompt,

package/src/clients/openrouter.ts

@@ -0,0 +1,17 @@
+import { HttpClient } from "./http";
+import { OpenRouterTextPricing } from "./pricing/openrouter";
+
+/**
+ * OpenRouter client — OpenAI-compatible API aggregator
+ * https://openrouter.ai/docs
+ * 39+ free models; append `:free` suffix to a model id for the free variant.
+ * One API key gives access to models from many providers.
+ * Set env var OPENROUTER_API_KEY to enable.
+ */
+export class GenericOpenRouterClient extends HttpClient {
+  constructor(apiKey = process.env.OPENROUTER_API_KEY) {
+    super("https://openrouter.ai/api");
+    if (apiKey) this.setJwt(apiKey);
+    this.setPrices(OpenRouterTextPricing);
+  }
+}

package/src/clients/pricing/anthropic.ts

@@ -1,80 +1,107 @@
-import { Models } from "../../types";
+/**
+ * Anthropic model IDs, pricing, and catalog.
+ * Single source of truth for all Anthropic/Claude models.
+ * Deprecation dates from: https://docs.anthropic.com/en/docs/about-claude/model-deprecations
+ */
+import { completions, ModelCatalogEntry, ModelPricing } from "./types";
 
-export const AnthropicTextPricing = {
-  [Models.anthropic.Opus4_6]: {
-    input: 5.0,
-    cache_write: 6.25,
-    cache_hit: 0.5,
-    output: 25.0,
-  },
-  [Models.anthropic.Sonnet4_6]: {
-    input: 3.0,
-    cache_write: 3.75,
-    cache_hit: 0.3,
-    output: 15.0,
-  },
-  [Models.anthropic.Opus4_5]: {
-    input: 5.0,
-    cache_write: 6.25,
-    cache_hit: 0.5,
-    output: 25.0,
-  },
-  [Models.anthropic.Opus4_1]: {
-    input: 15.0,
-    cache_write: 18.75,
-    cache_hit: 1.5,
-    output: 75.0,
-  },
-  [Models.anthropic.Opus4]: {
-    input: 15.0,
-    cache_write: 18.75,
-    cache_hit: 1.5,
-    output: 75.0,
-  },
-  [Models.anthropic.Sonnet4]: {
-    input: 3.0,
-    input_gt_200k: 6.0,
-    cache_write: 3.75,
-    cache_hit: 0.3,
-    output: 15.0,
-    output_gt_200k: 22.5,
-  },
-  [Models.anthropic.Sonnet4_5]: {
-    input: 3.0,
-    input_gt_200k: 6.0,
-    cache_write: 3.75,
-    cache_hit: 0.3,
-    output: 15.0,
-    output_gt_200k: 22.5,
-  },
-  [Models.anthropic.Haiku4_5]: {
-    input: 1,
-    cache_write: 1.25,
-    cache_hit: 0.1,
-    output: 5,
-  },
-  [Models.anthropic.Sonnet3_7]: {
-    input: 3.0,
-    cache_write: 3.75,
-    cache_hit: 0.3,
-    output: 15.0,
-  },
-  [Models.anthropic.Sonnet3_5]: {
-    input: 3.0,
-    cache_write: 3.75,
-    cache_hit: 0.3,
-    output: 15.0,
-  },
-  [Models.anthropic.Opus3]: {
-    input: 15.0,
-    cache_write: 18.75,
-    cache_hit: 1.5,
-    output: 75.0,
-  },
-  [Models.anthropic.Haiku3]: {
-    input: 0.25,
-    cache_write: 0.3125,
-    cache_hit: 0.025,
-    output: 1.25,
-  },
+// ─── Model IDs ────────────────────────────────────────────────────────────────
+
+export const AnthropicModels = {
+  // Active models
+  Opus4_7:   "claude-opus-4-7",
+  Opus4_6:   "claude-opus-4-6",
+  Opus4_6Fast: "claude-opus-4-6-fast",
+  Sonnet4_6: "claude-sonnet-4-6",
+  Opus4_5:   "claude-opus-4-5-20251101",
+  Opus4_1:   "claude-opus-4-1-20250805",
+  Sonnet4_5: "claude-sonnet-4-5-20250929",
+  Haiku4_5:  "claude-haiku-4-5-20251001",
+  // Deprecated models (per Anthropic deprecation page)
+  Opus4:     "claude-opus-4-20250514",   // deprecated 2026-04-14, retirement 2026-06-15
+  Sonnet4:   "claude-sonnet-4-20250514", // deprecated 2026-04-14, retirement 2026-06-15
+  Sonnet3_7: "claude-3-7-sonnet-20250219", // retired 2026-02-19
+  Sonnet3_5: "claude-3-5-sonnet-20241022", // retired 2025-10-28
+  Sonnet3_5_20240620: "claude-3-5-sonnet-20240620", // earlier version, retired
+  Haiku3_5_Latest: "claude-3-5-haiku-latest", // alias → claude-3-5-haiku-20241022, retired
+  Sonnet3:   "claude-3-sonnet-20240229", // retired
+  // models.dev alias IDs (versioned as -0 suffix instead of date)
+  Opus4_0:   "claude-opus-4-0",    // alias for claude-opus-4-20250514, deprecated
+  Sonnet4_0: "claude-sonnet-4-0",  // alias for claude-sonnet-4-20250514, deprecated
+  Haiku3_5:  "claude-3-5-haiku-20241022",  // retired 2026-02-19
+  Opus3:     "claude-3-opus-20240229",   // retired 2026-01-05
+  Haiku3:    "claude-3-haiku-20240307",  // retired 2026-04-20
+} as const;
+
+// ─── Active (non-deprecated) text models ──────────────────────────────────────
+
+export const AnthropicTextModels: string[] = [
+  AnthropicModels.Opus4_7,
+  AnthropicModels.Opus4_6,
+  AnthropicModels.Sonnet4_6,
+  AnthropicModels.Opus4_5,
+  AnthropicModels.Opus4_1,
+  AnthropicModels.Sonnet4_5,
+  AnthropicModels.Haiku4_5,
+];
+
+// Models in our catalog but not yet publicly available
+export const AnthropicLimitedAvailabilityModels: string[] = [
+  AnthropicModels.Opus4_6Fast, // 404 – not publicly available yet
+];
+
+// ─── All models for catalog (active + limited + deprecated/retired) ───────────
+// Deprecation/replacement metadata is embedded in AnthropicTextPricing entries.
+const AnthropicAllModels: string[] = [
+  ...AnthropicTextModels,
+  ...AnthropicLimitedAvailabilityModels,
+  // Deprecated (retirement 2026-06-15)
+  AnthropicModels.Opus4,
+  AnthropicModels.Sonnet4,
+  // Retired — kept for historical cost tracking
+  AnthropicModels.Sonnet3_7,
+  AnthropicModels.Sonnet3_5,
+  AnthropicModels.Sonnet3_5_20240620,
+  AnthropicModels.Haiku3_5_Latest,
+  AnthropicModels.Sonnet3,
+  AnthropicModels.Opus4_0,
+  AnthropicModels.Sonnet4_0,
+  AnthropicModels.Haiku3_5,
+  AnthropicModels.Opus3,
+  AnthropicModels.Haiku3,
+];
+
+// ─── Pricing (USD per 1M tokens) ──────────────────────────────────────────────
+
+export const AnthropicTextPricing: Record<string, ModelPricing> = {
+  [AnthropicModels.Opus4_7]:   { input: 5.0,  cache_write: 6.25,   cache_hit: 0.5,  output: 25.0 },
+  [AnthropicModels.Opus4_6]:   { input: 5.0,  cache_write: 6.25,   cache_hit: 0.5,  output: 25.0 },
+  [AnthropicModels.Opus4_6Fast]: { input: 30.0, cache_write: 37.5, cache_hit: 3.0,  output: 150.0, limitedAvailability: true },
+  [AnthropicModels.Sonnet4_6]: { input: 3.0,  cache_write: 3.75,   cache_hit: 0.3,  output: 15.0 },
+  [AnthropicModels.Opus4_5]:   { input: 5.0,  cache_write: 6.25,   cache_hit: 0.5,  output: 25.0 },
+  [AnthropicModels.Opus4_1]:   { input: 15.0, cache_write: 18.75,  cache_hit: 1.5,  output: 75.0 },
+  [AnthropicModels.Sonnet4_5]: { input: 3.0,  input_gt_200k: 6.0,  cache_write: 3.75, cache_hit: 0.3, output: 15.0, output_gt_200k: 22.5 },
+  [AnthropicModels.Haiku4_5]:  { input: 1.0,  cache_write: 1.25,   cache_hit: 0.1,  output: 5.0 },
+  // Deprecated — pricing retained for cost tracking
+  [AnthropicModels.Opus4]:     { input: 15.0, cache_write: 18.75,  cache_hit: 1.5,  output: 75.0,  deprecated: true, deprecationDate: "2026-06-15" },
+  [AnthropicModels.Sonnet4]:   { input: 3.0,  input_gt_200k: 6.0,  cache_write: 3.75, cache_hit: 0.3, output: 15.0, output_gt_200k: 22.5, deprecated: true, deprecationDate: "2026-06-15" },
+  // Retired — pricing retained for cost tracking of historical usage
+  [AnthropicModels.Sonnet3_7]: { input: 3.0,  cache_write: 3.75,   cache_hit: 0.3,  output: 15.0,  deprecated: true, deprecationDate: "2026-02-19" },
+  [AnthropicModels.Sonnet3_5]: { input: 3.0,  cache_write: 3.75,   cache_hit: 0.3,  output: 15.0,  deprecated: true, deprecationDate: "2025-10-28" },
+  [AnthropicModels.Sonnet3_5_20240620]: { input: 3.0, cache_write: 3.75, cache_hit: 0.3, output: 15.0, deprecated: true, deprecationDate: "2025-10-28" },
+  [AnthropicModels.Haiku3_5_Latest]:   { input: 0.8, cache_write: 1.0,   cache_hit: 0.08, output: 4.0,  deprecated: true, deprecationDate: "2026-02-19" },
+  [AnthropicModels.Sonnet3]:           { input: 3.0, cache_write: 3.75,  cache_hit: 0.3,  output: 15.0, deprecated: true },
+  [AnthropicModels.Opus4_0]:           { input: 15.0, cache_write: 18.75, cache_hit: 1.5, output: 75.0, deprecated: true, deprecationDate: "2026-06-15" },
+  [AnthropicModels.Sonnet4_0]:         { input: 3.0,  cache_write: 3.75,  cache_hit: 0.3,  output: 15.0, deprecated: true, deprecationDate: "2026-06-15" },
+  [AnthropicModels.Haiku3_5]:  { input: 0.8,  cache_write: 1.0,    cache_hit: 0.08, output: 4.0,   deprecated: true, deprecationDate: "2026-02-19" },
+  [AnthropicModels.Opus3]:     { input: 15.0, cache_write: 18.75,  cache_hit: 1.5,  output: 75.0,  deprecated: true, deprecationDate: "2026-01-05" },
+  [AnthropicModels.Haiku3]:    { input: 0.25, cache_write: 0.3125, cache_hit: 0.025, output: 1.25, deprecated: true, deprecationDate: "2026-04-20" },
 };
+
+// ─── Catalog ──────────────────────────────────────────────────────────────────
+// Metadata (deprecated, deprecationDate, limitedAvailability) is read directly
+// from the AnthropicTextPricing entries — no need for separate groups.
+
+export const ANTHROPIC_MODEL_CATALOG: ModelCatalogEntry[] = [
+  ...completions(AnthropicAllModels, "anthropic", AnthropicTextPricing),
+];

package/src/clients/pricing/cerebras.ts

@@ -0,0 +1,11 @@
+/**
+ * Cerebras pricing (USD per 1M tokens)
+ * Source: https://cerebras.ai/pricing
+ */
+export const CerebrasTextPricing: Record<string, { input: number; output: number }> = {
+  "llama3.1-8b": { input: 0.10, output: 0.10 },
+  "llama3.3-70b": { input: 0.85, output: 1.20 },
+  "qwen-3-235b-a22b-instruct-2507": { input: 0.60, output: 1.20 },
+  "gpt-oss-120b": { input: 0.25, output: 0.69 },
+  "zai-glm-4.7": { input: 2.25, output: 2.75 },
+};

package/src/clients/pricing/copilot.ts

@@ -0,0 +1,60 @@
+/**
+ * GitHub Copilot pricing — "premium request" multiplier system.
+ *
+ * GitHub Copilot does NOT charge per token. Instead each model call costs a
+ * certain number of "premium requests" from the subscriber's monthly allowance.
+ * All models are effectively $0/token from the perspective of knowhow pricing,
+ * but we store the multiplier as metadata for informational purposes.
+ *
+ * Copilot Individual: 300 premium requests/month
+ * Copilot Pro:        300 premium requests/month
+ * Copilot Business:  300 premium requests/month (per seat)
+ * Copilot Enterprise: 300 premium requests/month (per seat)
+ *
+ * Model IDs are as returned by https://api.githubcopilot.com/models
+ * Source: https://docs.github.com/en/copilot/reference/ai-models/supported-models
+ */
+
+/** Copilot premium request multipliers (informational only) */
+export const CopilotModelMultipliers: Record<string, number> = {
+  // OpenAI models
+  "gpt-4o": 0,          // free (base request)
+  "gpt-4.1": 0,         // free (base request)
+  "gpt-5-mini": 0,      // free (base request)
+  "gpt-5.2": 1,
+  "gpt-5.2-codex": 1,
+  "gpt-5.3-codex": 1,
+  "gpt-5.4": 1,
+  "gpt-5.4-mini": 0.33,
+  "gpt-5.4-nano": 0.25,
+  "gpt-5.5": 7.5,       // promotional until further notice
+  "gpt-5": 1,
+  "gpt-5.1": 1,
+  "gpt-5.1-codex": 1,
+  "gpt-5.1-codex-max": 1,
+  "gpt-5.1-codex-mini": 1,
+  // Anthropic models
+  "claude-haiku-4.5": 0.33,
+  "claude-sonnet-4": 1,
+  "claude-sonnet-4.5": 1,
+  "claude-sonnet-4.6": 1,
+  "claude-opus-4.5": 3,
+  "claude-opus-4.6": 3,
+  "claude-opus-41": 3,
+  "claude-opus-4.7": 7.5, // promotional until April 30, 2026
+  // Google models
+  "gemini-2.5-pro": 1,
+  "gemini-3-flash-preview": 0.33,
+  "gemini-3.1-pro-preview": 1,
+  "gemini-3-pro-preview": 1,
+  // xAI models
+  "grok-code-fast-1": 0.25,
+};
+
+/**
+ * Copilot pricing is all $0/token — consumption is via premium request allowance.
+ * This map is used by the check-model-pricing script to recognise registered models.
+ */
+export const CopilotTextPricing: Record<string, { input: number; output: number }> = Object.fromEntries(
+  Object.keys(CopilotModelMultipliers).map((id) => [id, { input: 0.0, output: 0.0 }])
+);

package/src/clients/pricing/deepseek.ts

@@ -0,0 +1,15 @@
+/**
+ * DeepSeek pricing (USD per 1M tokens)
+ * Source: https://platform.deepseek.com/pricing
+ */
+export const DeepSeekTextPricing: Record<string, { input: number; output: number; cached_input?: number }> = {
+  // DeepSeek V3 / deepseek-chat — updated 2026-04 (now routes to DeepSeek V3-0324)
+  // Source: api-docs.deepseek.com/quick_start/pricing
+  "deepseek-chat": { input: 0.28, output: 0.42, cached_input: 0.028 },
+  // DeepSeek R1 / deepseek-reasoner — same unified pricing as V3
+  // Source: api-docs.deepseek.com/quick_start/pricing (costgoat.com verification)
+  "deepseek-reasoner": { input: 0.28, output: 0.42, cached_input: 0.028 },
+  // DeepSeek V4 Flash & Pro
+  "deepseek-v4-flash": { input: 0.14, output: 0.28 },
+  "deepseek-v4-pro": { input: 1.74, output: 3.48 },
+};

package/src/clients/pricing/fireworks.ts

@@ -0,0 +1,32 @@
+/**
+ * Fireworks AI pricing (USD per 1M tokens)
+ * Source: https://fireworks.ai/pricing
+ */
+export const FireworksTextPricing: Record<string, { input: number; output: number; cache_hit?: number }> = {
+  // Moonshot AI
+  "accounts/fireworks/models/kimi-k2-6": { input: 0.95, cache_hit: 0.16, output: 4.0 },
+  "accounts/fireworks/models/kimi-k2-5": { input: 0.60, cache_hit: 0.10, output: 3.0 },
+
+  // MiniMax
+  "accounts/fireworks/models/minimax-m2-7": { input: 0.30, cache_hit: 0.06, output: 1.20 },
+  "accounts/fireworks/models/minimax-m2-5": { input: 0.30, cache_hit: 0.03, output: 1.20 },
+
+  // Qwen
+  "accounts/fireworks/models/qwen3-6-plus": { input: 0.50, cache_hit: 0.10, output: 3.0 },
+  "accounts/fireworks/models/qwen3p6-plus": { input: 0.50, cache_hit: 0.10, output: 3.0 },
+  "accounts/fireworks/models/qwen3-vl-30b-a3b-thinking": { input: 0.15, cache_hit: 0.08, output: 0.60 },
+  "accounts/fireworks/models/qwen3-vl-30b-a3b-instruct": { input: 0.15, cache_hit: 0.08, output: 0.60 },
+  "accounts/fireworks/models/qwen3-8b": { input: 0.20, cache_hit: 0.10, output: 0.20 },
+
+  // Z.ai
+  "accounts/fireworks/models/glm-5-1": { input: 1.40, cache_hit: 0.26, output: 4.40 },
+  "accounts/fireworks/models/glm-5": { input: 1.00, cache_hit: 0.20, output: 3.20 },
+  "accounts/fireworks/models/glm-4-7": { input: 0.60, cache_hit: 0.30, output: 2.20 },
+
+  // DeepSeek AI
+  "accounts/fireworks/models/deepseek-v3-2": { input: 0.56, cache_hit: 0.28, output: 1.68 },
+  "accounts/fireworks/models/deepseek-v3-1": { input: 0.56, cache_hit: 0.28, output: 1.68 },
+
+  // OpenAI OSS
+  "accounts/fireworks/models/gpt-oss-120b": { input: 0.15, cache_hit: 0.01, output: 0.60 },
+};

package/src/clients/pricing/github.ts

@@ -0,0 +1,69 @@
+/**
+ * GitHub Models pricing (USD per 1M tokens)
+ * Source: https://github.com/marketplace/models
+ * All models are free within rate limits for GitHub users.
+ * Low-tier (higher rate limits) and high-tier (lower rate limits) exist.
+ */
+export const GitHubModelsTextPricing: Record<string, { input: number; output: number }> = {
+  // OpenAI via GitHub (free)
+  "openai/gpt-4o": { input: 0.0, output: 0.0 },
+  "openai/gpt-4o-mini": { input: 0.0, output: 0.0 },
+  "openai/o1": { input: 0.0, output: 0.0 },
+  "openai/o1-mini": { input: 0.0, output: 0.0 },
+  "openai/o3-mini": { input: 0.0, output: 0.0 },
+  "openai/o4-mini": { input: 0.0, output: 0.0 },
+  "openai/gpt-4.1": { input: 0.0, output: 0.0 },
+  "openai/gpt-4.1-mini": { input: 0.0, output: 0.0 },
+  "openai/gpt-4.1-nano": { input: 0.0, output: 0.0 },
+  "openai/o3": { input: 0.0, output: 0.0 },
+  // DeepSeek via GitHub (free)
+  "deepseek/deepseek-r1": { input: 0.0, output: 0.0 },
+  "deepseek/deepseek-v3-0324": { input: 0.0, output: 0.0 },
+  // Microsoft Phi via GitHub (free)
+  "microsoft/phi-4": { input: 0.0, output: 0.0 },
+  "microsoft/phi-4-mini-instruct": { input: 0.0, output: 0.0 },
+  "microsoft/phi-4-multimodal-instruct": { input: 0.0, output: 0.0 },
+  "microsoft/phi-4-mini-reasoning": { input: 0.0, output: 0.0 },
+  "microsoft/phi-4-reasoning": { input: 0.0, output: 0.0 },
+  "microsoft/phi-3-medium-128k-instruct": { input: 0.0, output: 0.0 },
+  "microsoft/phi-3-medium-4k-instruct": { input: 0.0, output: 0.0 },
+  "microsoft/phi-3-mini-128k-instruct": { input: 0.0, output: 0.0 },
+  "microsoft/phi-3-mini-4k-instruct": { input: 0.0, output: 0.0 },
+  "microsoft/phi-3-small-128k-instruct": { input: 0.0, output: 0.0 },
+  "microsoft/phi-3-small-8k-instruct": { input: 0.0, output: 0.0 },
+  "microsoft/phi-3.5-mini-instruct": { input: 0.0, output: 0.0 },
+  "microsoft/phi-3.5-moe-instruct": { input: 0.0, output: 0.0 },
+  "microsoft/phi-3.5-vision-instruct": { input: 0.0, output: 0.0 },
+  "microsoft/mai-ds-r1": { input: 0.0, output: 0.0 },
+  // Meta Llama via GitHub (free)
+  "meta/llama-3.3-70b-instruct": { input: 0.0, output: 0.0 },
+  "meta/llama-3.2-11b-vision-instruct": { input: 0.0, output: 0.0 },
+  "meta/llama-3.1-405b-instruct": { input: 0.0, output: 0.0 },
+  "meta/llama-3.2-90b-vision-instruct": { input: 0.0, output: 0.0 },
+  "meta/llama-4-maverick-17b-128e-instruct-fp8": { input: 0.0, output: 0.0 },
+  "meta/llama-4-scout-17b-16e-instruct": { input: 0.0, output: 0.0 },
+  "meta/meta-llama-3-70b-instruct": { input: 0.0, output: 0.0 },
+  "meta/meta-llama-3-8b-instruct": { input: 0.0, output: 0.0 },
+  "meta/meta-llama-3.1-405b-instruct": { input: 0.0, output: 0.0 },
+  "meta/meta-llama-3.1-70b-instruct": { input: 0.0, output: 0.0 },
+  "meta/meta-llama-3.1-8b-instruct": { input: 0.0, output: 0.0 },
+  // Mistral via GitHub (free)
+  "mistral-ai/mistral-large-2411": { input: 0.0, output: 0.0 },
+  "mistral-ai/mistral-small-2503": { input: 0.0, output: 0.0 },
+  "mistral-ai/codestral-2501": { input: 0.0, output: 0.0 },
+  "mistral-ai/ministral-3b": { input: 0.0, output: 0.0 },
+  "mistral-ai/mistral-medium-2505": { input: 0.0, output: 0.0 },
+  "mistral-ai/mistral-nemo": { input: 0.0, output: 0.0 },
+  // Cohere via GitHub (free)
+  "cohere/cohere-command-r-plus-08-2024": { input: 0.0, output: 0.0 },
+  "cohere/cohere-command-r-08-2024": { input: 0.0, output: 0.0 },
+  "cohere/cohere-command-a": { input: 0.0, output: 0.0 },
+  // AI21 via GitHub (free)
+  "ai21-labs/ai21-jamba-1.5-mini": { input: 0.0, output: 0.0 },
+  "ai21-labs/ai21-jamba-1.5-large": { input: 0.0, output: 0.0 },
+  // Core42
+  "core42/jais-30b-chat": { input: 0.0, output: 0.0 },
+  // xAI via GitHub (free)
+  "xai/grok-3": { input: 0.0, output: 0.0 },
+  "xai/grok-3-mini": { input: 0.0, output: 0.0 },
+};