chore(sync): keep fireworks catalog entries curated

chore(sync): standardize fireworks api key env
chore(sync): add fireworks ai provider
2026-05-21 18:32:00 -05:00 · 2026-05-21 18:25:07 -05:00 · 2026-05-21 18:18:35 -05:00 · 2026-05-21 18:10:09 -05:00 · 2026-05-21 18:06:25 -05:00 · 2026-05-21 17:58:29 -05:00
223 changed files with 3114 additions and 1636 deletions
@@ -35,3 +35,4 @@ jobs:
      - run: bun sst deploy --stage=dev
        env:
          CLOUDFLARE_API_TOKEN: ${{ secrets.CLOUDFLARE_API_TOKEN }}
+          CLOUDFLARE_DEFAULT_ACCOUNT_ID: ${{ secrets.CLOUDFLARE_DEFAULT_ACCOUNT_ID }}
@@ -2,7 +2,7 @@ name: Sync Model Catalogs

 on:
  schedule:
-    - cron: "17 8 * * *"
+    - cron: "17 * * * *"
  workflow_dispatch:

 permissions:
@@ -13,16 +13,34 @@ permissions:
 concurrency: ${{ github.workflow }}-${{ github.ref }}

 jobs:
+  providers:
+    runs-on: ubuntu-latest
+    outputs:
+      matrix: ${{ steps.providers.outputs.matrix }}
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5
+        with:
+          ref: dev
+
+      - name: Setup Bun
+        uses: oven-sh/setup-bun@f4d14e03ff726c06358e5557344e1da148b56cf7
+        with:
+          bun-version: latest
+
+      - name: Install dependencies
+        run: bun install
+
+      - name: List sync providers
+        id: providers
+        run: echo "matrix=$(bun models:sync --list-providers)" >> "$GITHUB_OUTPUT"
+
  sync:
+    needs: providers
    runs-on: ubuntu-latest
    strategy:
      fail-fast: false
-      matrix:
-        include:
-          - group: aggregators
-            title: "chore(sync): update aggregator model catalogs"
-            branch: automation/sync-models-aggregators
-            labels: automation,model-sync,sync-group:aggregators,provider:openrouter
+      matrix: ${{ fromJSON(needs.providers.outputs.matrix) }}

    steps:
      - name: Checkout code
@@ -39,9 +57,16 @@ jobs:
        run: bun install

      - name: Sync model catalogs
-        run: bun models:sync ${{ matrix.group }}
+        run: bun models:sync ${{ matrix.provider }}
        env:
          OPENROUTER_API_KEY: ${{ secrets.OPENROUTER_API_KEY }}
+          GOOGLE_API_KEY: ${{ secrets.GOOGLE_API_KEY }}
+          GEMINI_API_KEY: ${{ secrets.GEMINI_API_KEY }}
+          GOOGLE_GENERATIVE_AI_API_KEY: ${{ secrets.GOOGLE_GENERATIVE_AI_API_KEY }}
+          FIREWORKS_API_KEY: ${{ secrets.FIREWORKS_API_KEY }}
+          XAI_API_KEY: ${{ secrets.XAI_API_KEY }}
+          CLOUDFLARE_WORKERS_AI_SYNC_ACCOUNT_ID: ${{ secrets.CLOUDFLARE_WORKERS_AI_SYNC_ACCOUNT_ID }}
+          CLOUDFLARE_WORKERS_AI_SYNC_API_TOKEN: ${{ secrets.CLOUDFLARE_WORKERS_AI_SYNC_API_TOKEN }}

      - name: Validate models
        run: bun validate
@@ -49,9 +74,9 @@ jobs:
      - name: Create pull request
        env:
          GH_TOKEN: ${{ github.token }}
-          BRANCH: ${{ matrix.branch }}
-          LABELS: ${{ matrix.labels }}
-          TITLE: ${{ matrix.title }}
+          BRANCH: automation/sync-models-${{ matrix.provider }}
+          LABELS: automation,model-sync,provider:${{ matrix.provider }}
+          TITLE: "chore(sync): update ${{ matrix.name }} model catalog"
        run: |
          if [ -z "$(git status --porcelain -- providers)" ]; then
            echo "No model catalog changes found."
@@ -3,6 +3,7 @@
 .idea
 dist
 .DS_Store
+.sync/
 node_modules
 data/tokenspeed-monitor.sqlite
 data/tokenspeed-monitor.sqlite-shm
@@ -17,6 +17,7 @@
  "scripts": {
    "validate": "bun ./packages/core/script/validate.ts",
    "compare:migrations": "bun ./packages/core/script/compare-model-migrations.ts",
+    "cloudflare:sync": "bun ./packages/core/script/sync-models.ts cloudflare-workers-ai",
    "chutes:generate": "bun ./packages/core/script/generate-chutes.ts",
    "databricks:generate": "bun ./packages/core/script/generate-databricks.ts",
    "helicone:generate": "bun ./packages/core/script/generate-helicone.ts",
@@ -1,421 +1,5 @@
 #!/usr/bin/env bun

-import path from "node:path";
-import { mkdir, readdir, rm } from "node:fs/promises";
-import { z } from "zod";
+import { main } from "../src/sync/index.js";

-import { AuthoredModel, AuthoredModelShape } from "../src/schema.js";
-import { google } from "./sync/google.js";
-import { openrouter } from "./sync/openrouter.js";
-
-const ExistingModel = AuthoredModelShape.partial()
-  .extend({
-    extends: z
-      .object({
-        from: z.string(),
-        omit: z.array(z.string()).optional(),
-      })
-      .strict()
-      .optional(),
-  })
-  .strict();
-
-export type ExistingModel = z.infer<typeof ExistingModel>;
-export type SyncedModel = Omit<z.infer<typeof AuthoredModelShape>, "id">;
-
-export interface SyncProvider<SourceModel> {
-  id: string;
-  name: string;
-  modelsDir: string;
-  skipCreates?: boolean;
-  sourceID?(model: SourceModel): string;
-  skippedNotice?(ids: string[]): string[];
-  fetchModels(): Promise<unknown>;
-  parseModels(raw: unknown): SourceModel[];
-  translateModel(
-    model: SourceModel,
-    context: { existing(id: string): ExistingModel | undefined },
-  ): { id: string; model: SyncedModel } | undefined;
-}
-
-export interface SyncResult {
-  id: string;
-  name: string;
-  status: "changed" | "unchanged";
-  created: number;
-  updated: number;
-  deleted: number;
-  unchanged: number;
-  notices: string[];
-  files: Array<{ status: "created" | "updated" | "deleted"; path: string }>;
-}
-
-export const providers: {
-  google: SyncProvider<any>;
-  openrouter: SyncProvider<any>;
-} = {
-  google,
-  openrouter,
-};
-
-export const groups = {
-  aggregators: ["openrouter"],
-  direct: ["google"],
-} as const;
-
-type ProviderID = keyof typeof providers;
-
-interface SyncOptions {
-  dryRun?: boolean;
-  newOnly?: boolean;
-}
-
-export async function syncProviderByID(id: ProviderID, options: SyncOptions = {}) {
-  return syncProvider(providers[id], options);
-}
-
-export async function syncProvider<SourceModel>(
-  provider: SyncProvider<SourceModel>,
-  options: SyncOptions = {},
-): Promise<SyncResult> {
-  console.log(`\nSyncing ${provider.name}...`);
-
-  const existing = await readExisting(provider.modelsDir);
-  const sourceModels = provider.parseModels(await provider.fetchModels());
-  const desired = new Map<string, { model: z.infer<typeof AuthoredModel>; content: string }>();
-  const skippedRemote: string[] = [];
-
-  for (const sourceModel of sourceModels) {
-    const translated = provider.translateModel(sourceModel, {
-      existing(id) {
-        return existing.get(`${id}.toml`)?.toml;
-      },
-    });
-    if (translated === undefined) {
-      if (provider.skipCreates) skippedRemote.push(provider.sourceID?.(sourceModel) ?? "unknown");
-      continue;
-    }
-
-    const relativePath = `${translated.id}.toml`;
-    if (provider.skipCreates && !existing.has(relativePath)) {
-      skippedRemote.push(translated.id);
-      continue;
-    }
-
-    if (desired.has(relativePath)) {
-      throw new Error(`Duplicate synced model path: ${provider.id}/${relativePath}`);
-    }
-
-    const parsed = AuthoredModel.safeParse({
-      id: translated.id,
-      ...translated.model,
-    });
-    if (!parsed.success) {
-      parsed.error.cause = { provider: provider.id, path: relativePath };
-      throw parsed.error;
-    }
-
-    desired.set(relativePath, {
-      model: parsed.data,
-      content: formatToml(parsed.data),
-    });
-  }
-
-  const files: SyncResult["files"] = [];
-  let unchanged = 0;
-
-  for (const [relativePath, file] of desired) {
-    const filePath = path.join(provider.modelsDir, relativePath);
-    const current = existing.get(relativePath);
-
-    if (current === undefined) {
-      files.push({ status: "created", path: filePath });
-      if (options.dryRun) {
-        console.log(`Would create ${relativePath}`);
-      } else {
-        await mkdir(path.dirname(filePath), { recursive: true });
-        await Bun.write(filePath, file.content);
-      }
-      continue;
-    }
-
-    if (!sameModel(relativePath, current.toml, file.model)) {
-      if (options.newOnly) {
-        unchanged++;
-        continue;
-      }
-
-      files.push({ status: "updated", path: filePath });
-      if (options.dryRun) {
-        console.log(`Would update ${relativePath}`);
-      } else {
-        if (current.symlink) await rm(filePath, { force: true });
-        await Bun.write(filePath, file.content);
-      }
-    } else {
-      unchanged++;
-    }
-  }
-
-  for (const relativePath of existing.keys()) {
-    if (desired.has(relativePath)) continue;
-    if (options.newOnly) {
-      console.log(`Skipping removal in new-only mode: ${relativePath}`);
-      unchanged++;
-      continue;
-    }
-
-    const filePath = path.join(provider.modelsDir, relativePath);
-    files.push({ status: "deleted", path: filePath });
-    if (options.dryRun) {
-      console.log(`Would remove ${relativePath}`);
-    } else {
-      await rm(filePath, { force: true });
-    }
-  }
-
-  const result = summarize(provider, files, unchanged, provider.skippedNotice?.(skippedRemote) ?? []);
-  console.log(
-    `${options.dryRun ? "Dry run: " : ""}${result.created} created, ${result.updated} updated, ${result.deleted} removed, ${result.unchanged} unchanged`,
-  );
-  return result;
-}
-
-export async function syncTargets(target: string, options: SyncOptions = {}) {
-  const ids = target in groups
-    ? groups[target as keyof typeof groups]
-    : target in providers
-      ? [target as ProviderID]
-      : undefined;
-
-  if (ids === undefined) {
-    throw new Error(`Unknown sync target: ${target}`);
-  }
-
-  const results: SyncResult[] = [];
-  for (const id of ids) {
-    results.push(await syncProviderByID(id as ProviderID, options));
-  }
-  return results;
-}
-
-async function readExisting(modelsDir: string) {
-  const existing = new Map<string, { text: string; toml: ExistingModel; symlink: boolean }>();
-
-  for (const { file, symlink } of await tomlFiles(modelsDir)) {
-    const text = await Bun.file(path.join(modelsDir, file)).text();
-    const parsed = ExistingModel.safeParse(Bun.TOML.parse(text));
-    if (!parsed.success) {
-      parsed.error.cause = { path: path.join(modelsDir, file) };
-      throw parsed.error;
-    }
-    existing.set(file, { text, toml: parsed.data, symlink });
-  }
-
-  return existing;
-}
-
-async function tomlFiles(root: string, dir = "") {
-  const result: Array<{ file: string; symlink: boolean }> = [];
-
-  for (const entry of await readdir(path.join(root, dir), { withFileTypes: true })) {
-    const file = path.join(dir, entry.name);
-    if (entry.isDirectory()) {
-      result.push(...await tomlFiles(root, file));
-    } else if (entry.name.endsWith(".toml") && (entry.isFile() || entry.isSymbolicLink())) {
-      result.push({ file, symlink: entry.isSymbolicLink() });
-    }
-  }
-
-  return result;
-}
-
-function summarize(
-  provider: { id: string; name: string },
-  files: SyncResult["files"],
-  unchanged: number,
-  notices: string[],
-): SyncResult {
-  return {
-    id: provider.id,
-    name: provider.name,
-    status: files.length > 0 ? "changed" : "unchanged",
-    created: files.filter((file) => file.status === "created").length,
-    updated: files.filter((file) => file.status === "updated").length,
-    deleted: files.filter((file) => file.status === "deleted").length,
-    unchanged,
-    notices,
-    files,
-  };
-}
-
-function sameModel(
-  relativePath: string,
-  current: ExistingModel,
-  desired: z.infer<typeof AuthoredModel>,
-) {
-  const parsed = AuthoredModel.safeParse({
-    id: relativePath.slice(0, -5),
-    ...current,
-  });
-  return parsed.success && stable(parsed.data) === stable(desired);
-}
-
-function stable(value: unknown): string {
-  if (Array.isArray(value)) {
-    const items = value.map(stable);
-    const ordered = value.every((item) => item === null || typeof item !== "object")
-      ? items.sort()
-      : items;
-    return `[${ordered.join(",")}]`;
-  }
-  if (value !== null && typeof value === "object") {
-    return `{${Object.entries(value)
-      .filter(([, item]) => item !== undefined)
-      .sort(([a], [b]) => a.localeCompare(b))
-      .map(([key, item]) => `${JSON.stringify(key)}:${stable(item)}`)
-      .join(",")}}`;
-  }
-  return JSON.stringify(value);
-}
-
-async function writeReport(target: string, results: SyncResult[]) {
-  await mkdir(".sync", { recursive: true });
-
-  const lines = [
-    `Updates model TOMLs for the \`${target}\` sync target.`,
-    "",
-    "| Provider | Status | Created | Updated | Deleted |",
-    "| --- | --- | ---: | ---: | ---: |",
-  ];
-
-  for (const result of results) {
-    lines.push(
-      `| ${result.name} | ${result.status} | ${result.created} | ${result.updated} | ${result.deleted} |`,
-    );
-  }
-
-  for (const result of results.filter((item) => item.files.length > 0)) {
-    lines.push("", `<details><summary>${result.name} changed files</summary>`, "");
-    for (const file of result.files) {
-      lines.push(`- ${file.status}: \`${file.path}\``);
-    }
-    lines.push("", "</details>");
-  }
-
-  const noticeResults = results.filter((item) => item.notices.length > 0);
-  if (noticeResults.length > 0) {
-    lines.push("", "## Notices");
-    for (const result of noticeResults) {
-      lines.push("", `### ${result.name}`);
-      for (const notice of result.notices) {
-        lines.push(`- ${notice}`);
-      }
-    }
-  }
-
-  lines.push("", "This PR was created automatically by the daily model sync workflow.");
-  await Bun.write(".sync/model-sync-report.md", `${lines.join("\n")}\n`);
-}
-
-function quote(value: string) {
-  return `"${value.replaceAll("\\", "\\\\").replaceAll('"', '\\"')}"`;
-}
-
-function formatInteger(n: number) {
-  return String(n).replace(/\B(?=(\d{3})+(?!\d))/g, "_");
-}
-
-function formatNumber(n: number) {
-  return Number.isInteger(n) ? formatInteger(n) : String(n);
-}
-
-function formatToml(model: z.infer<typeof AuthoredModel>) {
-  const lines: string[] = [];
-
-  lines.push(`name = ${quote(model.name)}`);
-  if (model.family !== undefined) lines.push(`family = ${quote(model.family)}`);
-  lines.push(`release_date = ${quote(model.release_date)}`);
-  lines.push(`last_updated = ${quote(model.last_updated)}`);
-  lines.push(`attachment = ${model.attachment}`);
-  lines.push(`reasoning = ${model.reasoning}`);
-  if (model.temperature !== undefined) lines.push(`temperature = ${model.temperature}`);
-  lines.push(`tool_call = ${model.tool_call}`);
-  if (model.structured_output !== undefined) {
-    lines.push(`structured_output = ${model.structured_output}`);
-  }
-  if (model.knowledge !== undefined) lines.push(`knowledge = ${quote(model.knowledge)}`);
-  lines.push(`open_weights = ${model.open_weights}`);
-  if (model.status !== undefined) lines.push(`status = ${quote(model.status)}`);
-
-  if (model.interleaved !== undefined) {
-    lines.push("");
-    if (model.interleaved === true) {
-      lines.push("interleaved = true");
-    } else {
-      lines.push("[interleaved]");
-      lines.push(`field = ${quote(model.interleaved.field)}`);
-    }
-  }
-
-  if (model.cost !== undefined) {
-    lines.push("", "[cost]");
-    lines.push(`input = ${formatNumber(model.cost.input)}`);
-    lines.push(`output = ${formatNumber(model.cost.output)}`);
-    if (model.cost.reasoning !== undefined) {
-      lines.push(`reasoning = ${formatNumber(model.cost.reasoning)}`);
-    }
-    if (model.cost.cache_read !== undefined) {
-      lines.push(`cache_read = ${formatNumber(model.cost.cache_read)}`);
-    }
-    if (model.cost.cache_write !== undefined) {
-      lines.push(`cache_write = ${formatNumber(model.cost.cache_write)}`);
-    }
-    if (model.cost.input_audio !== undefined) {
-      lines.push(`input_audio = ${formatNumber(model.cost.input_audio)}`);
-    }
-    if (model.cost.output_audio !== undefined) {
-      lines.push(`output_audio = ${formatNumber(model.cost.output_audio)}`);
-    }
-
-    for (const tier of model.cost.tiers ?? []) {
-      lines.push("", "[[cost.tiers]]");
-      lines.push(`tier = { size = ${formatInteger(tier.tier.size)} }`);
-      lines.push(`input = ${formatNumber(tier.input)}`);
-      lines.push(`output = ${formatNumber(tier.output)}`);
-      if (tier.reasoning !== undefined) lines.push(`reasoning = ${formatNumber(tier.reasoning)}`);
-      if (tier.cache_read !== undefined) lines.push(`cache_read = ${formatNumber(tier.cache_read)}`);
-      if (tier.cache_write !== undefined) lines.push(`cache_write = ${formatNumber(tier.cache_write)}`);
-    }
-  }
-
-  lines.push("", "[limit]");
-  lines.push(`context = ${formatInteger(model.limit.context)}`);
-  if (model.limit.input !== undefined) lines.push(`input = ${formatInteger(model.limit.input)}`);
-  lines.push(`output = ${formatInteger(model.limit.output)}`);
-
-  lines.push("", "[modalities]");
-  lines.push(`input = [${model.modalities.input.map(quote).join(", ")}]`);
-  lines.push(`output = [${model.modalities.output.map(quote).join(", ")}]`);
-
-  return `${lines.join("\n")}\n`;
-}
-
-export async function main(args = process.argv.slice(2)) {
-  const target = args.find((arg) => !arg.startsWith("-")) ?? "aggregators";
-  const results = await syncTargets(target, {
-    dryRun: args.includes("--dry-run"),
-    newOnly: args.includes("--new-only"),
-  });
-
-  await writeReport(target, results);
-
-  console.log("\nSync summary");
-  for (const result of results) {
-    console.log(
-      `${result.name}: ${result.created} created, ${result.updated} updated, ${result.deleted} deleted`,
-    );
-  }
-}
-
-if (import.meta.main) await main();
+await main();
@@ -1,144 +0,0 @@
-import { z } from "zod";
-
-import { ModelFamilyValues } from "../../src/family.js";
-import type { ExistingModel, SyncProvider } from "../sync-models.js";
-
-const API_ENDPOINT = "https://openrouter.ai/api/v1/models";
-
-const OpenRouterModel = z.object({
-  id: z.string(),
-  name: z.string(),
-  created: z.number(),
-  hugging_face_id: z.string().nullable(),
-  knowledge_cutoff: z.string().nullable(),
-  context_length: z.number(),
-  architecture: z.object({
-    input_modalities: z.array(z.string()),
-    output_modalities: z.array(z.string()),
-  }),
-  pricing: z.object({
-    prompt: z.string(),
-    completion: z.string(),
-    internal_reasoning: z.string().optional(),
-    input_cache_read: z.string().optional(),
-    input_cache_write: z.string().optional(),
-  }),
-  top_provider: z.object({
-    context_length: z.number().nullable(),
-    max_completion_tokens: z.number().nullable(),
-  }),
-  supported_parameters: z.array(z.string()),
-});
-
-const OpenRouterResponse = z.object({
-  data: z.array(OpenRouterModel),
-}).passthrough();
-
-type OpenRouterModel = z.infer<typeof OpenRouterModel>;
-
-export const openrouter = {
-  id: "openrouter",
-  name: "OpenRouter",
-  modelsDir: "providers/openrouter/models",
-  async fetchModels() {
-    const headers = process.env.OPENROUTER_API_KEY
-      ? { Authorization: `Bearer ${process.env.OPENROUTER_API_KEY}` }
-      : undefined;
-    const response = await fetch(API_ENDPOINT, { headers });
-    if (!response.ok) {
-      throw new Error(`OpenRouter request failed: ${response.status} ${response.statusText}`);
-    }
-    return response.json();
-  },
-  parseModels(raw) {
-    return OpenRouterResponse.parse(raw).data;
-  },
-  translateModel(model, context) {
-    return {
-      id: model.id,
-      model: buildModel(model, context.existing(model.id)),
-    };
-  },
-} satisfies SyncProvider<OpenRouterModel>;
-
-function dateFromTimestamp(timestamp: number) {
-  return new Date(timestamp * 1000).toISOString().slice(0, 10);
-}
-
-function price(value: string | undefined) {
-  if (value === undefined) return undefined;
-  const number = Number(value);
-  return Number.isFinite(number) && number >= 0
-    ? Math.round(number * 1_000_000_000_000) / 1_000_000
-    : undefined;
-}
-
-type Modality = "text" | "audio" | "image" | "video" | "pdf";
-
-function modalities(values: string[], fallback: Modality[]): Modality[] {
-  const allowed = new Set<Modality>(["text", "audio", "image", "video", "pdf"]);
-  const result = values
-    .map((value) => value.toLowerCase())
-    .map((value) => value === "file" ? "pdf" : value)
-    .filter((value): value is Modality => allowed.has(value as Modality));
-  return [...new Set(result.length > 0 ? result : fallback)];
-}
-
-function inferFamily(model: OpenRouterModel, name: string) {
-  const target = `${model.id} ${name}`.toLowerCase();
-  return [...ModelFamilyValues]
-    .sort((a, b) => b.length - a.length)
-    .find((family) => {
-      const value = family.toLowerCase().replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
-      if (family === "o") {
-        return new RegExp(`(^|[^a-z0-9])${value}(?=\\d|$|[^a-z0-9])`).test(target);
-      }
-      return new RegExp(`(^|[^a-z0-9])${value}(?=$|[^a-z0-9])`).test(target);
-    });
-}
-
-function buildModel(model: OpenRouterModel, existing: ExistingModel | undefined) {
-  const params = new Set(model.supported_parameters);
-  const name = model.name.replace(/^[^:]+:\s+/, "");
-  const input = modalities(model.architecture.input_modalities, ["text"]);
-  const output = modalities(model.architecture.output_modalities, ["text"]);
-  const prompt = price(model.pricing.prompt);
-  const completion = price(model.pricing.completion);
-  const reasoning = params.has("reasoning") || params.has("include_reasoning");
-  const context = model.top_provider.context_length ?? model.context_length;
-  const family = inferFamily(model, name);
-
-  return {
-    name,
-    family: existing?.family === "o" && family !== "o"
-      ? family
-      : (existing?.family ?? family),
-    release_date: dateFromTimestamp(model.created),
-    last_updated: dateFromTimestamp(model.created),
-    attachment: input.some((value) => value !== "text"),
-    reasoning,
-    temperature: params.has("temperature"),
-    tool_call: params.has("tools") || params.has("tool_choice"),
-    structured_output: params.has("structured_outputs"),
-    knowledge: model.knowledge_cutoff?.slice(0, 10) ?? existing?.knowledge,
-    open_weights: Boolean(model.hugging_face_id),
-    status: existing?.status,
-    interleaved: existing?.interleaved,
-    cost: prompt !== undefined && completion !== undefined
-      ? {
-          input: prompt,
-          output: completion,
-          reasoning: reasoning ? price(model.pricing.internal_reasoning) : undefined,
-          cache_read: price(model.pricing.input_cache_read),
-          cache_write: price(model.pricing.input_cache_write),
-          tiers: existing?.cost?.tiers,
-        }
-      : existing?.cost,
-    limit: {
-      context,
-      input: existing?.limit?.input,
-      output: model.top_provider.max_completion_tokens ?? existing?.limit?.output ?? context,
-    },
-    modalities: { input, output },
-  };
-}
@@ -82,6 +82,7 @@ export const ModelFamilyValues = [

  // xAI Grok
  "grok",
+  "grok-build",
  "grok-vision",
  "grok-beta",

@@ -0,0 +1,489 @@
+import path from "node:path";
+import { mkdir, readdir, rm } from "node:fs/promises";
+import { z } from "zod";
+
+import { AuthoredModel, AuthoredModelShape } from "../schema.js";
+import { cloudflareWorkersAi } from "./providers/cloudflare-workers-ai.js";
+import { fireworksAi } from "./providers/fireworks-ai.js";
+import { google } from "./providers/google.js";
+import { openrouter } from "./providers/openrouter.js";
+import { xai } from "./providers/xai.js";
+
+const ExtendsConfig = z
+  .object({
+    from: z.string(),
+    omit: z.array(z.string()).optional(),
+  })
+  .strict();
+
+const ExistingExtendsConfig = z
+  .object({
+    from: z.string(),
+    omit: z.array(z.string()).optional(),
+  })
+  .passthrough();
+
+const ExistingModel = AuthoredModelShape.partial()
+  .extend({
+    extends: ExistingExtendsConfig.optional(),
+  })
+  .strict();
+
+const SyncedExtendsModel = AuthoredModelShape.partial()
+  .extend({
+    id: z.string(),
+    extends: ExtendsConfig,
+  })
+  .strict();
+
+const SyncedAuthoredModel = z.union([AuthoredModel, SyncedExtendsModel]);
+
+export type ExistingModel = z.infer<typeof ExistingModel>;
+export type SyncedFullModel = Omit<z.infer<typeof AuthoredModelShape>, "id">;
+export type SyncedExtendsModel = Omit<z.infer<typeof SyncedExtendsModel>, "id">;
+export type SyncedModel = SyncedFullModel | SyncedExtendsModel;
+
+export interface SyncProvider<SourceModel> {
+  id: string;
+  name: string;
+  modelsDir: string;
+  skipCreates?: boolean;
+  deleteMissing?: boolean;
+  sourceID?(model: SourceModel): string;
+  skippedNotice?(ids: string[]): string[];
+  fetchModels(): Promise<unknown>;
+  parseModels(raw: unknown): SourceModel[];
+  translateModel(
+    model: SourceModel,
+    context: { existing(id: string): ExistingModel | undefined },
+  ): { id: string; model: SyncedModel } | undefined;
+}
+
+export interface SyncResult {
+  id: string;
+  name: string;
+  status: "changed" | "unchanged";
+  created: number;
+  updated: number;
+  deleted: number;
+  unchanged: number;
+  notices: string[];
+  files: Array<{ status: "created" | "updated" | "deleted"; path: string }>;
+}
+
+export const providers: {
+  "cloudflare-workers-ai": SyncProvider<any>;
+  "fireworks-ai": SyncProvider<any>;
+  google: SyncProvider<any>;
+  openrouter: SyncProvider<any>;
+  xai: SyncProvider<any>;
+} = {
+  "cloudflare-workers-ai": cloudflareWorkersAi,
+  "fireworks-ai": fireworksAi,
+  google,
+  openrouter,
+  xai,
+};
+
+export const groups = {
+  aggregators: ["openrouter"],
+  cloudflare: ["cloudflare-workers-ai"],
+  direct: ["fireworks-ai", "google", "xai"],
+} as const;
+
+type ProviderID = keyof typeof providers;
+
+interface SyncOptions {
+  dryRun?: boolean;
+  newOnly?: boolean;
+}
+
+export async function syncProviderByID(id: ProviderID, options: SyncOptions = {}) {
+  return syncProvider(providers[id], options);
+}
+
+export async function syncProvider<SourceModel>(
+  provider: SyncProvider<SourceModel>,
+  options: SyncOptions = {},
+): Promise<SyncResult> {
+  console.log(`\nSyncing ${provider.name}...`);
+
+  const existing = await readExisting(provider.modelsDir);
+  const sourceModels = provider.parseModels(await provider.fetchModels());
+  const desired = new Map<string, { model: z.infer<typeof SyncedAuthoredModel>; content: string }>();
+  const skippedRemote: string[] = [];
+
+  for (const sourceModel of sourceModels) {
+    const translated = provider.translateModel(sourceModel, {
+      existing(id) {
+        return existing.get(`${id}.toml`)?.toml;
+      },
+    });
+    if (translated === undefined) {
+      if (provider.skipCreates) skippedRemote.push(provider.sourceID?.(sourceModel) ?? "unknown");
+      continue;
+    }
+
+    const relativePath = `${translated.id}.toml`;
+    if (provider.skipCreates && !existing.has(relativePath)) {
+      skippedRemote.push(translated.id);
+      continue;
+    }
+
+    if (desired.has(relativePath)) {
+      throw new Error(`Duplicate synced model path: ${provider.id}/${relativePath}`);
+    }
+
+    const parsed = SyncedAuthoredModel.safeParse({
+      id: translated.id,
+      ...translated.model,
+    });
+    if (!parsed.success) {
+      parsed.error.cause = { provider: provider.id, path: relativePath };
+      throw parsed.error;
+    }
+
+    desired.set(relativePath, {
+      model: parsed.data,
+      content: formatToml(parsed.data),
+    });
+  }
+
+  const files: SyncResult["files"] = [];
+  let unchanged = 0;
+
+  for (const [relativePath, file] of desired) {
+    const filePath = path.join(provider.modelsDir, relativePath);
+    const current = existing.get(relativePath);
+
+    if (current === undefined) {
+      files.push({ status: "created", path: filePath });
+      if (options.dryRun) {
+        console.log(`Would create ${relativePath}`);
+      } else {
+        await mkdir(path.dirname(filePath), { recursive: true });
+        await Bun.write(filePath, file.content);
+      }
+      continue;
+    }
+
+    if (!sameModel(relativePath, current.toml, file.model)) {
+      if (options.newOnly) {
+        unchanged++;
+        continue;
+      }
+
+      files.push({ status: "updated", path: filePath });
+      if (options.dryRun) {
+        console.log(`Would update ${relativePath}`);
+      } else {
+        if (current.symlink) await rm(filePath, { force: true });
+        await Bun.write(filePath, file.content);
+      }
+    } else {
+      unchanged++;
+    }
+  }
+
+  for (const relativePath of existing.keys()) {
+    if (desired.has(relativePath)) continue;
+    if (provider.deleteMissing === false) {
+      unchanged++;
+      continue;
+    }
+    if (options.newOnly) {
+      console.log(`Skipping removal in new-only mode: ${relativePath}`);
+      unchanged++;
+      continue;
+    }
+
+    const filePath = path.join(provider.modelsDir, relativePath);
+    files.push({ status: "deleted", path: filePath });
+    if (options.dryRun) {
+      console.log(`Would remove ${relativePath}`);
+    } else {
+      await rm(filePath, { force: true });
+    }
+  }
+
+  const result = summarize(provider, files, unchanged, provider.skippedNotice?.(skippedRemote) ?? []);
+  console.log(
+    `${options.dryRun ? "Dry run: " : ""}${result.created} created, ${result.updated} updated, ${result.deleted} removed, ${result.unchanged} unchanged`,
+  );
+  return result;
+}
+
+export async function syncTargets(target: string, options: SyncOptions = {}) {
+  const ids = target in groups
+    ? groups[target as keyof typeof groups]
+    : target in providers
+      ? [target as ProviderID]
+      : undefined;
+
+  if (ids === undefined) {
+    throw new Error(`Unknown sync target: ${target}`);
+  }
+
+  const results: SyncResult[] = [];
+  for (const id of ids) {
+    results.push(await syncProviderByID(id as ProviderID, options));
+  }
+  return results;
+}
+
+export function syncProviderMatrix() {
+  return {
+    include: Object.values(providers).map((provider) => ({
+      provider: provider.id,
+      name: provider.name,
+    })),
+  };
+}
+
+async function readExisting(modelsDir: string) {
+  const existing = new Map<string, { text: string; toml: ExistingModel; symlink: boolean }>();
+
+  for (const { file, symlink } of await tomlFiles(modelsDir)) {
+    const text = await Bun.file(path.join(modelsDir, file)).text();
+    const parsed = ExistingModel.safeParse(Bun.TOML.parse(text));
+    if (!parsed.success) {
+      parsed.error.cause = { path: path.join(modelsDir, file) };
+      throw parsed.error;
+    }
+    existing.set(file, { text, toml: parsed.data, symlink });
+  }
+
+  return existing;
+}
+
+async function tomlFiles(root: string, dir = "") {
+  const result: Array<{ file: string; symlink: boolean }> = [];
+
+  for (const entry of await readdir(path.join(root, dir), { withFileTypes: true })) {
+    const file = path.join(dir, entry.name);
+    if (entry.isDirectory()) {
+      result.push(...await tomlFiles(root, file));
+    } else if (entry.name.endsWith(".toml") && (entry.isFile() || entry.isSymbolicLink())) {
+      result.push({ file, symlink: entry.isSymbolicLink() });
+    }
+  }
+
+  return result;
+}
+
+function summarize(
+  provider: { id: string; name: string },
+  files: SyncResult["files"],
+  unchanged: number,
+  notices: string[],
+): SyncResult {
+  return {
+    id: provider.id,
+    name: provider.name,
+    status: files.length > 0 ? "changed" : "unchanged",
+    created: files.filter((file) => file.status === "created").length,
+    updated: files.filter((file) => file.status === "updated").length,
+    deleted: files.filter((file) => file.status === "deleted").length,
+    unchanged,
+    notices,
+    files,
+  };
+}
+
+function sameModel(
+  relativePath: string,
+  current: ExistingModel,
+  desired: z.infer<typeof SyncedAuthoredModel>,
+) {
+  const parsed = SyncedAuthoredModel.safeParse({
+    id: relativePath.slice(0, -5),
+    ...current,
+  });
+  return parsed.success && stable(parsed.data) === stable(desired);
+}
+
+function stable(value: unknown): string {
+  if (Array.isArray(value)) {
+    const items = value.map(stable);
+    const ordered = value.every((item) => item === null || typeof item !== "object")
+      ? items.sort()
+      : items;
+    return `[${ordered.join(",")}]`;
+  }
+  if (value !== null && typeof value === "object") {
+    return `{${Object.entries(value)
+      .filter(([, item]) => item !== undefined)
+      .sort(([a], [b]) => a.localeCompare(b))
+      .map(([key, item]) => `${JSON.stringify(key)}:${stable(item)}`)
+      .join(",")}}`;
+  }
+  return JSON.stringify(value);
+}
+
+async function writeReport(target: string, results: SyncResult[]) {
+  await mkdir(".sync", { recursive: true });
+
+  const lines = [
+    `Updates model TOMLs for the \`${target}\` sync target.`,
+    "",
+    "| Provider | Status | Created | Updated | Deleted |",
+    "| --- | --- | ---: | ---: | ---: |",
+  ];
+
+  for (const result of results) {
+    lines.push(
+      `| ${result.name} | ${result.status} | ${result.created} | ${result.updated} | ${result.deleted} |`,
+    );
+  }
+
+  for (const result of results.filter((item) => item.files.length > 0)) {
+    lines.push("", `<details><summary>${result.name} changed files</summary>`, "");
+    for (const file of result.files) {
+      lines.push(`- ${file.status}: \`${file.path}\``);
+    }
+    lines.push("", "</details>");
+  }
+
+  const noticeResults = results.filter((item) => item.notices.length > 0);
+  if (noticeResults.length > 0) {
+    lines.push("", "## Notices");
+    for (const result of noticeResults) {
+      lines.push("", `### ${result.name}`);
+      for (const notice of result.notices) {
+        lines.push(`- ${notice}`);
+      }
+    }
+  }
+
+  lines.push("", "This PR was created automatically by the daily model sync workflow.");
+  await Bun.write(".sync/model-sync-report.md", `${lines.join("\n")}\n`);
+}
+
+function quote(value: string) {
+  return `"${value.replaceAll("\\", "\\\\").replaceAll('"', '\\"')}"`;
+}
+
+function formatInteger(n: number) {
+  return String(n).replace(/\B(?=(\d{3})+(?!\d))/g, "_");
+}
+
+function formatNumber(n: number) {
+  return Number.isInteger(n) ? formatInteger(n) : String(n);
+}
+
+function formatToml(model: z.infer<typeof SyncedAuthoredModel>) {
+  const lines: string[] = [];
+  const extendsLines: string[] = [];
+
+  if ("extends" in model) {
+    extendsLines.push("[extends]");
+    extendsLines.push(`from = ${quote(model.extends.from)}`);
+    if (model.extends.omit !== undefined) {
+      extendsLines.push(`omit = [${model.extends.omit.map(quote).join(", ")}]`);
+    }
+  }
+
+  if (model.name !== undefined) lines.push(`name = ${quote(model.name)}`);
+  if (model.family !== undefined) lines.push(`family = ${quote(model.family)}`);
+  if (model.release_date !== undefined) lines.push(`release_date = ${quote(model.release_date)}`);
+  if (model.last_updated !== undefined) lines.push(`last_updated = ${quote(model.last_updated)}`);
+  if (model.attachment !== undefined) lines.push(`attachment = ${model.attachment}`);
+  if (model.reasoning !== undefined) lines.push(`reasoning = ${model.reasoning}`);
+  if (model.temperature !== undefined) lines.push(`temperature = ${model.temperature}`);
+  if (model.tool_call !== undefined) lines.push(`tool_call = ${model.tool_call}`);
+  if (model.structured_output !== undefined) {
+    lines.push(`structured_output = ${model.structured_output}`);
+  }
+  if (model.knowledge !== undefined) lines.push(`knowledge = ${quote(model.knowledge)}`);
+  if (model.open_weights !== undefined) lines.push(`open_weights = ${model.open_weights}`);
+  if (model.status !== undefined) lines.push(`status = ${quote(model.status)}`);
+
+  if (extendsLines.length > 0) {
+    if (lines.length > 0) lines.push("");
+    lines.push(...extendsLines);
+  }
+
+  if (model.interleaved !== undefined) {
+    lines.push("");
+    if (model.interleaved === true) {
+      lines.push("interleaved = true");
+    } else {
+      lines.push("[interleaved]");
+      lines.push(`field = ${quote(model.interleaved.field)}`);
+    }
+  }
+
+  if (model.cost !== undefined) {
+    lines.push("", "[cost]");
+    lines.push(`input = ${formatNumber(model.cost.input)}`);
+    lines.push(`output = ${formatNumber(model.cost.output)}`);
+    if (model.cost.reasoning !== undefined) {
+      lines.push(`reasoning = ${formatNumber(model.cost.reasoning)}`);
+    }
+    if (model.cost.cache_read !== undefined) {
+      lines.push(`cache_read = ${formatNumber(model.cost.cache_read)}`);
+    }
+    if (model.cost.cache_write !== undefined) {
+      lines.push(`cache_write = ${formatNumber(model.cost.cache_write)}`);
+    }
+    if (model.cost.input_audio !== undefined) {
+      lines.push(`input_audio = ${formatNumber(model.cost.input_audio)}`);
+    }
+    if (model.cost.output_audio !== undefined) {
+      lines.push(`output_audio = ${formatNumber(model.cost.output_audio)}`);
+    }
+
+    for (const tier of model.cost.tiers ?? []) {
+      lines.push("", "[[cost.tiers]]");
+      lines.push(`tier = { size = ${formatInteger(tier.tier.size)} }`);
+      lines.push(`input = ${formatNumber(tier.input)}`);
+      lines.push(`output = ${formatNumber(tier.output)}`);
+      if (tier.reasoning !== undefined) lines.push(`reasoning = ${formatNumber(tier.reasoning)}`);
+      if (tier.cache_read !== undefined) lines.push(`cache_read = ${formatNumber(tier.cache_read)}`);
+      if (tier.cache_write !== undefined) lines.push(`cache_write = ${formatNumber(tier.cache_write)}`);
+    }
+  }
+
+  if (model.limit !== undefined) {
+    lines.push("", "[limit]");
+    if (model.limit.context !== undefined) lines.push(`context = ${formatInteger(model.limit.context)}`);
+    if (model.limit.input !== undefined) lines.push(`input = ${formatInteger(model.limit.input)}`);
+    if (model.limit.output !== undefined) lines.push(`output = ${formatInteger(model.limit.output)}`);
+  }
+
+  if (model.modalities !== undefined) {
+    lines.push("", "[modalities]");
+    if (model.modalities.input !== undefined) {
+      lines.push(`input = [${model.modalities.input.map(quote).join(", ")}]`);
+    }
+    if (model.modalities.output !== undefined) {
+      lines.push(`output = [${model.modalities.output.map(quote).join(", ")}]`);
+    }
+  }
+
+  return `${lines.join("\n")}\n`;
+}
+
+export async function main(args = process.argv.slice(2)) {
+  if (args.includes("--list-providers")) {
+    console.log(JSON.stringify(syncProviderMatrix()));
+    return;
+  }
+
+  const target = args.find((arg) => !arg.startsWith("-")) ?? "aggregators";
+  const results = await syncTargets(target, {
+    dryRun: args.includes("--dry-run"),
+    newOnly: args.includes("--new-only"),
+  });
+
+  await writeReport(target, results);
+
+  console.log("\nSync summary");
+  for (const result of results) {
+    console.log(
+      `${result.name}: ${result.created} created, ${result.updated} updated, ${result.deleted} deleted`,
+    );
+  }
+}
+
+if (import.meta.main) await main();
@@ -0,0 +1,176 @@
+import { z } from "zod";
+
+import type { ExistingModel, SyncProvider } from "../index.js";
+import {
+  buildOpenRouterModel,
+  OpenRouterModel,
+  OpenRouterResponse,
+} from "./openrouter.js";
+
+const API_BASE = "https://api.cloudflare.com/client/v4/accounts";
+
+const CloudflareOpenRouterResponse = z.object({
+  result: z.union([OpenRouterResponse, z.array(OpenRouterModel)]).optional(),
+  result_info: z.object({
+    page: z.number().optional(),
+    total_pages: z.number().optional(),
+  }).passthrough().optional(),
+}).passthrough();
+
+const CloudflareModel = z.object({
+  id: z.string(),
+  name: z.string(),
+  created: z.number(),
+  hugging_face_id: z.string().nullable().optional(),
+  context_length: z.number(),
+  max_output_length: z.number().nullable().optional(),
+  input_modalities: z.array(z.string()).optional(),
+  output_modalities: z.array(z.string()).optional(),
+  pricing: z.object({
+    prompt: z.string(),
+    completion: z.string(),
+    internal_reasoning: z.string().optional(),
+    input_cache_read: z.string().optional(),
+    input_cache_write: z.string().optional(),
+  }),
+  supported_features: z.array(z.string()).optional(),
+  supported_sampling_parameters: z.array(z.string()).optional(),
+}).passthrough();
+
+const CloudflareResponse = z.object({
+  data: z.array(CloudflareModel),
+}).passthrough();
+
+type CloudflareModel = z.infer<typeof CloudflareModel>;
+
+export const cloudflareWorkersAi = {
+  id: "cloudflare-workers-ai",
+  name: "Cloudflare Workers AI",
+  modelsDir: "providers/cloudflare-workers-ai/models",
+  async fetchModels() {
+    const accountID = process.env.CLOUDFLARE_WORKERS_AI_SYNC_ACCOUNT_ID;
+    const token = process.env.CLOUDFLARE_WORKERS_AI_SYNC_API_TOKEN;
+    if (accountID === undefined || token === undefined) {
+      throw new Error(
+        "Cloudflare Workers AI sync requires CLOUDFLARE_WORKERS_AI_SYNC_ACCOUNT_ID and CLOUDFLARE_WORKERS_AI_SYNC_API_TOKEN",
+      );
+    }
+
+    const first = await fetchPage(accountID, token, 1);
+    const models = parseCloudflareModels(first);
+    const pageInfo = CloudflareOpenRouterResponse.safeParse(first).success
+      ? CloudflareOpenRouterResponse.parse(first).result_info
+      : undefined;
+
+    for (let page = 2; page <= (pageInfo?.total_pages ?? 1); page++) {
+      models.push(...parseCloudflareModels(await fetchPage(accountID, token, page)));
+    }
+
+    return { data: models };
+  },
+  parseModels(raw) {
+    return parseCloudflareModels(raw);
+  },
+  translateModel(model, context) {
+    const normalized = normalizeModel(model);
+    const id = normalized.id.replace(/^workers-ai\//, "");
+    return {
+      id,
+      model: buildWorkersAiModel(normalized, context.existing(id)),
+    };
+  },
+} satisfies SyncProvider<CloudflareModel>;
+
+function buildWorkersAiModel(model: z.infer<typeof OpenRouterModel>, existing: ExistingModel | undefined) {
+  const synced = buildOpenRouterModel(model, existing);
+  return {
+    ...synced,
+    name: existing?.name ?? synced.name,
+    release_date: existing?.release_date ?? synced.release_date,
+    last_updated: existing?.last_updated ?? synced.last_updated,
+    limit: {
+      ...synced.limit,
+      output: existing?.limit?.output ?? synced.limit.output,
+    },
+  };
+}
+
+async function fetchPage(accountID: string, token: string, page: number) {
+  const url = new URL(`${API_BASE}/${accountID}/ai/models/search`);
+  url.searchParams.set("format", "openrouter");
+  url.searchParams.set("per_page", "1000");
+  url.searchParams.set("page", String(page));
+
+  const response = await fetch(url, {
+    headers: { Authorization: `Bearer ${token}` },
+  });
+  if (!response.ok) {
+    throw new Error(
+      `Cloudflare Workers AI models request failed: ${response.status} ${response.statusText}${await responseDetails(response)}`,
+    );
+  }
+  return response.json();
+}
+
+function parseCloudflareModels(raw: unknown) {
+  const cloudflare = CloudflareResponse.safeParse(raw);
+  if (cloudflare.success) return cloudflare.data.data;
+
+  const direct = OpenRouterResponse.safeParse(raw);
+  if (direct.success) return direct.data.data;
+
+  const wrapped = CloudflareOpenRouterResponse.parse(raw);
+  if (wrapped.result === undefined) {
+    throw new Error("Cloudflare Workers AI response did not include model data");
+  }
+  return Array.isArray(wrapped.result) ? wrapped.result : wrapped.result.data;
+}
+
+function normalizeModel(model: CloudflareModel) {
+  if ("architecture" in model && "top_provider" in model && "supported_parameters" in model) {
+    return OpenRouterModel.parse(model);
+  }
+
+  return OpenRouterModel.parse({
+    id: model.id.startsWith("@cf/") ? model.id : `@cf/${model.id.replace(/^@cf\//, "")}`,
+    name: model.name,
+    created: model.created,
+    hugging_face_id: model.hugging_face_id ?? null,
+    knowledge_cutoff: null,
+    context_length: model.context_length,
+    architecture: {
+      input_modalities: model.input_modalities ?? ["text"],
+      output_modalities: model.output_modalities ?? ["text"],
+    },
+    pricing: model.pricing,
+    top_provider: {
+      context_length: model.context_length,
+      max_completion_tokens: model.max_output_length ?? null,
+    },
+    supported_parameters: [
+      ...model.supported_sampling_parameters ?? [],
+      ...model.supported_features ?? [],
+    ],
+  });
+}
+
+async function responseDetails(response: Response) {
+  const text = await response.text();
+  if (text.length === 0) return "";
+
+  try {
+    const body = z.object({
+      errors: z.array(z.object({
+        code: z.union([z.string(), z.number()]).optional(),
+        message: z.string().optional(),
+      }).passthrough()).optional(),
+    }).passthrough().parse(JSON.parse(text));
+    const details = body.errors
+      ?.map((error) => [error.code, error.message].filter(Boolean).join(": "))
+      .filter((message) => message.length > 0)
+      .join("; ");
+    return details === undefined || details.length === 0 ? "" : ` (${details})`;
+  } catch {
+    return "";
+  }
+}
@@ -0,0 +1,75 @@
+import { z } from "zod";
+
+import type { SyncProvider } from "../index.js";
+
+const API_BASE = "https://api.fireworks.ai/v1";
+
+const FireworksModel = z.object({
+  name: z.string(),
+}).passthrough();
+
+const FireworksResponse = z.object({
+  models: z.array(FireworksModel).optional(),
+  nextPageToken: z.string().optional(),
+}).passthrough();
+
+type FireworksModel = z.infer<typeof FireworksModel>;
+
+export const fireworksAi = {
+  id: "fireworks-ai",
+  name: "Fireworks AI",
+  modelsDir: "providers/fireworks-ai/models",
+  skipCreates: true,
+  deleteMissing: false,
+  sourceID(model) {
+    return model.name;
+  },
+  skippedNotice(ids) {
+    if (ids.length === 0) return [];
+    return [
+      `${ids.length} Fireworks AI models returned by the API were not created because the Models API does not provide pricing, output token limits, release dates, or complete capability metadata. Existing models are preserved unchanged.`,
+      `Skipped remote IDs: ${ids.map((id) => `\`${id}\``).join(", ")}`,
+    ];
+  },
+  async fetchModels() {
+    const key = process.env.FIREWORKS_API_KEY;
+    if (key === undefined) {
+      throw new Error("Fireworks AI sync requires FIREWORKS_API_KEY");
+    }
+
+    const models: FireworksModel[] = [];
+    let pageToken: string | undefined;
+
+    do {
+      const url = new URL(`${API_BASE}/accounts/fireworks/models`);
+      url.searchParams.set("pageSize", "200");
+      url.searchParams.set("filter", "supports_serverless = true");
+      if (pageToken !== undefined) url.searchParams.set("pageToken", pageToken);
+
+      const response = await fetch(url, {
+        headers: { Authorization: `Bearer ${key}` },
+      });
+      if (!response.ok) {
+        throw new Error(`Fireworks AI models request failed: ${response.status} ${response.statusText}`);
+      }
+
+      const page = FireworksResponse.parse(await response.json());
+      models.push(...page.models ?? []);
+      pageToken = page.nextPageToken || undefined;
+    } while (pageToken !== undefined);
+
+    return { models };
+  },
+  parseModels(raw) {
+    return FireworksResponse.parse(raw).models ?? [];
+  },
+  translateModel(model, context) {
+    const existing = context.existing(model.name);
+    if (existing === undefined) return undefined;
+
+    return {
+      id: model.name,
+      model: existing,
+    };
+  },
+} satisfies SyncProvider<FireworksModel>;
@@ -1,6 +1,6 @@
 import { z } from "zod";

-import type { ExistingModel, SyncProvider, SyncedModel } from "../sync-models.js";
+import type { ExistingModel, SyncProvider, SyncedModel } from "../index.js";

 const API_ENDPOINT = "https://generativelanguage.googleapis.com/v1beta/models";

@@ -0,0 +1,334 @@
+import { z } from "zod";
+import { readFileSync, readdirSync } from "node:fs";
+import path from "node:path";
+
+import { ModelFamilyValues } from "../../family.js";
+import type { ExistingModel, SyncProvider, SyncedFullModel, SyncedModel } from "../index.js";
+
+const API_ENDPOINT = "https://openrouter.ai/api/v1/models";
+const PROVIDERS_DIR = path.join(import.meta.dirname, "..", "..", "..", "..", "..", "providers");
+const modelFilesByProvider = new Map<string, Set<string>>();
+const canonicalTomlByModel = new Map<string, Record<string, unknown>>();
+
+const CANONICAL_PROVIDER_PREFIXES = {
+  anthropic: "anthropic",
+  cohere: "cohere",
+  deepseek: "deepseek",
+  google: "google",
+  meta: "llama",
+  "meta-llama": "llama",
+  minimax: "minimax",
+  mistralai: "mistral",
+  moonshotai: "moonshotai",
+  openai: "openai",
+  "x-ai": "xai",
+  xai: "xai",
+  xiaomi: "xiaomi",
+  zai: "zai",
+  "z-ai": "zai",
+} as const;
+
+export const OpenRouterModel = z.object({
+  id: z.string(),
+  name: z.string(),
+  created: z.number(),
+  hugging_face_id: z.string().nullable(),
+  knowledge_cutoff: z.string().nullable(),
+  context_length: z.number(),
+  architecture: z.object({
+    input_modalities: z.array(z.string()),
+    output_modalities: z.array(z.string()),
+  }),
+  pricing: z.object({
+    prompt: z.string(),
+    completion: z.string(),
+    internal_reasoning: z.string().optional(),
+    input_cache_read: z.string().optional(),
+    input_cache_write: z.string().optional(),
+  }),
+  top_provider: z.object({
+    context_length: z.number().nullable(),
+    max_completion_tokens: z.number().nullable(),
+  }),
+  supported_parameters: z.array(z.string()),
+});
+
+export const OpenRouterResponse = z.object({
+  data: z.array(OpenRouterModel),
+}).passthrough();
+
+export type OpenRouterModel = z.infer<typeof OpenRouterModel>;
+
+export const openrouter = {
+  id: "openrouter",
+  name: "OpenRouter",
+  modelsDir: "providers/openrouter/models",
+  async fetchModels() {
+    const headers = process.env.OPENROUTER_API_KEY
+      ? { Authorization: `Bearer ${process.env.OPENROUTER_API_KEY}` }
+      : undefined;
+    const response = await fetch(API_ENDPOINT, { headers });
+    if (!response.ok) {
+      throw new Error(`OpenRouter request failed: ${response.status} ${response.statusText}`);
+    }
+    return response.json();
+  },
+  parseModels(raw) {
+    return OpenRouterResponse.parse(raw).data;
+  },
+  translateModel(model, context) {
+    return {
+      id: model.id,
+      model: buildOpenRouterModel(model, context.existing(model.id)),
+    };
+  },
+} satisfies SyncProvider<OpenRouterModel>;
+
+function dateFromTimestamp(timestamp: number) {
+  return new Date(timestamp * 1000).toISOString().slice(0, 10);
+}
+
+function price(value: string | undefined) {
+  if (value === undefined) return undefined;
+  const number = Number(value);
+  return Number.isFinite(number) && number >= 0
+    ? Math.round(number * 1_000_000_000_000) / 1_000_000
+    : undefined;
+}
+
+type Modality = "text" | "audio" | "image" | "video" | "pdf";
+
+function modalities(values: string[], fallback: Modality[]): Modality[] {
+  const allowed = new Set<Modality>(["text", "audio", "image", "video", "pdf"]);
+  const result = values
+    .map((value) => value.toLowerCase())
+    .map((value) => value === "file" ? "pdf" : value)
+    .filter((value): value is Modality => allowed.has(value as Modality));
+  return [...new Set(result.length > 0 ? result : fallback)];
+}
+
+function inferFamily(model: OpenRouterModel, name: string) {
+  const target = `${model.id} ${name}`.toLowerCase();
+  return [...ModelFamilyValues]
+    .sort((a, b) => b.length - a.length)
+    .find((family) => {
+      const value = family.toLowerCase().replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
+      if (family === "o") {
+        return new RegExp(`(^|[^a-z0-9])${value}(?=\\d|$|[^a-z0-9])`).test(target);
+      }
+      return new RegExp(`(^|[^a-z0-9])${value}(?=$|[^a-z0-9])`).test(target);
+    });
+}
+
+export function buildOpenRouterModel(model: OpenRouterModel, existing: ExistingModel | undefined): SyncedModel {
+  const params = new Set(model.supported_parameters);
+  const name = model.name.replace(/^[^:]+:\s+/, "");
+  const input = modalities(model.architecture.input_modalities, ["text"]);
+  const output = modalities(model.architecture.output_modalities, ["text"]);
+  const prompt = price(model.pricing.prompt);
+  const completion = price(model.pricing.completion);
+  const reasoning = params.has("reasoning") || params.has("include_reasoning");
+  const context = model.top_provider.context_length ?? model.context_length;
+  const family = inferFamily(model, name);
+  const releaseDate = dateFromTimestamp(model.created);
+  const familyValue = existing?.family === "o" && family !== "o"
+    ? family
+    : (existing?.family ?? family);
+  const attachment = input.some((value) => value !== "text");
+  const toolCall = params.has("tools") || params.has("tool_choice");
+  const structuredOutput = params.has("structured_outputs");
+  const knowledge = model.knowledge_cutoff?.slice(0, 10) ?? existing?.knowledge;
+  const openWeights = Boolean(model.hugging_face_id);
+  const cost = prompt !== undefined && completion !== undefined
+    ? {
+        input: prompt,
+        output: completion,
+        reasoning: reasoning ? price(model.pricing.internal_reasoning) : undefined,
+        cache_read: price(model.pricing.input_cache_read),
+        cache_write: price(model.pricing.input_cache_write),
+        tiers: existing?.cost?.tiers,
+      }
+    : existing?.cost;
+  const limit = {
+    context,
+    input: existing?.limit?.input,
+    output: model.top_provider.max_completion_tokens ?? existing?.limit?.output ?? context,
+  };
+  const canonical = resolveCanonicalModel(model.id);
+
+  if (canonical !== undefined) {
+    return {
+      extends: {
+        from: canonical.from,
+        omit: canonicalOmit(canonical.provider, canonical.modelID, cost, limit),
+      },
+      ...canonicalRuntimeOverrides(canonical.provider, canonical.modelID, {
+        name: model.id.endsWith(":free") ? name : undefined,
+        attachment,
+        reasoning,
+      }),
+      temperature: params.has("temperature"),
+      tool_call: toolCall,
+      structured_output: structuredOutput,
+      status: existing?.status,
+      interleaved: existing?.interleaved,
+      cost,
+      limit,
+      modalities: { input, output },
+    };
+  }
+
+  return {
+    name,
+    family: familyValue,
+    release_date: releaseDate,
+    last_updated: releaseDate,
+    attachment,
+    reasoning,
+    temperature: params.has("temperature"),
+    tool_call: toolCall,
+    structured_output: structuredOutput,
+    knowledge,
+    open_weights: openWeights,
+    status: existing?.status,
+    interleaved: existing?.interleaved,
+    cost,
+    limit,
+    modalities: { input, output },
+  } satisfies SyncedFullModel;
+}
+
+function resolveCanonicalModel(openrouterID: string) {
+  const [prefix, ...modelParts] = openrouterID.split("/");
+  if (prefix === undefined || modelParts.length === 0) return undefined;
+  if (openrouterID.startsWith("~/") || prefix.startsWith("~")) return undefined;
+
+  const provider = CANONICAL_PROVIDER_PREFIXES[prefix as keyof typeof CANONICAL_PROVIDER_PREFIXES];
+  if (provider === undefined) return undefined;
+
+  const modelID = modelParts.join("/").replace(/:free$/, "");
+  const candidates = canonicalCandidates(provider, modelID);
+  const match = candidates.find((candidate) => {
+    return canonicalModelExists(provider, candidate);
+  });
+
+  return match === undefined
+    ? undefined
+    : {
+        from: `${provider}/${match}`,
+        provider,
+        modelID: match,
+      };
+}
+
+function canonicalModelExists(provider: string, modelID: string) {
+  let files = modelFilesByProvider.get(provider);
+  if (files === undefined) {
+    try {
+      files = new Set(readdirSync(path.join(PROVIDERS_DIR, provider, "models")));
+    } catch {
+      files = new Set();
+    }
+    modelFilesByProvider.set(provider, files);
+  }
+  return files.has(`${modelID}.toml`);
+}
+
+function canonicalOmit(
+  provider: string,
+  modelID: string,
+  cost: SyncedFullModel["cost"],
+  limit: SyncedFullModel["limit"],
+) {
+  const toml = canonicalToml(provider, modelID);
+  const omit = ["provider", "experimental"].filter((key) => toml[key] !== undefined);
+
+  const baseCost = toml.cost;
+  if (baseCost !== undefined && baseCost !== null && typeof baseCost === "object" && !Array.isArray(baseCost)) {
+    if (cost === undefined) {
+      omit.push("cost");
+    } else {
+      for (const key of ["reasoning", "cache_read", "cache_write", "input_audio", "output_audio", "tiers"] as const) {
+        if ((baseCost as Record<string, unknown>)[key] !== undefined && cost[key] === undefined) {
+          omit.push(`cost.${key}`);
+        }
+      }
+      if (hasLegacyContextOver200k(baseCost) && cost.tiers === undefined) {
+        omit.push("cost.context_over_200k");
+      }
+    }
+  }
+
+  const baseLimit = toml.limit;
+  if (
+    baseLimit !== undefined &&
+    baseLimit !== null &&
+    typeof baseLimit === "object" &&
+    !Array.isArray(baseLimit) &&
+    (baseLimit as Record<string, unknown>).input !== undefined &&
+    limit.input === undefined
+  ) {
+    omit.push("limit.input");
+  }
+
+  return omit.length > 0 ? omit : undefined;
+}
+
+function hasLegacyContextOver200k(cost: object) {
+  const tiers = (cost as { tiers?: unknown }).tiers;
+  if (!Array.isArray(tiers) || tiers.length !== 1) return false;
+
+  const tier = tiers[0];
+  if (tier === null || typeof tier !== "object" || Array.isArray(tier)) return false;
+  const tierConfig = (tier as { tier?: unknown }).tier;
+  if (tierConfig === null || typeof tierConfig !== "object" || Array.isArray(tierConfig)) return false;
+
+  const size = (tierConfig as { size?: unknown }).size;
+  return typeof size === "number" && size >= 200_000;
+}
+
+function canonicalRuntimeOverrides(
+  provider: string,
+  modelID: string,
+  values: Pick<SyncedFullModel, "name" | "attachment" | "reasoning">,
+) {
+  const toml = canonicalToml(provider, modelID);
+  return Object.fromEntries(
+    Object.entries(values).filter(([key, value]) => value !== undefined && toml[key] !== value),
+  );
+}
+
+function canonicalToml(provider: string, modelID: string) {
+  const key = `${provider}/${modelID}`;
+  let toml = canonicalTomlByModel.get(key);
+  if (toml === undefined) {
+    const filePath = path.join(PROVIDERS_DIR, provider, "models", `${modelID}.toml`);
+    toml = Bun.TOML.parse(readFileSync(filePath, "utf8")) as Record<string, unknown>;
+    canonicalTomlByModel.set(key, toml);
+  }
+  return toml;
+}
+
+function canonicalCandidates(provider: string, modelID: string) {
+  const candidates = [modelID];
+
+  if (provider === "anthropic") {
+    candidates.push(modelID.replace(/(claude-(?:opus|sonnet|haiku)-\d+)\.(\d+)/, "$1-$2"));
+    candidates.push(modelID.replace(/^claude-3\.5-/, "claude-3-5-"));
+  }
+
+  if (provider === "llama") {
+    candidates.push(modelID.replace(/^llama-(\d+)-(\d+)/, "llama-$1.$2"));
+    candidates.push(modelID.replace(/^llama-(4)-(maverick|scout)$/, "llama-$1-$2-17b"));
+  }
+
+  if (provider === "mistral") {
+    candidates.push(modelID.replace(/-latest$/, ""));
+  }
+
+  if (provider === "minimax") {
+    candidates.push(modelID.replace(/^minimax-m/, "MiniMax-M"));
+  }
+
+  return [...new Set(candidates)];
+}
@@ -0,0 +1,211 @@
+import { z } from "zod";
+
+import type { ExistingModel, SyncProvider, SyncedModel } from "../index.js";
+
+const API_BASE = "https://api.x.ai/v1";
+
+const XAIModel = z.object({
+  id: z.string(),
+  canonical_id: z.string().optional(),
+  created: z.number().int().nonnegative(),
+  aliases: z.array(z.string()).optional(),
+  input_modalities: z.array(z.string()).optional(),
+  output_modalities: z.array(z.string()).optional(),
+  prompt_text_token_price: z.number().int().nonnegative().optional(),
+  cached_prompt_text_token_price: z.number().int().nonnegative().optional(),
+  completion_text_token_price: z.number().int().nonnegative().optional(),
+  max_prompt_length: z.number().int().nonnegative().optional(),
+}).passthrough();
+
+const XAIModelList = z.object({
+  models: z.array(XAIModel),
+}).passthrough();
+
+const XAIResponse = z.object({
+  models: z.array(XAIModel),
+});
+
+const XAIAPIKey = z.object({
+  acls: z.array(z.string()),
+}).passthrough();
+
+type XAIModel = z.infer<typeof XAIModel>;
+
+export const xai = {
+  id: "xai",
+  name: "xAI",
+  modelsDir: "providers/xai/models",
+  skipCreates: true,
+  sourceID(model) {
+    return model.id;
+  },
+  skippedNotice(ids) {
+    if (ids.length === 0) return [];
+    return [
+      `${ids.length} xAI models returned by the API were not created because the Models API does not provide enough authoritative metadata for the catalog, especially output token limits and some feature/capability flags. Existing models are still updated from API-authoritative fields.`,
+      `Skipped remote IDs: ${ids.map((id) => `\`${id}\``).join(", ")}`,
+    ];
+  },
+  async fetchModels() {
+    const key = process.env.XAI_API_KEY;
+    if (key === undefined) throw new Error("xAI sync requires XAI_API_KEY");
+    await assertFullModelAccess(key);
+
+    const models = await Promise.all([
+      fetchTypedModels(key, "language-models"),
+      fetchTypedModels(key, "image-generation-models"),
+      fetchTypedModels(key, "video-generation-models"),
+    ]);
+
+    return { models: models.flat() };
+  },
+  parseModels(raw) {
+    const models = XAIResponse.parse(raw).models;
+    const seen = new Set<string>();
+    const expanded: XAIModel[] = [];
+
+    for (const model of models) {
+      if (!seen.has(model.id)) {
+        seen.add(model.id);
+        expanded.push(model);
+      }
+    }
+
+    for (const model of models) {
+      for (const alias of model.aliases ?? []) {
+        if (seen.has(alias)) continue;
+        seen.add(alias);
+        expanded.push({ ...model, id: alias, canonical_id: model.id });
+      }
+    }
+
+    return expanded;
+  },
+  translateModel(model, context) {
+    const existing = context.existing(model.id);
+    if (existing === undefined) return undefined;
+
+    return {
+      id: model.id,
+      model: buildModel(model, existing),
+    };
+  },
+} satisfies SyncProvider<XAIModel>;
+
+async function assertFullModelAccess(key: string) {
+  const response = await fetch(`${API_BASE}/api-key`, {
+    headers: { Authorization: `Bearer ${key}` },
+  });
+  if (!response.ok) {
+    throw new Error(`xAI API key metadata request failed: ${response.status} ${response.statusText}`);
+  }
+
+  const apiKey = XAIAPIKey.parse(await response.json());
+  if (!apiKey.acls.includes("api-key:model:*")) {
+    throw new Error("xAI sync requires XAI_API_KEY to include api-key:model:* so the model list is not ACL-filtered");
+  }
+}
+
+async function fetchTypedModels(key: string, endpoint: string) {
+  const response = await fetch(`${API_BASE}/${endpoint}`, {
+    headers: { Authorization: `Bearer ${key}` },
+  });
+  if (!response.ok) {
+    throw new Error(`xAI ${endpoint} request failed: ${response.status} ${response.statusText}`);
+  }
+
+  return XAIModelList.parse(await response.json()).models;
+}
+
+function dateFromTimestamp(timestamp: number) {
+  return new Date(timestamp * 1000).toISOString().slice(0, 10);
+}
+
+type Modality = "text" | "audio" | "image" | "video" | "pdf";
+
+function modalities(values: string[] | undefined, fallback: Modality[]) {
+  const allowed = new Set<Modality>(["text", "audio", "image", "video", "pdf"]);
+  const result = (values ?? [])
+    .map((value) => value.toLowerCase())
+    .filter((value): value is Modality => allowed.has(value as Modality));
+  if (result.includes("image")) result.push("pdf");
+  return [...new Set(result.length > 0 ? result : fallback)];
+}
+
+function tokenPrice(value: number | undefined) {
+  if (value === undefined) return undefined;
+  return value / 10_000;
+}
+
+function preservedCostTiers(existing: ExistingModel) {
+  // The xAI models API exposes base pricing only; long-context tiers are curated from xAI docs/console.
+  return existing.cost?.tiers;
+}
+
+function cost(model: XAIModel, existing: ExistingModel) {
+  const input = tokenPrice(model.prompt_text_token_price);
+  const output = tokenPrice(model.completion_text_token_price);
+  if (input === undefined || output === undefined) return existing.cost;
+
+  return {
+    input,
+    output,
+    reasoning: existing.cost?.reasoning,
+    cache_read: tokenPrice(model.cached_prompt_text_token_price),
+    cache_write: existing.cost?.cache_write,
+    input_audio: existing.cost?.input_audio,
+    output_audio: existing.cost?.output_audio,
+    tiers: preservedCostTiers(existing),
+  };
+}
+
+function buildModel(model: XAIModel, existing: ExistingModel): SyncedModel {
+  const name = existing.name;
+  const attachment = existing.attachment;
+  const reasoning = existing.reasoning;
+  const toolCall = existing.tool_call;
+  const openWeights = existing.open_weights;
+  const limit = existing.limit;
+  const releaseDate = existing.release_date;
+  const lastUpdated = existing.last_updated;
+
+  if (
+    name === undefined
+    || attachment === undefined
+    || reasoning === undefined
+    || toolCall === undefined
+    || openWeights === undefined
+    || limit === undefined
+    || (model.canonical_id !== undefined && releaseDate === undefined)
+    || (model.canonical_id !== undefined && lastUpdated === undefined)
+  ) {
+    throw new Error(`xAI model ${model.id} has incomplete local TOML metadata required for sync`);
+  }
+
+  const input = modalities(model.input_modalities, existing.modalities?.input ?? ["text"]);
+  const output = modalities(model.output_modalities, existing.modalities?.output ?? ["text"]);
+  const created = dateFromTimestamp(model.created);
+
+  return {
+    name,
+    family: existing.family,
+    release_date: model.canonical_id === undefined ? created : releaseDate!,
+    last_updated: model.canonical_id === undefined ? created : lastUpdated!,
+    attachment: input.some((value) => value !== "text"),
+    reasoning,
+    temperature: existing.temperature,
+    tool_call: toolCall,
+    structured_output: existing.structured_output,
+    knowledge: existing.knowledge,
+    open_weights: openWeights,
+    status: existing.status,
+    interleaved: existing.interleaved,
+    cost: cost(model, existing),
+    limit: {
+      input: limit.input,
+      context: model.max_prompt_length ?? limit.context,
+      output: limit.output,
+    },
+    modalities: { input, output },
+  };
+}
@@ -0,0 +1,23 @@
+name = "Qwen3.7 Max"
+family = "qwen"
+release_date = "2026-05-21"
+last_updated = "2026-05-21"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+open_weights = false
+
+[cost]
+input = 2.50
+output = 7.50
+cache_read = 0.50
+cache_write = 3.125
+
+[limit]
+context = 1_000_000
+output = 65_536
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -0,0 +1,22 @@
+name = "Gemma Sea Lion V4 27B It"
+family = "gemma"
+release_date = "2025-09-23"
+last_updated = "2025-09-23"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = false
+structured_output = false
+open_weights = true
+
+[cost]
+input = 0.351
+output = 0.555
+
+[limit]
+context = 128_000
+output = 128_000
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -0,0 +1,22 @@
+name = "Deepseek R1 Distill Qwen 32B"
+family = "deepseek"
+release_date = "2025-01-22"
+last_updated = "2025-01-22"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = false
+structured_output = false
+open_weights = true
+
+[cost]
+input = 0.497
+output = 4.881
+
+[limit]
+context = 80_000
+output = 80_000
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -0,0 +1,22 @@
+name = "Gemma 3 12B It"
+family = "gemma"
+release_date = "2025-03-18"
+last_updated = "2025-03-18"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = false
+structured_output = false
+open_weights = true
+
+[cost]
+input = 0.345
+output = 0.556
+
+[limit]
+context = 80_000
+output = 80_000
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -6,15 +6,16 @@ attachment = true
 reasoning = true
 temperature = true
 tool_call = true
+structured_output = true
 open_weights = true

 [cost]
-input = 0.10
-output = 0.30
+input = 0.1
+output = 0.3

 [limit]
-context = 256000
-output = 16384
+context = 256_000
+output = 16_384

 [modalities]
 input = ["text", "image"]
@@ -1,17 +1,17 @@
-name = "Trinity Large Preview"
-family = "trinity"
-release_date = "2026-01-27"
-last_updated = "2026-01-27"
+name = "Granite 4.0 H Micro"
+family = "granite"
+release_date = "2025-10-07"
+last_updated = "2025-10-07"
 attachment = false
 reasoning = false
 temperature = true
 tool_call = true
-structured_output = true
+structured_output = false
 open_weights = true

 [cost]
-input = 0.15
-output = 0.45
+input = 0.017
+output = 0.112

 [limit]
 context = 131_000
@@ -0,0 +1,22 @@
+name = "Llama 2 7B Chat fp16"
+family = "llama"
+release_date = "2023-11-07"
+last_updated = "2023-11-07"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = false
+structured_output = false
+open_weights = true
+
+[cost]
+input = 0.556
+output = 6.667
+
+[limit]
+context = 4_096
+output = 4_096
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -0,0 +1,22 @@
+name = "Llama 3 8B Instruct Awq"
+family = "llama"
+release_date = "2024-05-09"
+last_updated = "2024-05-09"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = false
+structured_output = false
+open_weights = true
+
+[cost]
+input = 0.123
+output = 0.266
+
+[limit]
+context = 8_192
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -0,0 +1,22 @@
+name = "Llama 3 8B Instruct"
+family = "llama"
+release_date = "2024-04-18"
+last_updated = "2024-04-18"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = false
+structured_output = false
+open_weights = true
+
+[cost]
+input = 0.282
+output = 0.827
+
+[limit]
+context = 7_968
+output = 7_968
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -0,0 +1,22 @@
+name = "Llama 3.1 8B Instruct Awq"
+family = "llama"
+release_date = "2024-07-25"
+last_updated = "2024-07-25"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = false
+structured_output = false
+open_weights = true
+
+[cost]
+input = 0.123
+output = 0.266
+
+[limit]
+context = 8_192
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -0,0 +1,22 @@
+name = "Llama 3.1 8B Instruct fp8"
+family = "llama"
+release_date = "2024-07-25"
+last_updated = "2024-07-25"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = false
+structured_output = false
+open_weights = true
+
+[cost]
+input = 0.152
+output = 0.287
+
+[limit]
+context = 32_000
+output = 32_000
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -0,0 +1,22 @@
+name = "Llama 3.2 11B Vision Instruct"
+family = "llama"
+release_date = "2024-09-25"
+last_updated = "2024-09-25"
+attachment = true
+reasoning = false
+temperature = true
+tool_call = false
+structured_output = false
+open_weights = true
+
+[cost]
+input = 0.0485
+output = 0.676
+
+[limit]
+context = 128_000
+output = 128_000
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
@@ -0,0 +1,22 @@
+name = "Llama 3.2 1B Instruct"
+family = "llama"
+release_date = "2024-09-25"
+last_updated = "2024-09-25"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = false
+structured_output = false
+open_weights = true
+
+[cost]
+input = 0.027
+output = 0.201
+
+[limit]
+context = 60_000
+output = 60_000
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -0,0 +1,22 @@
+name = "Llama 3.2 3B Instruct"
+family = "llama"
+release_date = "2024-09-25"
+last_updated = "2024-09-25"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = false
+structured_output = false
+open_weights = true
+
+[cost]
+input = 0.0509
+output = 0.335
+
+[limit]
+context = 80_000
+output = 80_000
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -0,0 +1,22 @@
+name = "Llama 3.3 70B Instruct fp8 Fast"
+family = "llama"
+release_date = "2024-12-06"
+last_updated = "2024-12-06"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = true
+structured_output = false
+open_weights = true
+
+[cost]
+input = 0.293
+output = 2.253
+
+[limit]
+context = 24_000
+output = 24_000
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -6,6 +6,7 @@ attachment = true
 reasoning = false
 temperature = true
 tool_call = true
+structured_output = false
 open_weights = true

 [cost]
@@ -13,8 +14,8 @@ input = 0.27
 output = 0.85

 [limit]
-context = 128000
-output = 16384
+context = 131_000
+output = 16_384

 [modalities]
 input = ["text", "image"]
@@ -0,0 +1,22 @@
+name = "Llama Guard 3 8B"
+family = "llama"
+release_date = "2025-01-22"
+last_updated = "2025-01-22"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = false
+structured_output = false
+open_weights = true
+
+[cost]
+input = 0.484
+output = 0.03
+
+[limit]
+context = 131_072
+output = 131_072
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -0,0 +1,22 @@
+name = "Mistral 7B Instruct V0.1"
+family = "mistral"
+release_date = "2023-11-07"
+last_updated = "2023-11-07"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = false
+structured_output = false
+open_weights = true
+
+[cost]
+input = 0.11
+output = 0.19
+
+[limit]
+context = 2_824
+output = 2_824
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -0,0 +1,22 @@
+name = "Mistral Small 3.1 24B Instruct"
+family = "mistral-small"
+release_date = "2025-03-18"
+last_updated = "2025-03-18"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = true
+structured_output = false
+open_weights = true
+
+[cost]
+input = 0.351
+output = 0.555
+
+[limit]
+context = 128_000
+output = 128_000
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -4,9 +4,9 @@ release_date = "2026-04-20"
 last_updated = "2026-04-20"
 attachment = true
 reasoning = true
-structured_output = true
 temperature = true
 tool_call = true
+structured_output = true
 knowledge = "2025-01"
 open_weights = true

@@ -15,11 +15,11 @@ field = "reasoning_content"

 [cost]
 input = 0.95
-output = 4.00
+output = 4
 cache_read = 0.16

 [limit]
-context = 256_000
+context = 262_144
 output = 256_000

 [modalities]
@@ -6,14 +6,15 @@ attachment = false
 reasoning = true
 temperature = true
 tool_call = true
+structured_output = true
 open_weights = true

 [interleaved]
 field = "reasoning_content"

 [cost]
-input = 0.50
-output = 1.50
+input = 0.5
+output = 1.5

 [limit]
 context = 256_000
@@ -1,10 +1,12 @@
 name = "GPT OSS 120B"
+family = "gpt-oss"
 release_date = "2025-08-05"
 last_updated = "2025-08-05"
 attachment = false
 reasoning = true
 temperature = true
 tool_call = true
+structured_output = true
 open_weights = true

 [cost]
@@ -12,8 +14,8 @@ input = 0.35
 output = 0.75

 [limit]
-context = 128000
-output = 16384
+context = 128_000
+output = 16_384

 [modalities]
 input = ["text"]
@@ -1,10 +1,12 @@
 name = "GPT OSS 20B"
+family = "gpt-oss"
 release_date = "2025-08-05"
 last_updated = "2025-08-05"
 attachment = false
 reasoning = true
 temperature = true
 tool_call = true
+structured_output = true
 open_weights = true

 [cost]
@@ -12,8 +14,8 @@ input = 0.2
 output = 0.3

 [limit]
-context = 128000
-output = 16384
+context = 128_000
+output = 16_384

 [modalities]
 input = ["text"]
@@ -0,0 +1,22 @@
+name = "Qwen2.5 Coder 32B Instruct"
+family = "qwen"
+release_date = "2025-02-27"
+last_updated = "2025-02-27"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = false
+structured_output = false
+open_weights = true
+
+[cost]
+input = 0.66
+output = 1
+
+[limit]
+context = 32_768
+output = 32_768
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -0,0 +1,22 @@
+name = "Qwen3 30B A3b fp8"
+family = "qwen"
+release_date = "2025-04-30"
+last_updated = "2025-04-30"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = false
+open_weights = true
+
+[cost]
+input = 0.0509
+output = 0.335
+
+[limit]
+context = 32_768
+output = 32_768
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -0,0 +1,22 @@
+name = "Qwq 32B"
+family = "qwen"
+release_date = "2025-03-05"
+last_updated = "2025-03-05"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = false
+structured_output = false
+open_weights = true
+
+[cost]
+input = 0.66
+output = 1
+
+[limit]
+context = 24_000
+output = 24_000
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -6,12 +6,13 @@ attachment = false
 reasoning = true
 temperature = true
 tool_call = true
+structured_output = true
 knowledge = "2025-04"
 open_weights = true

 [cost]
-input = 0.06
-output = 0.40
+input = 0.0605
+output = 0.4

 [limit]
 context = 131_072
@@ -0,0 +1,22 @@
+name = "DeepSeek V3.2"
+family = "deepseek"
+release_date = "2025-07-22"
+last_updated = "2025-07-22"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = true
+open_weights = false
+
+[cost]
+input = 0.28
+output = 0.38
+cache_read = 0.06
+
+[limit]
+context = 163_840
+output = 163_840
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -0,0 +1,14 @@
+[extends]
+from = "deepseek/deepseek-v4-flash"
+
+[cost]
+input = 0.12
+output = 0.21
+cache_read = 0.02
+
+[limit]
+context = 1_000_000
+output = 131_072
+
+[provider]
+npm = "@ai-sdk/openai-compatible"
@@ -0,0 +1,26 @@
+name = "DeepSeek V4 Pro (Precision)"
+family = "deepseek-thinking"
+release_date = "2026-04-24"
+last_updated = "2026-04-24"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = true
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 1.25
+output = 2.50
+cache_read = 0.10
+
+[limit]
+context = 1_000_000
+output = 131_072
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -0,0 +1,14 @@
+[extends]
+from = "deepseek/deepseek-v4-pro"
+
+[cost]
+input = 0.40
+output = 0.85
+cache_read = 0.003
+
+[limit]
+context = 1_000_000
+output = 131_072
+
+[provider]
+npm = "@ai-sdk/openai-compatible"
@@ -0,0 +1,14 @@
+[extends]
+from = "google/gemma-4-31b-it"
+
+[cost]
+input = 0.10
+output = 0.30
+cache_read = 0.02
+
+[limit]
+context = 262_144
+output = 262_144
+
+[provider]
+npm = "@ai-sdk/openai-compatible"
@@ -0,0 +1,10 @@
+[extends]
+from = "zhipuai/glm-4.7-flash"
+
+[cost]
+input = 0.04
+output = 0.30
+cache_read = 0.008
+
+[provider]
+npm = "@ai-sdk/openai-compatible"
@@ -0,0 +1,14 @@
+[extends]
+from = "zhipuai/glm-4.7"
+
+[cost]
+input = 0.25
+output = 1.10
+cache_read = 0.05
+
+[limit]
+context = 202_752
+output = 202_752
+
+[provider]
+npm = "@ai-sdk/openai-compatible"
@@ -0,0 +1,26 @@
+name = "GLM 5.1 (Precision)"
+family = "glm"
+release_date = "2026-03-27"
+last_updated = "2026-03-27"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = false
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 0.75
+output = 2.90
+cache_read = 0.15
+
+[limit]
+context = 202_752
+output = 202_752
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -0,0 +1,14 @@
+[extends]
+from = "zhipuai/glm-5.1"
+
+[cost]
+input = 0.45
+output = 2.10
+cache_read = 0.09
+
+[limit]
+context = 202_752
+output = 202_752
+
+[provider]
+npm = "@ai-sdk/openai-compatible"
@@ -0,0 +1,14 @@
+[extends]
+from = "zhipuai/glm-5"
+
+[cost]
+input = 0.48
+output = 1.90
+cache_read = 0.10
+
+[limit]
+context = 202_752
+output = 202_752
+
+[provider]
+npm = "@ai-sdk/openai-compatible"
@@ -0,0 +1,21 @@
+name = "Experiment!: Greg"
+release_date = "2026-01-27"
+last_updated = "2026-01-27"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = false
+open_weights = false
+
+[cost]
+input = 0.10
+output = 0.20
+cache_read = 0.02
+
+[limit]
+context = 229_376
+output = 229_376
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -0,0 +1,26 @@
+name = "Kimi K2.5 (Lightning)"
+family = "kimi-k2.5"
+release_date = "2026-02-06"
+last_updated = "2026-02-06"
+attachment = false
+reasoning = true
+temperature = false
+tool_call = true
+structured_output = true
+open_weights = true
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 1.00
+output = 3.00
+cache_read = 0.20
+
+[limit]
+context = 131_072
+output = 32_768
+
+[modalities]
+input = ["text", "image", "video"]
+output = ["text"]
@@ -0,0 +1,10 @@
+[extends]
+from = "moonshotai/kimi-k2.5"
+
+[cost]
+input = 0.35
+output = 1.70
+cache_read = 0.07
+
+[provider]
+npm = "@ai-sdk/openai-compatible"
@@ -0,0 +1,26 @@
+name = "Kimi K2.6 (Precision)"
+family = "kimi-k2.6"
+release_date = "2026-04-21"
+last_updated = "2026-04-21"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = true
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 0.55
+output = 2.70
+cache_read = 0.11
+
+[limit]
+context = 262_144
+output = 262_144
+
+[modalities]
+input = ["text", "image", "video"]
+output = ["text"]
@@ -0,0 +1,10 @@
+[extends]
+from = "moonshotai/kimi-k2.6"
+
+[cost]
+input = 0.50
+output = 1.99
+cache_read = 0.10
+
+[provider]
+npm = "@ai-sdk/openai-compatible"
@@ -0,0 +1,25 @@
+name = "MiMo-V2.5-Pro (Precision)"
+family = "mimo"
+release_date = "2026-04-22"
+last_updated = "2026-04-22"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+open_weights = true
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 0.80
+output = 2.50
+cache_read = 0.16
+
+[limit]
+context = 1_000_000
+output = 131_072
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -0,0 +1,10 @@
+[extends]
+from = "xiaomi/mimo-v2.5-pro"
+
+[cost]
+input = 0.50
+output = 1.50
+cache_read = 0.10
+
+[provider]
+npm = "@ai-sdk/openai-compatible"
@@ -0,0 +1,12 @@
+reasoning = false
+
+[extends]
+from = "minimax/MiniMax-M2.5"
+
+[cost]
+input = 0.11
+output = 0.95
+cache_read = 0.02
+
+[provider]
+npm = "@ai-sdk/openai-compatible"
@@ -0,0 +1,14 @@
+[extends]
+from = "alibaba/qwen3.5-397b-a17b"
+
+[cost]
+input = 0.35
+output = 1.75
+cache_read = 0.07
+
+[limit]
+context = 262_144
+output = 262_144
+
+[provider]
+npm = "@ai-sdk/openai-compatible"
@@ -0,0 +1,26 @@
+name = "Qwen3.5 9B"
+family = "qwen"
+release_date = "2026-03-13"
+last_updated = "2026-03-13"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = true
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 0.04
+output = 0.15
+cache_read = 0.008
+
+[limit]
+context = 262_144
+output = 262_144
+
+[modalities]
+input = ["text", "image", "video", "audio"]
+output = ["text"]
@@ -0,0 +1,14 @@
+[extends]
+from = "alibaba/qwen3.6-27b"
+
+[cost]
+input = 0.20
+output = 1.50
+cache_read = 0.04
+
+[limit]
+context = 262_144
+output = 262_144
+
+[provider]
+npm = "@ai-sdk/openai-compatible"
@@ -0,0 +1,5 @@
+name = "CrofAI"
+env = ["CROF_API_KEY"]
+npm = "@ai-sdk/openai-compatible"
+api = "https://crof.ai/v1"
+doc = "https://crof.ai/docs"
@@ -0,0 +1,14 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 90.158706 90.190743" fill="currentColor">
+	<g transform="matrix(0.32039341,0,0,0.32039341,-193.03703,-175.63967)">
+		<polygon points="720.6,711.5 602.5,711.5 691.7,622.3 809.8,622.3" />
+		<polygon points="720.6,711.5 809.8,622.3 809.8,711.5" />
+		<polygon points="720.6,711.5 809.8,711.5 809.8,739.5 719.7,829.6 719.2,829.6 719.2,712.8" />
+		<polygon points="602.5,711.5 720.6,711.5 602.5,829.6" />
+		<path d="m 809.8,739.5 v 89.7 c -0.1,0.1 -0.3,0.3 -0.5,0.5 h -89.7 z" />
+		<polygon points="691.7,622.3 602.5,622.3 676.5,548.2 765.7,548.2" />
+		<polygon points="825.9,606.2 707.8,606.2 765.7,548.2 883.9,548.2" />
+		<polygon points="602.5,622.3 691.7,622.3 602.5,711.5" />
+		<polygon points="602.5,829.6 719.2,712.8 719.2,829.6" />
+		<path d="m 809.8,622.3 74,-74 c 0,0 0,130.6 0,206.9 -23.2,23.2 -49.3,49.3 -74,74 z" />
+	</g>
+</svg>
@@ -0,0 +1,15 @@
+name = "MiniMax M2.5"
+structured_output = true
+
+[extends]
+from = "minimax/MiniMax-M2.5"
+
+[cost]
+input = 0.24
+output = 0.9
+cache_read = 0.03
+cache_write = 0
+
+[limit]
+context = 196608
+output = 196608
@@ -0,0 +1,12 @@
+[extends]
+from = "moonshotai/kimi-k2.6"
+
+[cost]
+input = 0.78
+output = 3.5
+cache_read = 0.2
+cache_write = 0
+
+[modalities]
+input = [ "text", "image" ]
+output = [ "text" ]
@@ -0,0 +1,15 @@
+name = "Llama 3.3 70B Instruct"
+structured_output = true
+
+[extends]
+from = "llama/llama-3.3-70b-instruct"
+
+[cost]
+input = 0.12
+output = 0.38
+cache_read = 0
+cache_write = 0
+
+[limit]
+context = 131072
+output = 131072
@@ -0,0 +1,14 @@
+name = "GLM 5.1"
+
+[extends]
+from = "zai/glm-5.1"
+
+[cost]
+input = 1.4
+output = 4.4
+cache_read = 0.26
+cache_write = 0
+
+[limit]
+context = 202752
+output = 202752
@@ -0,0 +1,5 @@
+name = "Inceptron"
+npm = "@ai-sdk/openai-compatible"
+env = ["INCEPTRON_API_KEY"]
+api = "https://api.inceptron.io/v1"
+doc = "https://docs.inceptron.io"
@@ -0,0 +1,2 @@
+[extends]
+from = "google/gemini-3.5-flash"
@@ -0,0 +1,2 @@
+[extends]
+from = "xai/grok-4.20-0309-non-reasoning"
@@ -0,0 +1,2 @@
+[extends]
+from = "xai/grok-4.20-0309-reasoning"
@@ -0,0 +1,2 @@
+[extends]
+from = "alibaba/qwen3.7-max"
@@ -0,0 +1,28 @@
+name = "Gemini 3.5 Flash"
+family = "gemini-flash"
+release_date = "2026-05-19"
+last_updated = "2026-05-19"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+knowledge = "2025-01"
+open_weights = false
+
+[cost]
+input = 1.50
+output = 9.00
+cache_read = 0.15
+input_audio = 1.50
+
+[limit]
+context = 1_048_576
+output = 65_536
+
+[modalities]
+input = ["text", "image", "video", "audio", "pdf"]
+output = ["text"]
+
+[provider]
+npm = "@ai-sdk/google"
@@ -0,0 +1,23 @@
+name = "Grok Build 0.1"
+family = "grok-build"
+release_date = "2026-05-20"
+last_updated = "2026-05-20"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = false
+
+[cost]
+input = 1.00
+output = 2.00
+cache_read = 0.20
+
+[limit]
+context = 256_000
+output = 256_000
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
@@ -8,6 +8,7 @@ temperature = true
 tool_call = true
 knowledge = "2025-01"
 open_weights = true
+status = "deprecated"

 [cost]
 input = 0
@@ -8,6 +8,7 @@ temperature = true
 tool_call = true
 knowledge = "2025-04"
 open_weights = false
+status = "deprecated"

 [cost]
 input = 0
@@ -1,14 +1,9 @@
-name = "Claude Haiku 4.5"
-family = "claude-haiku"
-release_date = "2025-10-15"
-last_updated = "2025-10-15"
-attachment = true
-reasoning = true
 temperature = true
 tool_call = true
 structured_output = true
-knowledge = "2025-02-28"
-open_weights = false
+
+[extends]
+from = "anthropic/claude-haiku-4-5"

 [cost]
 input = 1
@@ -1,14 +1,9 @@
-name = "Claude Opus 4.1"
-family = "claude-opus"
-release_date = "2025-08-05"
-last_updated = "2025-08-05"
-attachment = true
-reasoning = true
 temperature = true
 tool_call = true
 structured_output = true
-knowledge = "2025-01-31"
-open_weights = false
+
+[extends]
+from = "anthropic/claude-opus-4-1"

 [cost]
 input = 15
@@ -1,14 +1,9 @@
-name = "Claude Opus 4.5"
-family = "claude-opus"
-release_date = "2025-11-24"
-last_updated = "2025-11-24"
-attachment = true
-reasoning = true
 temperature = true
 tool_call = true
 structured_output = true
-knowledge = "2025-05-30"
-open_weights = false
+
+[extends]
+from = "anthropic/claude-opus-4-5"

 [cost]
 input = 5
@@ -1,14 +1,10 @@
-name = "Claude Opus 4.6"
-family = "claude-opus"
-release_date = "2026-02-04"
-last_updated = "2026-02-04"
-attachment = true
-reasoning = true
 temperature = true
 tool_call = true
 structured_output = true
-knowledge = "2025-05-31"
-open_weights = false
+
+[extends]
+from = "anthropic/claude-opus-4-6"
+omit = ["experimental"]

 [cost]
 input = 5
@@ -1,14 +1,10 @@
-name = "Claude Opus 4.7"
-family = "claude-opus"
-release_date = "2026-04-16"
-last_updated = "2026-04-16"
-attachment = true
-reasoning = true
 temperature = false
 tool_call = true
 structured_output = true
-knowledge = "2026-01-31"
-open_weights = false
+
+[extends]
+from = "anthropic/claude-opus-4-7"
+omit = ["experimental"]

 [cost]
 input = 5
@@ -1,14 +1,9 @@
-name = "Claude Sonnet 4.5"
-family = "claude-sonnet"
-release_date = "2025-09-29"
-last_updated = "2025-09-29"
-attachment = true
-reasoning = true
 temperature = true
 tool_call = true
 structured_output = true
-knowledge = "2025-01-31"
-open_weights = false
+
+[extends]
+from = "anthropic/claude-sonnet-4-5"

 [cost]
 input = 3
@@ -1,14 +1,9 @@
-name = "Claude Sonnet 4.6"
-family = "claude-sonnet"
-release_date = "2026-02-17"
-last_updated = "2026-02-17"
-attachment = true
-reasoning = true
 temperature = true
 tool_call = true
 structured_output = true
-knowledge = "2025-08-31"
-open_weights = false
+
+[extends]
+from = "anthropic/claude-sonnet-4-6"

 [cost]
 input = 3
@@ -1,14 +1,9 @@
-name = "Command R (08-2024)"
-family = "command-r"
-release_date = "2024-08-30"
-last_updated = "2024-08-30"
-attachment = false
-reasoning = false
 temperature = true
 tool_call = true
 structured_output = true
-knowledge = "2024-03-31"
-open_weights = false
+
+[extends]
+from = "cohere/command-r-08-2024"

 [cost]
 input = 0.15
@@ -1,14 +1,9 @@
-name = "Command R+ (08-2024)"
-family = "command-r"
-release_date = "2024-08-30"
-last_updated = "2024-08-30"
-attachment = false
-reasoning = false
 temperature = true
 tool_call = true
 structured_output = true
-knowledge = "2024-03-31"
-open_weights = false
+
+[extends]
+from = "cohere/command-r-plus-08-2024"

 [cost]
 input = 2.5
@@ -1,14 +1,9 @@
-name = "Command R7B (12-2024)"
-family = "command-r"
-release_date = "2024-12-14"
-last_updated = "2024-12-14"
-attachment = false
-reasoning = false
 temperature = true
 tool_call = false
 structured_output = true
-knowledge = "2024-08-31"
-open_weights = false
+
+[extends]
+from = "cohere/command-r7b-12-2024"

 [cost]
 input = 0.0375
@@ -1,14 +1,11 @@
-name = "DeepSeek V3"
-family = "deepseek"
-release_date = "2024-12-26"
-last_updated = "2024-12-26"
 attachment = false
-reasoning = false
 temperature = true
 tool_call = true
 structured_output = true
-knowledge = "2024-07-31"
-open_weights = true
+
+[extends]
+from = "deepseek/deepseek-chat"
+omit = ["cost.cache_read"]

 [cost]
 input = 0.32
@@ -1,13 +1,9 @@
-name = "DeepSeek V4 Flash"
-family = "deepseek"
-release_date = "2026-04-24"
-last_updated = "2026-04-24"
-attachment = false
-reasoning = true
 temperature = true
 tool_call = true
 structured_output = true
-open_weights = true
+
+[extends]
+from = "deepseek/deepseek-v4-flash"

 [interleaved]
 field = "reasoning_content"
@@ -1,13 +1,11 @@
 name = "DeepSeek V4 Flash (free)"
-family = "deepseek"
-release_date = "2026-04-24"
-last_updated = "2026-04-24"
-attachment = false
-reasoning = true
 temperature = false
 tool_call = true
 structured_output = false
-open_weights = true
+
+[extends]
+from = "deepseek/deepseek-v4-flash"
+omit = ["cost.cache_read"]

 [cost]
 input = 0
@@ -1,13 +1,9 @@
-name = "DeepSeek V4 Pro"
-family = "deepseek"
-release_date = "2026-04-24"
-last_updated = "2026-04-24"
-attachment = false
-reasoning = true
 temperature = true
 tool_call = true
 structured_output = true
-open_weights = true
+
+[extends]
+from = "deepseek/deepseek-v4-pro"

 [interleaved]
 field = "reasoning_content"
@@ -17,7 +17,7 @@ cache_read = 0.025
 cache_write = 0.083333

 [limit]
-context = 1_048_576
+context = 1_000_000
 output = 8_192

 [modalities]
@@ -1,14 +1,10 @@
-name = "Nano Banana (Gemini 2.5 Flash Image)"
-family = "gemini"
-release_date = "2025-10-07"
-last_updated = "2025-10-07"
-attachment = true
 reasoning = false
 temperature = true
 tool_call = false
 structured_output = true
-knowledge = "2025-01-31"
-open_weights = false
+
+[extends]
+from = "google/gemini-2.5-flash-image"

 [cost]
 input = 0.3
@@ -1,14 +1,10 @@
-name = "Gemini 2.5 Flash Lite"
-family = "gemini-flash-lite"
-release_date = "2025-07-22"
-last_updated = "2025-07-22"
-attachment = true
-reasoning = true
 temperature = true
 tool_call = true
 structured_output = true
-knowledge = "2025-01-31"
-open_weights = false
+
+[extends]
+from = "google/gemini-2.5-flash-lite"
+omit = ["cost.input_audio"]

 [cost]
 input = 0.1
@@ -1,14 +1,10 @@
-name = "Gemini 2.5 Flash"
-family = "gemini-flash"
-release_date = "2025-06-17"
-last_updated = "2025-06-17"
-attachment = true
-reasoning = true
 temperature = true
 tool_call = true
 structured_output = true
-knowledge = "2025-01-31"
-open_weights = false
+
+[extends]
+from = "google/gemini-2.5-flash"
+omit = ["cost.input_audio"]

 [cost]
 input = 0.3
@@ -1,14 +1,10 @@
-name = "Gemini 2.5 Pro"
-family = "gemini"
-release_date = "2025-06-17"
-last_updated = "2025-06-17"
-attachment = true
-reasoning = true
 temperature = true
 tool_call = true
 structured_output = true
-knowledge = "2025-01-31"
-open_weights = false
+
+[extends]
+from = "google/gemini-2.5-pro"
+omit = ["cost.tiers", "cost.context_over_200k"]

 [cost]
 input = 1.25
@@ -1,14 +1,10 @@
-name = "Gemini 3 Flash Preview"
-family = "gemini-flash"
-release_date = "2025-12-17"
-last_updated = "2025-12-17"
-attachment = true
-reasoning = true
 temperature = true
 tool_call = true
 structured_output = true
-knowledge = "2025-01"
-open_weights = false
+
+[extends]
+from = "google/gemini-3-flash-preview"
+omit = ["cost.input_audio"]

 [interleaved]
 field = "reasoning_details"
@@ -1,14 +1,9 @@
-name = "Nano Banana 2 (Gemini 3.1 Flash Image Preview)"
-family = "gemini-flash"
-release_date = "2026-02-26"
-last_updated = "2026-02-26"
-attachment = true
-reasoning = true
 temperature = true
 tool_call = false
 structured_output = true
-knowledge = "2025-01"
-open_weights = false
+
+[extends]
+from = "google/gemini-3.1-flash-image-preview"

 [cost]
 input = 0.5
@@ -1,13 +1,10 @@
-name = "Gemini 3.1 Flash Lite Preview"
-family = "gemini-flash-lite"
-release_date = "2026-03-03"
-last_updated = "2026-03-03"
-attachment = true
-reasoning = true
 temperature = true
 tool_call = true
 structured_output = true
-open_weights = false
+
+[extends]
+from = "google/gemini-3.1-flash-lite-preview"
+omit = ["cost.input_audio"]

 [cost]
 input = 0.25
@@ -1,13 +1,10 @@
-name = "Gemini 3.1 Flash Lite"
-family = "gemini"
-release_date = "2026-05-07"
-last_updated = "2026-05-07"
-attachment = true
-reasoning = true
 temperature = true
 tool_call = true
 structured_output = true
-open_weights = false
+
+[extends]
+from = "google/gemini-3.1-flash-lite"
+omit = ["cost.input_audio"]

 [cost]
 input = 0.25
@@ -1,14 +1,9 @@
-name = "Gemini 3.1 Pro Preview Custom Tools"
-family = "gemini-pro"
-release_date = "2026-02-25"
-last_updated = "2026-02-25"
-attachment = true
-reasoning = true
 temperature = true
 tool_call = true
 structured_output = true
-knowledge = "2025-01"
-open_weights = false
+
+[extends]
+from = "google/gemini-3.1-pro-preview-customtools"

 [interleaved]
 field = "reasoning_details"
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Aiden Cline	c8dc9cae24	chore(sync): keep fireworks catalog entries curated	2026-05-21 18:32:00 -05:00
Aiden Cline	964775cf77	chore(sync): standardize fireworks api key env	2026-05-21 18:25:07 -05:00
Aiden Cline	7509cf1797	chore(sync): add fireworks ai provider	2026-05-21 18:18:35 -05:00
Aiden Cline	2738f81d1a	Merge pull request #1828 from anomalyco/refactor/sync-core-layout refactor: move sync implementation into core src	2026-05-21 18:10:09 -05:00
Aiden Cline	89b834086a	refactor: move sync implementation into core src	2026-05-21 18:06:25 -05:00
Aiden Cline	1ab2ff8163	Merge pull request #1826 from smakosh/add-llmgateway-models feat: add new LLM Gateway text models	2026-05-21 17:58:29 -05:00
Frank	9468676683	update zen models	2026-05-21 18:42:36 -04:00
Claude	6cdd2f054b	Merge upstream/dev into add-llmgateway-models; resolve gemini-3.5-flash conflict # Conflicts: # providers/google/models/gemini-3.5-flash.toml	2026-05-21 21:48:48 +00:00
Aiden Cline	b13abc9141	Merge pull request #1827 from anomalyco/update-xai-pricing fix xAI long-context pricing	2026-05-21 16:45:43 -05:00
Aiden Cline	e5ba264751	fix xAI long-context pricing	2026-05-21 16:41:36 -05:00
smakosh	a7811fb522	refactor: use extends for gemini and qwen models Add canonical google/gemini-3.5-flash and alibaba/qwen3.7-max defs and have the llmgateway entries extend them, per PR review. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-21 23:38:57 +02:00
smakosh	605fae75d9	feat: add new LLM Gateway text models Add Grok 4.20 (reasoning/non-reasoning), Gemini 3.5 Flash, and Qwen3.7 Max to the llmgateway provider. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-21 23:04:47 +02:00
Aiden Cline	26b05268ae	Merge pull request #1824 from anomalyco/fix/vercel-gemini-35-flash Add new Vercel AI Gateway models	2026-05-21 15:30:35 -05:00
Aiden Cline	1aee13d2e5	Add new Vercel AI Gateway models	2026-05-21 13:21:25 -05:00
Aiden Cline	0a924e6bb2	Merge pull request #1822 from anomalyco/automation/sync-models-xai chore(sync): update xAI model catalog	2026-05-21 13:05:20 -05:00
Aiden Cline	9769b2b11d	Merge pull request #1823 from anomalyco/automation/sync-models-openrouter chore(sync): update OpenRouter model catalog	2026-05-21 13:05:13 -05:00
github-actions[bot]	1b0db099cf	chore(sync): update OpenRouter model catalog	2026-05-21 17:56:13 +00:00
github-actions[bot]	16ed78587c	chore(sync): update xAI model catalog	2026-05-21 17:56:11 +00:00
Frank	0b88965165	update zen models	2026-05-21 13:41:55 -04:00
Aiden Cline	acc704ce39	Merge pull request #1797 from arnavchachra/add-crof-provider add crof.ai provider with 21 models	2026-05-21 11:43:02 -05:00
Aiden Cline	51ad3b264e	Merge pull request #1821 from anomalyco/sync-provider-ci chore: automate provider sync jobs	2026-05-21 11:23:49 -05:00
Aiden Cline	146b6c7084	Merge pull request #1819 from Inceptron-Software/add_inceptron_provider Add Inceptron provider	2026-05-21 11:21:27 -05:00
Aiden Cline	0e3cbe3c64	chore: automate provider sync jobs	2026-05-21 11:21:12 -05:00
Aiden Cline	604d4d66a4	Merge pull request #1820 from Suat-B/codex/xpersona-image-input-20260521 Add image input modality to Xpersona model	2026-05-21 11:00:49 -05:00
SuatB	f5090028b8	Add image input modality to Xpersona model	2026-05-21 09:46:28 -05:00
Frank	4bad8faf29	update zen models	2026-05-21 09:05:13 -04:00
Oskar Gustafsson	0df2ccf586	Add Inceptron provider	2026-05-21 09:24:43 +02:00
Aiden Cline	bafdc00b45	Merge pull request #1812 from anomalyco/openrouter-extends-sync Sync OpenRouter models with extends	2026-05-20 21:07:21 -05:00
Aiden Cline	49840c013b	Merge pull request #1814 from neonn0d/feat/stepfun-ai feat(stepfun-ai): add international StepFun platform	2026-05-20 20:58:37 -05:00
Aiden Cline	eccae0b54e	sync openrouter models with extends	2026-05-20 20:32:11 -05:00
Aiden Cline	4cca29405f	Merge pull request #1817 from dpuyosa/dev Venice: Remove Grok 4.1 Fast and add Grok Build 0.1	2026-05-20 20:26:45 -05:00
Aiden Cline	e40d9dd338	Merge pull request #1818 from anomalyco/cloudflare-sync-env chore(sync): isolate cloudflare credentials	2026-05-20 20:26:22 -05:00
Aiden Cline	6a74991397	chore(sync): isolate cloudflare credentials	2026-05-20 20:19:47 -05:00
dpuyosa	035999cb58	[venice] Replace Grok 4.1 Fast with Grok Build 0.1 - Remove deprecated grok-41-fast model entry - Add grok-build-0-1 with 200K token tiered pricing - Update context to 256K and output limit to 65,536	2026-05-21 02:36:19 +02:00
Frank	cec56bf1bc	update zen models	2026-05-20 19:43:25 -04:00
Aiden Cline	85f0cdcb2f	Merge pull request #1816 from anomalyco/xai-sync Add PDF input modality to Grok models	2026-05-20 18:12:47 -05:00
Aiden Cline	ef80d4df4e	Infer PDF modality for xAI image models	2026-05-20 18:12:11 -05:00
Aiden Cline	af0ef00109	Update xAI Grok PDF modalities	2026-05-20 18:06:33 -05:00
Aiden Cline	5fdcea6b36	Merge pull request #1815 from anomalyco/cloudflare-ai-gateway chore(sync): add cloudflare workers ai sync	2026-05-20 18:02:28 -05:00
Aiden Cline	31e56480b4	chore(sync): add cloudflare workers ai sync	2026-05-20 16:55:29 -05:00
Aiden Cline	92a621594e	Merge pull request #1813 from anomalyco/sync-xai chore(sync): add xai model sync	2026-05-20 16:02:38 -05:00
Aiden Cline	d2db353ceb	chore: ignore sync reports	2026-05-20 16:01:51 -05:00
Aiden Cline	900ae509d2	Merge pull request #1808 from ajussak/scaleway Added Mistral Medium 3.5 128B from Scaleway	2026-05-20 15:58:00 -05:00
neo	9d60164243	feat(stepfun-ai): add international StepFun platform StepFun runs two separate platforms with distinct accounts/keys: platform.stepfun.com (China, already covered by providers/stepfun) and platform.stepfun.ai (international). Keys are not interchangeable across the two — .ai keys are rejected by api.stepfun.com as invalid_api_key. Stepfun's own opencode integration guide instructs users to point at https://api.stepfun.ai/step_plan/v1. This adds providers/stepfun-ai for that endpoint, symlinking the shared chat models. Follows the moonshotai / moonshotai-cn pattern.	2026-05-20 20:43:47 +02:00
Adrien Jussak	cd3e99025f	Update Mistral Medium 3.5 128B model configuration to extend from mistral-medium-2604 and adjust context window size.	2026-05-20 20:43:45 +02:00
Aiden Cline	1098981eb6	chore(sync): add xai model sync	2026-05-20 13:23:02 -05:00
Aiden Cline	27a151cf53	Merge pull request #1811 from anomalyco/xai-grok-build-model Add xAI Grok Build model	2026-05-20 13:05:06 -05:00
Aiden Cline	41ff42ab7f	Merge pull request #1810 from fhennerkes/dev poe: add Gemini-3.5-Flash model	2026-05-20 13:04:47 -05:00
Aiden Cline	adf1cbdecd	add xai grok build model	2026-05-20 13:04:13 -05:00
Frank	10ddc78ce0	update zen models	2026-05-20 14:02:01 -04:00
fhennerkes	7ae897e440	poe: add Gemini-3.5-Flash model Add new Google model from Poe API (released 2026-05-19). Uses extends format inheriting from google/gemini-3.5-flash with Poe-specific overrides (name format, no temperature, markup pricing, limited input modalities). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>	2026-05-20 10:52:15 -07:00
Aiden Cline	02e452c2e8	Merge pull request #1805 from anomalyco/sync-google sync google models	2026-05-20 10:53:57 -05:00
Aiden Cline	f05b63fff5	Merge pull request #1807 from anomalyco/automation/sync-models-aggregators chore(sync): update aggregator model catalogs	2026-05-20 10:34:30 -05:00
Adrien Jussak	e277d60236	Add Mistral Medium 3.5 128B to Scaleway	2026-05-20 16:14:37 +02:00
github-actions[bot]	b01277a737	chore(sync): update aggregator model catalogs	2026-05-20 09:25:37 +00:00
Frank	a5da5aa429	update zen models	2026-05-20 04:14:28 -04:00
Aiden Cline	11e1d5623a	Merge pull request #1806 from Cahl-Dee/grid-model-updates-2026-05 Grid model updates 2026-05	2026-05-19 23:42:26 -05:00
Carl DiClementi	e789d7c1f2	Merge branch 'anomalyco:dev' into grid-model-updates-2026-05	2026-05-19 16:06:23 -05:00
Cahl-Dee	899668ad49	added new code and agent models and updated existing text models	2026-05-19 16:04:58 -05:00
arnavchachra	9420048dfe	fix crof model limits and reasoning flag to match Crof API	2026-05-18 21:51:19 +05:30
arnavchachra	8db6c27634	add crof provider with 21 models	2026-05-18 21:40:11 +05:30