@juspay/neurolink 12.4.4 → 12.5.1

package/dist/utils/modelChoices.js

@@ -2,10 +2,45 @@
  * Centralized model choices for CLI commands
  * Derives choices from model enums to ensure consistency
  */
-import { AIProviderName, OpenAIModels, AnthropicModels, GoogleAIModels, BedrockModels, VertexModels, MistralModels, OllamaModels, AzureOpenAIModels, LiteLLMModels, HuggingFaceModels, SageMakerModels, OpenRouterModels, DeepSeekModels, NvidiaNimModels, XaiModels, GroqModels, CerebrasModels, SambanovaModels, CohereModels, TogetherAIModels, FireworksModels, PerplexityModels, CloudflareModels, VoyageModels, JinaModels, StabilityModels, IdeogramModels, RecraftModels, ReplicateModels, } from "../constants/enums.js";
+import { AIProviderName, OpenAIModels, AnthropicModels, GoogleAIModels, BedrockModels, VertexModels, OllamaModels, AzureOpenAIModels, LiteLLMModels, HuggingFaceModels, SageMakerModels, OpenRouterModels, DeepSeekModels, NvidiaNimModels, CohereModels, VoyageModels, JinaModels, StabilityModels, IdeogramModels, RecraftModels, ReplicateModels, } from "../constants/enums.js";
+import { getCatalogJsonEntries } from "../providers/catalog/loader.js";
+/**
+ * Looks up the JSON catalog entry for a provider, if it is one of the 9
+ * JSON-catalog providers. Used to short-circuit the hand tables below so a
+ * catalog provider's models are always sourced from its catalog JSON, never
+ * from stale hand-written duplicates.
+ */
+function catalogEntryFor(provider) {
+    return getCatalogJsonEntries().find((entry) => entry.id === provider);
+}
+/**
+ * Derives CLI-prompt entries for a catalog provider. Uses the curated
+ * `models.topModels` ordering when the JSON provides one (mapping each id to
+ * its own `models.catalog[id].description`); falls back to the full
+ * `models.catalog` in file key-order when `topModels` is absent, matching
+ * the loader's documented fallback contract.
+ */
+function catalogTopModels(entry) {
+    if (entry.models.topModels) {
+        return entry.models.topModels.map((model) => ({
+            model,
+            description: entry.models.catalog[model].description,
+        }));
+    }
+    return Object.entries(entry.models.catalog).map(([model, spec]) => ({
+        model,
+        description: spec.description,
+    }));
+}
 /**
  * Top models per provider with descriptions for CLI prompts
  * These are curated lists of the most commonly used/recommended models
+ *
+ * Covers every provider EXCEPT the 9 JSON-catalog providers (cerebras,
+ * cloudflare, fireworks, groq, mistral, perplexity, sambanova, together-ai,
+ * xai) — their entries are derived from `models.catalog[*].description` by
+ * `catalogTopModels()` and merged in at the accessor-function level below,
+ * never hand-duplicated here.
  */
 const TOP_MODELS_CONFIG = {
     [AIProviderName.OPENAI]: [
@@ -125,25 +160,6 @@ const TOP_MODELS_CONFIG = {
         },
         { model: AzureOpenAIModels.GPT_3_5_TURBO, description: "Legacy model" },
     ],
-    [AIProviderName.MISTRAL]: [
-        {
-            model: MistralModels.MISTRAL_LARGE_LATEST,
-            description: "Recommended - Flagship model",
-        },
-        {
-            model: MistralModels.MISTRAL_SMALL_LATEST,
-            description: "Cost-effective",
-        },
-        {
-            model: MistralModels.CODESTRAL_LATEST,
-            description: "Specialized for code",
-        },
-        {
-            model: MistralModels.MAGISTRAL_MEDIUM_LATEST,
-            description: "Reasoning model",
-        },
-        { model: MistralModels.MISTRAL_NEMO, description: "Efficient base model" },
-    ],
     [AIProviderName.OLLAMA]: [
         {
             model: OllamaModels.LLAMA4_LATEST,
@@ -270,75 +286,6 @@ const TOP_MODELS_CONFIG = {
             description: "Use whatever model llama-server has loaded",
         },
     ],
-    [AIProviderName.XAI]: [
-        {
-            model: XaiModels.GROK_3,
-            description: "Recommended - Latest flagship Grok",
-        },
-        { model: XaiModels.GROK_3_MINI, description: "Faster + cheaper Grok 3" },
-        {
-            model: XaiModels.GROK_2_VISION_LATEST,
-            description: "Multimodal (text + images)",
-        },
-        { model: XaiModels.GROK_2_LATEST, description: "Previous flagship" },
-        { model: XaiModels.GROK_BETA, description: "Pre-release / experimental" },
-    ],
-    [AIProviderName.GROQ]: [
-        {
-            model: GroqModels.LLAMA_3_3_70B_VERSATILE,
-            description: "Recommended - Production default; sub-100ms",
-        },
-        {
-            model: GroqModels.LLAMA_3_1_8B_INSTANT,
-            description: "Lowest latency tier",
-        },
-        {
-            model: GroqModels.LLAMA_3_2_90B_VISION_PREVIEW,
-            description: "Multimodal (vision)",
-        },
-        { model: GroqModels.GEMMA_2_9B_IT, description: "Google Gemma 2 9B" },
-        {
-            model: GroqModels.MIXTRAL_8X7B_32768,
-            description: "Mistral 8x7B MoE, 32K context",
-        },
-    ],
-    [AIProviderName.SAMBANOVA]: [
-        {
-            model: SambanovaModels.META_LLAMA_3_3_70B_INSTRUCT,
-            description: "Recommended - Meta Llama 3.3 70B; production, 128K context",
-        },
-        {
-            model: SambanovaModels.GPT_OSS_120B,
-            description: "OpenAI GPT-OSS 120B (open-weight)",
-        },
-        {
-            model: SambanovaModels.DEEPSEEK_V3_1,
-            description: "DeepSeek V3.1 (reasoning)",
-        },
-        {
-            model: SambanovaModels.DEEPSEEK_V3_2,
-            description: "DeepSeek V3.2 (reasoning; vendor preview, 32K context)",
-        },
-        {
-            model: SambanovaModels.MINIMAX_M2_7,
-            description: "MiniMax M2.7, 192K context",
-        },
-        { model: SambanovaModels.MINIMAX_M3, description: "MiniMax M3 (vision)" },
-        {
-            model: SambanovaModels.GEMMA_4_31B_IT,
-            description: "Google Gemma 4 31B IT (vision; vendor preview)",
-        },
-    ],
-    [AIProviderName.CEREBRAS]: [
-        {
-            model: CerebrasModels.GPT_OSS_120B,
-            description: "Recommended - OpenAI GPT-OSS 120B (open-weight); wafer-scale speed",
-        },
-        {
-            model: CerebrasModels.GEMMA_4_31B,
-            description: "Google Gemma 4 31B",
-        },
-    ],
     [AIProviderName.COHERE]: [
         {
             model: CohereModels.COMMAND_R_PLUS,
@@ -347,74 +294,6 @@ const TOP_MODELS_CONFIG = {
         { model: CohereModels.COMMAND_R, description: "Smaller RAG-tuned chat" },
         { model: CohereModels.COMMAND_R7B, description: "Most compact" },
     ],
-    [AIProviderName.TOGETHER_AI]: [
-        {
-            model: TogetherAIModels.LLAMA_3_3_70B_INSTRUCT_TURBO,
-            description: "Recommended - Llama 3.3 70B Turbo",
-        },
-        {
-            model: TogetherAIModels.LLAMA_3_1_405B_INSTRUCT_TURBO,
-            description: "Flagship 405B",
-        },
-        {
-            model: TogetherAIModels.QWEN_2_5_72B_INSTRUCT_TURBO,
-            description: "Qwen 2.5 72B Turbo",
-        },
-        {
-            model: TogetherAIModels.DEEPSEEK_R1,
-            description: "DeepSeek R1 reasoning",
-        },
-        {
-            model: TogetherAIModels.MIXTRAL_8X22B_INSTRUCT,
-            description: "Mistral 8x22B MoE",
-        },
-    ],
-    [AIProviderName.FIREWORKS]: [
-        {
-            model: FireworksModels.DEEPSEEK_V4_PRO,
-            description: "Recommended - DeepSeek V4 Pro",
-        },
-        { model: FireworksModels.GLM_5P1, description: "GLM 5.1 (Zhipu)" },
-        { model: FireworksModels.KIMI_K2P6, description: "Kimi K2.6 (Moonshot)" },
-        { model: FireworksModels.GPT_OSS_120B, description: "GPT-OSS 120B" },
-    ],
-    [AIProviderName.PERPLEXITY]: [
-        {
-            model: PerplexityModels.SONAR,
-            description: "Recommended - Sonar with web grounding",
-        },
-        { model: PerplexityModels.SONAR_PRO, description: "Better reasoning" },
-        {
-            model: PerplexityModels.SONAR_REASONING,
-            description: "Explicit reasoning traces",
-        },
-        {
-            model: PerplexityModels.SONAR_REASONING_PRO,
-            description: "Flagship reasoning + web",
-        },
-        {
-            model: PerplexityModels.SONAR_DEEP_RESEARCH,
-            description: "Long-form research with citations",
-        },
-    ],
-    [AIProviderName.CLOUDFLARE]: [
-        {
-            model: CloudflareModels.LLAMA_3_3_70B_FAST,
-            description: "Recommended - Llama 3.3 70B FP8 fast",
-        },
-        {
-            model: CloudflareModels.LLAMA_3_1_70B_INSTRUCT,
-            description: "Llama 3.1 70B",
-        },
-        {
-            model: CloudflareModels.LLAMA_3_1_8B_FAST,
-            description: "Llama 3.1 8B fast",
-        },
-        {
-            model: CloudflareModels.LLAMA_3_2_11B_VISION,
-            description: "Multimodal (vision)",
-        },
-    ],
     [AIProviderName.REPLICATE]: [
         {
             model: "meta/meta-llama-3-70b-instruct",
@@ -517,7 +396,15 @@ const TOP_MODELS_CONFIG = {
     [AIProviderName.AUTO]: [],
 };
 /**
- * Default models per provider (first choice/recommended)
+ * Default models per provider (first choice/recommended).
+ *
+ * Covers every provider EXCEPT the 9 JSON-catalog providers — those default
+ * models come from `models.default` in the catalog JSON, read directly by
+ * `getDefaultModel()` below. (Two of the 9 — cerebras, sambanova — never had
+ * a hand entry here at all; `getDefaultModel()` now resolves them too.)
+ *
+ * AUTO is also excluded — it never had an entry here either (matches
+ * pre-existing behavior: `getDefaultModel(AUTO)` returns `undefined`).
  */
 export const DEFAULT_MODELS = {
     [AIProviderName.OPENAI]: OpenAIModels.GPT_4O,
@@ -526,7 +413,6 @@ export const DEFAULT_MODELS = {
     [AIProviderName.VERTEX]: VertexModels.GEMINI_2_5_FLASH,
     [AIProviderName.BEDROCK]: BedrockModels.CLAUDE_4_5_SONNET,
     [AIProviderName.AZURE]: AzureOpenAIModels.GPT_4O,
-    [AIProviderName.MISTRAL]: MistralModels.MISTRAL_LARGE_LATEST,
     [AIProviderName.OLLAMA]: OllamaModels.LLAMA4_LATEST,
     [AIProviderName.LITELLM]: LiteLLMModels.OPENAI_GPT_4O,
     [AIProviderName.HUGGINGFACE]: HuggingFaceModels.LLAMA_3_3_70B_INSTRUCT,
@@ -539,13 +425,7 @@ export const DEFAULT_MODELS = {
     // an empty default is the documented signal to use that path.
     [AIProviderName.LM_STUDIO]: "",
     [AIProviderName.LLAMACPP]: "",
-    [AIProviderName.XAI]: XaiModels.GROK_3,
-    [AIProviderName.GROQ]: GroqModels.LLAMA_3_3_70B_VERSATILE,
     [AIProviderName.COHERE]: CohereModels.COMMAND_R_PLUS,
-    [AIProviderName.TOGETHER_AI]: TogetherAIModels.LLAMA_3_3_70B_INSTRUCT_TURBO,
-    [AIProviderName.FIREWORKS]: FireworksModels.DEEPSEEK_V4_PRO,
-    [AIProviderName.PERPLEXITY]: PerplexityModels.SONAR,
-    [AIProviderName.CLOUDFLARE]: CloudflareModels.LLAMA_3_3_70B_FAST,
     [AIProviderName.REPLICATE]: "meta/meta-llama-3-70b-instruct",
     [AIProviderName.VOYAGE]: VoyageModels.VOYAGE_3_5,
     [AIProviderName.JINA]: JinaModels.JINA_EMBEDDINGS_V3,
@@ -554,7 +434,11 @@ export const DEFAULT_MODELS = {
     [AIProviderName.RECRAFT]: RecraftModels.RECRAFT_V3,
 };
 /**
- * Model enum mappings for getAllModels
+ * Model enum mappings for getAllModels.
+ *
+ * Covers every provider EXCEPT the 9 JSON-catalog providers — those model
+ * lists come from `Object.keys(models.catalog)` in the catalog JSON, read
+ * directly by `getAllModels()` below.
  */
 const MODEL_ENUMS = {
     [AIProviderName.OPENAI]: OpenAIModels,
@@ -563,7 +447,6 @@ const MODEL_ENUMS = {
     [AIProviderName.VERTEX]: VertexModels,
     [AIProviderName.BEDROCK]: BedrockModels,
     [AIProviderName.AZURE]: AzureOpenAIModels,
-    [AIProviderName.MISTRAL]: MistralModels,
     [AIProviderName.OLLAMA]: OllamaModels,
     [AIProviderName.LITELLM]: LiteLLMModels,
     [AIProviderName.HUGGINGFACE]: HuggingFaceModels,
@@ -574,15 +457,7 @@ const MODEL_ENUMS = {
     [AIProviderName.NVIDIA_NIM]: NvidiaNimModels,
     [AIProviderName.LM_STUDIO]: null,
     [AIProviderName.LLAMACPP]: null,
-    [AIProviderName.XAI]: XaiModels,
-    [AIProviderName.GROQ]: GroqModels,
-    [AIProviderName.CEREBRAS]: CerebrasModels,
-    [AIProviderName.SAMBANOVA]: SambanovaModels,
     [AIProviderName.COHERE]: CohereModels,
-    [AIProviderName.TOGETHER_AI]: TogetherAIModels,
-    [AIProviderName.FIREWORKS]: FireworksModels,
-    [AIProviderName.PERPLEXITY]: PerplexityModels,
-    [AIProviderName.CLOUDFLARE]: CloudflareModels,
     [AIProviderName.REPLICATE]: ReplicateModels,
     [AIProviderName.VOYAGE]: VoyageModels,
     [AIProviderName.JINA]: JinaModels,
@@ -600,7 +475,10 @@ const MODEL_ENUMS = {
  * @returns Array of ModelChoice objects for CLI prompts
  */
 export function getTopModelChoices(provider, limit = 5) {
-    const config = TOP_MODELS_CONFIG[provider];
+    const catalogEntry = catalogEntryFor(provider);
+    const config = catalogEntry
+        ? catalogTopModels(catalogEntry)
+        : TOP_MODELS_CONFIG[provider];
     if (!config || config.length === 0) {
         return [];
     }
@@ -632,6 +510,10 @@ export function getTopModelChoices(provider, limit = 5) {
  * @returns Array of model identifier strings
  */
 export function getAllModels(provider) {
+    const catalogEntry = catalogEntryFor(provider);
+    if (catalogEntry) {
+        return Object.keys(catalogEntry.models.catalog);
+    }
     const modelEnum = MODEL_ENUMS[provider];
     if (!modelEnum) {
         return [];
@@ -662,6 +544,10 @@ export function getAllProviderChoices() {
  * @returns Default model string for the provider
  */
 export function getDefaultModel(provider) {
+    const catalogEntry = catalogEntryFor(provider);
+    if (catalogEntry) {
+        return catalogEntry.models.default;
+    }
     return DEFAULT_MODELS[provider];
 }
 /**
@@ -714,23 +600,31 @@ export function getModelChoicesWithDefault(provider, currentModel, limit = 5) {
  */
 export function getPopularModelsAcrossProviders() {
     const popularModels = [];
+    // Take top 2 from each provider, ignoring blank auto-discovery sentinels.
+    // (Auto-discovery is surfaced separately by `getTopModelChoices` for
+    // LM Studio / llama.cpp; we don't want it to appear in the cross-
+    // provider popular-models list as an empty value.)
+    const pushTop2 = (provider, config) => {
+        config
+            .filter((item) => item.model.length > 0)
+            .slice(0, 2)
+            .forEach((item) => {
+            popularModels.push({
+                provider,
+                model: item.model,
+                description: item.description,
+            });
+        });
+    };
     for (const [provider, config] of Object.entries(TOP_MODELS_CONFIG)) {
         if (config && config.length > 0) {
-            // Take top 2 from each provider, ignoring blank auto-discovery sentinels.
-            // (Auto-discovery is surfaced separately by `getTopModelChoices` for
-            // LM Studio / llama.cpp; we don't want it to appear in the cross-
-            // provider popular-models list as an empty value.)
-            config
-                .filter((item) => item.model.length > 0)
-                .slice(0, 2)
-                .forEach((item) => {
-                popularModels.push({
-                    provider: provider,
-                    model: item.model,
-                    description: item.description,
-                });
-            });
+            pushTop2(provider, config);
         }
     }
+    // JSON-catalog providers are excluded from TOP_MODELS_CONFIG above; merge
+    // their models in here, derived the same way getTopModelChoices() does.
+    for (const entry of getCatalogJsonEntries()) {
+        pushTop2(entry.id, catalogTopModels(entry));
+    }
     return popularModels;
 }

package/dist/utils/pricing.js

@@ -1,4 +1,69 @@
 import { getManifestForProvider, resolveManifestEntryExact, } from "../models/manifestRegistry.js";
+import { getCatalogJsonEntries } from "../providers/catalog/loader.js";
+/**
+ * Per-model pricing for the JSON-catalog providers, derived from
+ * models.catalog[*].pricingPerMTok (dollars-per-million → dollars-per-token,
+ * matching this file's unit convention). `_default` is the rate of the
+ * provider's own models.default model — mirroring what every hand-written
+ * catalog-provider block already did pre-migration (e.g. xai's `_default`
+ * was always grok-3's rate, xai's default model).
+ *
+ * `pricingPerMTok.cachedInput`, where present, maps to this file's
+ * `cacheRead` field — same semantics (price for a cached-read hit), just a
+ * different field name in the two shapes.
+ *
+ * A provider contributes no block here when its catalog carries zero
+ * `pricingPerMTok` data anywhere — fireworks (a fully refreshed model
+ * roster with no sourced pricing yet) and cloudflare (bills per "neuron",
+ * not per token — there's no USD-per-token figure to source). Both stay
+ * entirely hand-written below with their pre-existing symbolic/estimated
+ * rates, since the JSON has nothing to derive. Mistral's catalog likewise
+ * has no pricingPerMTok data, and its hand block below uses coarse model
+ * names ("mistral-large", "codestral", …) that don't match the JSON's real
+ * ids ("mistral-large-latest", "codestral-latest", …) — it stays
+ * hand-written too.
+ */
+const CATALOG_PRICING = Object.fromEntries(getCatalogJsonEntries().flatMap((entry) => {
+    const priced = Object.fromEntries(Object.entries(entry.models.catalog).flatMap(([modelId, spec]) => spec.pricingPerMTok
+        ? [
+            [
+                modelId,
+                {
+                    input: spec.pricingPerMTok.input / 1_000_000,
+                    output: spec.pricingPerMTok.output / 1_000_000,
+                    ...(spec.pricingPerMTok.cachedInput !== undefined
+                        ? { cacheRead: spec.pricingPerMTok.cachedInput / 1_000_000 }
+                        : {}),
+                },
+            ],
+        ]
+        : []));
+    if (Object.keys(priced).length === 0) {
+        return [];
+    }
+    const defaultRate = entry.models.catalog[entry.models.default]?.pricingPerMTok;
+    return [
+        [
+            entry.id,
+            {
+                ...(defaultRate
+                    ? {
+                        _default: {
+                            input: defaultRate.input / 1_000_000,
+                            output: defaultRate.output / 1_000_000,
+                            // Without this, a request falling back to `_default`
+                            // bills cached tokens at the full input rate.
+                            ...(defaultRate.cachedInput !== undefined
+                                ? { cacheRead: defaultRate.cachedInput / 1_000_000 }
+                                : {}),
+                        },
+                    }
+                    : {}),
+                ...priced,
+            },
+        ],
+    ];
+}));
 /**
  * Per-token pricing data (USD per token). Updated Feb 2026.
  * Sources:
@@ -483,65 +548,9 @@ const PRICING = {
     llamacpp: {
         _default: { input: 0, output: 0 },
     },
-    xai: {
-        _default: { input: 3.0 / 1_000_000, output: 15.0 / 1_000_000 },
-        "grok-3": { input: 3.0 / 1_000_000, output: 15.0 / 1_000_000 },
-        "grok-3-mini": { input: 0.3 / 1_000_000, output: 0.5 / 1_000_000 },
-        "grok-2-latest": { input: 2.0 / 1_000_000, output: 10.0 / 1_000_000 },
-        "grok-2-vision-latest": {
-            input: 2.0 / 1_000_000,
-            output: 10.0 / 1_000_000,
-        },
-        "grok-beta": { input: 5.0 / 1_000_000, output: 15.0 / 1_000_000 },
-    },
-    groq: {
-        _default: { input: 0.59 / 1_000_000, output: 0.79 / 1_000_000 },
-        "llama-3.3-70b-versatile": {
-            input: 0.59 / 1_000_000,
-            output: 0.79 / 1_000_000,
-        },
-        "llama-3.1-8b-instant": {
-            input: 0.05 / 1_000_000,
-            output: 0.08 / 1_000_000,
-        },
-        "llama-3.2-90b-vision-preview": {
-            input: 0.9 / 1_000_000,
-            output: 0.9 / 1_000_000,
-        },
-        "llama-3.2-11b-vision-preview": {
-            input: 0.18 / 1_000_000,
-            output: 0.18 / 1_000_000,
-        },
-        "gemma2-9b-it": { input: 0.2 / 1_000_000, output: 0.2 / 1_000_000 },
-        "mixtral-8x7b-32768": {
-            input: 0.24 / 1_000_000,
-            output: 0.24 / 1_000_000,
-        },
-    },
-    // inference-docs.cerebras.ai model pages, checked 2026-08-27. The
-    // gemma-4-31b page's prose and structured data disagree ($2.15/$2.70 vs
-    // $0.99/$1.49); the structured data feeds the vendor's rendered pricing
-    // card, so it's used here.
-    cerebras: {
-        _default: { input: 0.35 / 1_000_000, output: 0.75 / 1_000_000 },
-        "gpt-oss-120b": { input: 0.35 / 1_000_000, output: 0.75 / 1_000_000 },
-        "gemma-4-31b": { input: 0.99 / 1_000_000, output: 1.49 / 1_000_000 },
-    },
-    // cloud.sambanova.ai/plans/pricing, checked 2026-08-27. MiniMax-M2.7
-    // also has a $0.06/M cached-input rate the flat model here can't express.
-    sambanova: {
-        _default: { input: 0.6 / 1_000_000, output: 1.2 / 1_000_000 },
-        "Meta-Llama-3.3-70B-Instruct": {
-            input: 0.6 / 1_000_000,
-            output: 1.2 / 1_000_000,
-        },
-        "gpt-oss-120b": { input: 0.22 / 1_000_000, output: 0.59 / 1_000_000 },
-        "DeepSeek-V3.1": { input: 3.0 / 1_000_000, output: 4.5 / 1_000_000 },
-        "DeepSeek-V3.2": { input: 3.0 / 1_000_000, output: 4.5 / 1_000_000 },
-        "MiniMax-M2.7": { input: 0.6 / 1_000_000, output: 2.4 / 1_000_000 },
-        "MiniMax-M3": { input: 0.6 / 1_000_000, output: 2.4 / 1_000_000 },
-        "gemma-4-31B-it": { input: 0.38 / 1_000_000, output: 1.15 / 1_000_000 },
-    },
+    // xai, groq, cerebras, sambanova — derived from the JSON catalog, see
+    // CATALOG_PRICING above.
+    ...CATALOG_PRICING,
     cohere: {
         _default: { input: 2.5 / 1_000_000, output: 10.0 / 1_000_000 },
         "command-r-plus": { input: 2.5 / 1_000_000, output: 10.0 / 1_000_000 },
@@ -551,45 +560,7 @@ const PRICING = {
             output: 0.15 / 1_000_000,
         },
     },
-    "together-ai": {
-        _default: { input: 0.88 / 1_000_000, output: 0.88 / 1_000_000 },
-        "meta-llama/Llama-3.3-70B-Instruct-Turbo": {
-            input: 0.88 / 1_000_000,
-            output: 0.88 / 1_000_000,
-        },
-        "meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": {
-            input: 3.5 / 1_000_000,
-            output: 3.5 / 1_000_000,
-        },
-        "meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo": {
-            input: 0.88 / 1_000_000,
-            output: 0.88 / 1_000_000,
-        },
-        "meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo": {
-            input: 0.18 / 1_000_000,
-            output: 0.18 / 1_000_000,
-        },
-        "mistralai/Mixtral-8x22B-Instruct-v0.1": {
-            input: 1.2 / 1_000_000,
-            output: 1.2 / 1_000_000,
-        },
-        "mistralai/Mixtral-8x7B-Instruct-v0.1": {
-            input: 0.6 / 1_000_000,
-            output: 0.6 / 1_000_000,
-        },
-        "Qwen/Qwen2.5-72B-Instruct-Turbo": {
-            input: 1.2 / 1_000_000,
-            output: 1.2 / 1_000_000,
-        },
-        "deepseek-ai/DeepSeek-R1": {
-            input: 7.0 / 1_000_000,
-            output: 7.0 / 1_000_000,
-        },
-        "deepseek-ai/DeepSeek-V3": {
-            input: 1.25 / 1_000_000,
-            output: 1.25 / 1_000_000,
-        },
-    },
+    // together-ai — derived from the JSON catalog, see CATALOG_PRICING above.
     fireworks: {
         _default: { input: 0.9 / 1_000_000, output: 0.9 / 1_000_000 },
         "accounts/fireworks/models/llama-v3p1-70b-instruct": {
@@ -625,20 +596,7 @@ const PRICING = {
             output: 3.0 / 1_000_000,
         },
     },
-    perplexity: {
-        _default: { input: 1.0 / 1_000_000, output: 1.0 / 1_000_000 },
-        sonar: { input: 1.0 / 1_000_000, output: 1.0 / 1_000_000 },
-        "sonar-pro": { input: 3.0 / 1_000_000, output: 15.0 / 1_000_000 },
-        "sonar-reasoning": { input: 1.0 / 1_000_000, output: 5.0 / 1_000_000 },
-        "sonar-reasoning-pro": {
-            input: 2.0 / 1_000_000,
-            output: 8.0 / 1_000_000,
-        },
-        "sonar-deep-research": {
-            input: 2.0 / 1_000_000,
-            output: 8.0 / 1_000_000,
-        },
-    },
+    // perplexity — derived from the JSON catalog, see CATALOG_PRICING above.
     cloudflare: {
         // Cloudflare bills per "neuron"; symbolic per-token rate so cost
         // attribution dashboards have non-zero values.
@@ -679,6 +637,20 @@ const PRICING = {
         _default: { input: 0, output: 0.04 / 1_000 },
     },
 };
+/**
+ * Derives the catalog-provider rows of PROVIDER_ALIASES from each entry's
+ * own `id` and `aliases` — the same stripped-lowercase key format the hand
+ * rows below use ("together-ai" -> "togetherai", "workers-ai" -> "workersai").
+ */
+function buildCatalogProviderAliases() {
+    return Object.fromEntries(getCatalogJsonEntries().flatMap((entry) => {
+        const stripped = (s) => s.toLowerCase().replace(/[^a-z]/g, "");
+        return [
+            [stripped(entry.id), entry.id],
+            ...entry.aliases.map((alias) => [stripped(alias), entry.id]),
+        ];
+    }));
+}
 /**
  * Map of normalized provider aliases to canonical PRICING keys.
  * After stripping non-alpha characters, e.g. "google-ai" becomes "googleai".
@@ -691,8 +663,6 @@ const PROVIDER_ALIASES = {
     openai: "openai",
     vertex: "vertex",
     google: "google",
-    mistral: "mistral",
-    mistralai: "mistral",
     azure: "openai",
     azureopenai: "openai",
     bedrock: "anthropic",
@@ -706,20 +676,7 @@ const PROVIDER_ALIASES = {
     nvidia: "nvidia-nim",
     lmstudio: "lm-studio",
     llamacpp: "llamacpp",
-    xai: "xai",
-    grok: "xai",
-    groq: "groq",
-    cerebras: "cerebras",
-    sambanova: "sambanova",
     cohere: "cohere",
-    togetherai: "together-ai",
-    together: "together-ai",
-    fireworks: "fireworks",
-    perplexity: "perplexity",
-    pplx: "perplexity",
-    cloudflare: "cloudflare",
-    workersai: "cloudflare",
-    cfai: "cloudflare",
     replicate: "replicate",
     voyage: "voyage",
     voyageai: "voyage",
@@ -730,6 +687,14 @@ const PROVIDER_ALIASES = {
     sd: "stability",
     ideogram: "ideogram",
     recraft: "recraft",
+    // 9 JSON-catalog providers (cerebras, cloudflare, fireworks, groq,
+    // mistral, perplexity, sambanova, together-ai, xai), derived from each
+    // entry's id/aliases above.
+    ...buildCatalogProviderAliases(),
+    // Not in mistral.json's `aliases` array (only "grok", "together",
+    // "pplx", "workers-ai"/"cf-ai" are real catalog aliases) — a pre-existing
+    // lenient alias kept explicitly since it can't be derived from the catalog.
+    mistralai: "mistral",
 };
 /**
  * Look up per-token rates for a provider/model combination.

package/dist/utils/providerConfig.d.ts

@@ -136,7 +136,7 @@ export declare function createNvidiaNimConfig(): ProviderConfigOptions;
  */
 export declare function createXaiConfig(): ProviderConfigOptions;
 /**
- * Creates Cerebras provider configuration.
+ * Creates SambaNova provider configuration.
  */
 export declare function createSambanovaConfig(): ProviderConfigOptions;
 export declare function createCerebrasConfig(): ProviderConfigOptions;