updates

feat: add deepseek-v4-flash to the fireworks-ai provider

.github/workflows/sync-models.yml

@@ -0,0 +1,83 @@
+name: Sync Model Catalogs
+
+on:
+  schedule:
+    - cron: "17 8 * * *"
+  workflow_dispatch:
+
+permissions:
+  contents: write
+  issues: write
+  pull-requests: write
+
+concurrency: ${{ github.workflow }}-${{ github.ref }}
+
+jobs:
+  sync:
+    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
+
+    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: Sync model catalogs
+        run: bun models:sync ${{ matrix.group }}
+        env:
+          OPENROUTER_API_KEY: ${{ secrets.OPENROUTER_API_KEY }}
+
+      - name: Validate models
+        run: bun validate
+
+      - name: Create pull request
+        env:
+          GH_TOKEN: ${{ github.token }}
+          BRANCH: ${{ matrix.branch }}
+          LABELS: ${{ matrix.labels }}
+          TITLE: ${{ matrix.title }}
+        run: |
+          if [ -z "$(git status --porcelain -- providers)" ]; then
+            echo "No model catalog changes found."
+            exit 0
+          fi
+
+          git config user.name "github-actions[bot]"
+          git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
+          git checkout -B "$BRANCH"
+          git add providers
+          git commit -m "$TITLE"
+          git push --force-with-lease origin "$BRANCH"
+
+          label_args=()
+          IFS=',' read -ra labels <<< "$LABELS"
+          for label in "${labels[@]}"; do
+            gh label create "$label" --color "0E8A16" --description "Automated model catalog sync" >/dev/null 2>&1 || true
+            label_args+=(--label "$label")
+          done
+
+          pr_number="$(gh pr list --head "$BRANCH" --base dev --json number --jq '.[0].number')"
+          if [ -n "$pr_number" ]; then
+            gh pr edit "$pr_number" --title "$TITLE" --body-file .sync/model-sync-report.md
+            for label in "${labels[@]}"; do
+              gh pr edit "$pr_number" --add-label "$label"
+            done
+          else
+            gh pr create --base dev --head "$BRANCH" --title "$TITLE" --body-file .sync/model-sync-report.md "${label_args[@]}"
+          fi

package.json

@@ -25,7 +25,7 @@
     "wandb:generate": "bun ./packages/core/script/generate-wandb.ts",
     "digitalocean:generate": "bun ./packages/core/script/generate-digitalocean.ts",
     "ambient:generate": "bun ./packages/core/script/generate-ambient.ts",
-    "openrouter:sync": "bun ./packages/core/script/sync-openrouter.ts"
+    "models:sync": "bun ./packages/core/script/sync-models.ts"
   },
   "dependencies": {
     "@cloudflare/workers-types": "^4.20260424.1",

packages/core/script/generate-chutes.ts

@@ -138,7 +138,7 @@ function humanizeModelName(modelId: string): string {
   return modelPart.replace(/-/g, " ");
 }
 
-// ── Family inference (same approach as generate-vercel.ts) ───────────
+// ── Family inference ───────────
 
 function isSubstring(target: string, family: string): boolean {
   return target.toLowerCase().includes(family.toLowerCase());

packages/core/script/sync-models.ts

@@ -0,0 +1,421 @@
+#!/usr/bin/env bun
+
+import path from "node:path";
+import { mkdir, readdir, rm } from "node:fs/promises";
+import { z } from "zod";
+
+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();

packages/core/script/sync-openrouter.ts

@@ -1,318 +0,0 @@
-#!/usr/bin/env bun
-
-import path from "node:path";
-import { mkdir, rm } from "node:fs/promises";
-import { z } from "zod";
-
-import { ModelFamilyValues } from "../src/family.js";
-import { AuthoredModel, AuthoredModelShape } from "../src/schema.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();
-
-const ExistingModel = AuthoredModelShape.partial()
-  .extend({
-    extends: z
-      .object({
-        from: z.string(),
-        omit: z.array(z.string()).optional(),
-      })
-      .strict()
-      .optional(),
-  })
-  .strict();
-
-type OpenRouterModel = z.infer<typeof OpenRouterModel>;
-type ExistingModel = z.infer<typeof ExistingModel>;
-
-function dateFromTimestamp(timestamp: number | undefined) {
-  if (timestamp === undefined) return new Date().toISOString().slice(0, 10);
-  return new Date(timestamp * 1000).toISOString().slice(0, 10);
-}
-
-function formatInteger(n: number) {
-  return String(n).replace(/\B(?=(\d{3})+(?!\d))/g, "_");
-}
-
-function quote(value: string) {
-  return `"${value.replaceAll("\\", "\\\\").replaceAll('"', '\\"')}"`;
-}
-
-function price(value: string | undefined) {
-  if (value === undefined) return undefined;
-  const number = Number(value);
-  return Number.isFinite(number) && number >= 0
-    ? Math.round(number * 1_000_000_000_000) / 1_000_000
-    : undefined;
-}
-
-function modality(value: string) {
-  return value === "file" ? "pdf" : value;
-}
-
-function modalities(values: string[] | undefined, fallback: string[]) {
-  const allowed = new Set(["text", "audio", "image", "video", "pdf"]);
-  const result = (values ?? fallback)
-    .map((value) => modality(value.toLowerCase()))
-    .filter((value) => allowed.has(value));
-  return [...new Set(result.length > 0 ? result : fallback)];
-}
-
-function inferFamily(model: OpenRouterModel, name: string) {
-  const target = `${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 ?? 0;
-  const maxOutput = model.top_provider?.max_completion_tokens ?? existing?.limit?.output ?? context;
-  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") || params.has("response_format"),
-    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: maxOutput,
-    },
-    modalities: { input, output },
-  };
-}
-
-function formatToml(model: ReturnType<typeof buildModel>) {
-  const lines: string[] = [];
-
-  lines.push(`name = ${quote(model.name)}`);
-  if (model.family) 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}`);
-  lines.push(`temperature = ${model.temperature}`);
-  lines.push(`tool_call = ${model.tool_call}`);
-  lines.push(`structured_output = ${model.structured_output}`);
-  if (model.knowledge) lines.push(`knowledge = ${quote(model.knowledge)}`);
-  lines.push(`open_weights = ${model.open_weights}`);
-  if (model.status) 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) {
-    lines.push("");
-    lines.push("[cost]");
-    if (model.cost.input !== undefined) lines.push(`input = ${model.cost.input}`);
-    if (model.cost.output !== undefined) lines.push(`output = ${model.cost.output}`);
-    if (model.cost.reasoning !== undefined) {
-      lines.push(`reasoning = ${model.cost.reasoning}`);
-    }
-    if (model.cost.cache_read !== undefined) {
-      lines.push(`cache_read = ${model.cost.cache_read}`);
-    }
-    if (model.cost.cache_write !== undefined) {
-      lines.push(`cache_write = ${model.cost.cache_write}`);
-    }
-
-    for (const tier of model.cost.tiers ?? []) {
-      lines.push("");
-      lines.push("[[cost.tiers]]");
-      lines.push(`tier = { size = ${formatInteger(tier.tier.size)} }`);
-      if (tier.input !== undefined) lines.push(`input = ${tier.input}`);
-      if (tier.output !== undefined) lines.push(`output = ${tier.output}`);
-      if (tier.cache_read !== undefined) lines.push(`cache_read = ${tier.cache_read}`);
-      if (tier.cache_write !== undefined) lines.push(`cache_write = ${tier.cache_write}`);
-    }
-  }
-
-  lines.push("");
-  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 = ${formatInteger(model.limit.output)}`);
-
-  lines.push("");
-  lines.push("[modalities]");
-  lines.push(`input = [${model.modalities.input.map(quote).join(", ")}]`);
-  lines.push(`output = [${model.modalities.output.map(quote).join(", ")}]`);
-
-  return `${lines.join("\n")}\n`;
-}
-
-async function main() {
-  const dryRun = process.argv.includes("--dry-run");
-  const modelsDir = path.join(
-    import.meta.dirname,
-    "..",
-    "..",
-    "..",
-    "providers",
-    "openrouter",
-    "models",
-  );
-  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} ${response.statusText}`);
-  }
-
-  const parsed = OpenRouterResponse.safeParse(await response.json());
-  if (!parsed.success) throw parsed.error;
-
-  const existingFiles = new Set<string>();
-  for await (const file of new Bun.Glob("**/*.toml").scan({ cwd: modelsDir })) {
-    existingFiles.add(file);
-  }
-
-  let created = 0;
-  let updated = 0;
-  let removed = 0;
-  let unchanged = 0;
-  const apiFiles = new Set<string>();
-
-  for (const apiModel of parsed.data.data) {
-    const relativePath = `${apiModel.id}.toml`;
-    const filePath = path.join(modelsDir, relativePath);
-    const file = Bun.file(filePath);
-    const current = await file.exists() ? await file.text() : undefined;
-    const existing = current === undefined
-      ? undefined
-      : ExistingModel.parse(Bun.TOML.parse(current));
-    const model = buildModel(apiModel, existing);
-    const next = formatToml(model);
-
-    const valid = AuthoredModel.safeParse({
-      id: relativePath.slice(0, -5),
-      ...Bun.TOML.parse(next),
-    });
-    if (!valid.success) {
-      valid.error.cause = { relativePath };
-      throw valid.error;
-    }
-
-    apiFiles.add(relativePath);
-
-    if (current === undefined) {
-      created++;
-      if (dryRun) {
-        console.log(`Would create ${relativePath}`);
-      } else {
-        await mkdir(path.dirname(filePath), { recursive: true });
-        await Bun.write(filePath, next);
-      }
-      continue;
-    }
-
-    if (current !== next) {
-      updated++;
-      if (dryRun) {
-        console.log(`Would update ${relativePath}`);
-      } else {
-        await Bun.write(filePath, next);
-      }
-    } else {
-      unchanged++;
-    }
-  }
-
-  for (const relativePath of existingFiles) {
-    if (apiFiles.has(relativePath)) continue;
-
-    removed++;
-    if (dryRun) {
-      console.log(`Would remove ${relativePath}`);
-    } else {
-      await rm(path.join(modelsDir, relativePath));
-    }
-  }
-
-  console.log(
-    `${dryRun ? "Dry run: " : ""}${created} created, ${updated} updated, ${removed} removed, ${unchanged} unchanged`,
-  );
-}
-
-await main();

packages/core/script/sync/google.ts

@@ -0,0 +1,138 @@
+import { z } from "zod";
+
+import type { ExistingModel, SyncProvider, SyncedModel } from "../sync-models.js";
+
+const API_ENDPOINT = "https://generativelanguage.googleapis.com/v1beta/models";
+
+const GoogleModel = z.object({
+  name: z.string(),
+  baseModelId: z.string().optional(),
+  version: z.string().optional(),
+  displayName: z.string().optional(),
+  description: z.string().optional(),
+  inputTokenLimit: z.number().int().nonnegative(),
+  outputTokenLimit: z.number().int().nonnegative(),
+  supportedGenerationMethods: z.array(z.string()).optional(),
+  temperature: z.number().optional(),
+  topP: z.number().optional(),
+  topK: z.number().optional(),
+  maxTemperature: z.number().optional(),
+  thinking: z.boolean().optional(),
+}).passthrough();
+
+const GoogleResponse = z.object({
+  models: z.array(GoogleModel).optional(),
+  nextPageToken: z.string().optional(),
+}).passthrough();
+
+type GoogleModel = z.infer<typeof GoogleModel>;
+
+export const google = {
+  id: "google",
+  name: "Google",
+  modelsDir: "providers/google/models",
+  skipCreates: true,
+  sourceID(model) {
+    return model.name.replace(/^models\//, "");
+  },
+  skippedNotice(ids) {
+    if (ids.length === 0) return [];
+    return [
+      `${ids.length} Google models returned by the API were not created because the Models API does not provide authoritative modalities, pricing, knowledge cutoff, release date, tool calling, or structured output metadata. Existing models are still updated from API-authoritative fields.`,
+      `Skipped remote IDs: ${ids.map((id) => `\`${id}\``).join(", ")}`,
+    ];
+  },
+  async fetchModels() {
+    const key = process.env.GOOGLE_API_KEY
+      ?? process.env.GEMINI_API_KEY
+      ?? process.env.GOOGLE_GENERATIVE_AI_API_KEY;
+    if (key === undefined) {
+      throw new Error("Google sync requires GOOGLE_API_KEY, GEMINI_API_KEY, or GOOGLE_GENERATIVE_AI_API_KEY");
+    }
+
+    const models: GoogleModel[] = [];
+    let pageToken: string | undefined;
+
+    do {
+      const url = new URL(API_ENDPOINT);
+      url.searchParams.set("key", key);
+      url.searchParams.set("pageSize", "1000");
+      if (pageToken !== undefined) url.searchParams.set("pageToken", pageToken);
+
+      const response = await fetch(url);
+      if (!response.ok) {
+        throw new Error(`Google models request failed: ${response.status} ${response.statusText}`);
+      }
+
+      const page = GoogleResponse.parse(await response.json());
+      models.push(...page.models ?? []);
+      pageToken = page.nextPageToken;
+    } while (pageToken !== undefined);
+
+    return { models };
+  },
+  parseModels(raw) {
+    return GoogleResponse.parse(raw).models ?? [];
+  },
+  translateModel(model, context) {
+    const id = model.name.replace(/^models\//, "");
+    const existing = context.existing(id);
+    if (existing === undefined) return undefined;
+
+    return {
+      id,
+      model: buildModel(model, existing),
+    };
+  },
+} satisfies SyncProvider<GoogleModel>;
+
+function buildModel(model: GoogleModel, existing: ExistingModel): SyncedModel {
+  const name = existing.name;
+  const releaseDate = existing.release_date;
+  const lastUpdated = existing.last_updated;
+  const attachment = existing.attachment;
+  const reasoning = existing.reasoning;
+  const toolCall = existing.tool_call;
+  const openWeights = existing.open_weights;
+  const limit = existing.limit;
+  const modalities = existing.modalities;
+
+  if (
+    name === undefined
+    || releaseDate === undefined
+    || lastUpdated === undefined
+    || attachment === undefined
+    || reasoning === undefined
+    || toolCall === undefined
+    || openWeights === undefined
+    || limit === undefined
+    || modalities === undefined
+  ) {
+    throw new Error(`Google model ${model.name} has incomplete local TOML metadata required for sync`);
+  }
+
+  return {
+    name: model.displayName ?? name,
+    family: existing.family,
+    release_date: releaseDate,
+    last_updated: lastUpdated,
+    attachment,
+    reasoning: model.thinking ?? reasoning,
+    temperature: model.temperature !== undefined || model.maxTemperature !== undefined
+      ? true
+      : existing.temperature,
+    tool_call: toolCall,
+    structured_output: existing.structured_output,
+    knowledge: existing.knowledge,
+    open_weights: openWeights,
+    status: existing.status,
+    interleaved: existing.interleaved,
+    cost: existing.cost,
+    limit: {
+      input: limit.input,
+      context: model.inputTokenLimit,
+      output: model.outputTokenLimit,
+    },
+    modalities,
+  };
+}

packages/core/script/sync/openrouter.ts

@@ -0,0 +1,144 @@
+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 },
+  };
+}

providers/amazon-bedrock/models/moonshot.kimi-k2-thinking.toml

@@ -1,3 +1,4 @@
+# See https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-thinking.html
 name = "Kimi K2 Thinking"
 release_date = "2025-12-02"
 family = "kimi-thinking"
@@ -15,8 +16,8 @@ input = 0.6
 output = 2.5
 
 [limit]
-context = 256_000
-output = 256_000
+context = 262_143
+output = 16_000
 
 [modalities]
 input = ["text"]

providers/amazon-bedrock/models/moonshotai.kimi-k2.5.toml

@@ -1,3 +1,4 @@
+# See https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-moonshot-ai-kimi-k2-5.html
 name = "Kimi K2.5"
 family = "kimi"
 release_date = "2026-02-06"
@@ -15,8 +16,8 @@ input = 0.6
 output = 3
 
 [limit]
-context = 256_000
-output = 256_000
+context = 262_143
+output = 16_000
 
 [modalities]
 input = ["text", "image"]

providers/azure/models/grok-3-mini.toml

@@ -1,2 +0,0 @@
-[extends]
-from = "xai/grok-3-mini"

providers/azure/models/grok-3.toml

@@ -1,2 +0,0 @@
-[extends]
-from = "xai/grok-3"

providers/azure/models/grok-4-fast-non-reasoning.toml

@@ -1,2 +0,0 @@
-[extends]
-from = "xai/grok-4-fast-non-reasoning"

providers/azure/models/grok-4.toml

@@ -1,2 +0,0 @@
-[extends]
-from = "xai/grok-4"

providers/azure/models/grok-code-fast-1.toml

@@ -1,2 +0,0 @@
-[extends]
-from = "xai/grok-code-fast-1"

providers/berget/models/moonshotai/Kimi-K2.6.toml

@@ -0,0 +1,9 @@
+release_date = "2026-05-07"
+last_updated = "2026-05-07"
+
+[extends]
+from = "moonshotai/kimi-k2.6"
+
+[cost]
+input = 0.83
+output = 3.85

providers/fireworks-ai/models/accounts/fireworks/models/deepseek-v4-flash.toml

@@ -0,0 +1,7 @@
+[extends]
+from = "deepseek/deepseek-v4-flash"
+
+[cost]
+input = 0.14
+output = 0.28
+cache_read = 0.03

providers/gitlab/models/duo-chat-gpt-5-5.toml

@@ -0,0 +1,24 @@
+name = "Agentic Chat (GPT-5.5)"
+family = "gpt"
+release_date = "2026-04-23"
+last_updated = "2026-04-23"
+knowledge = "2025-08-31"
+attachment = true
+reasoning = true
+temperature = false
+tool_call = true
+structured_output = true
+open_weights = false
+
+[cost]
+input = 0
+output = 0
+
+[limit]
+context = 1_050_000
+input = 922_000
+output = 128_000
+
+[modalities]
+input = ["text", "image", "pdf"]
+output = ["text"]

providers/google-vertex/models/gemini-2.5-flash-lite-preview-09-2025.toml

@@ -1,3 +0,0 @@
-[extends]
-from = "google/gemini-2.5-flash-lite-preview-09-2025"
-omit = ["structured_output"]

providers/google-vertex/models/gemini-2.5-flash-preview-04-17.toml

@@ -1,2 +0,0 @@
-[extends]
-from = "google/gemini-2.5-flash-preview-04-17"

providers/google-vertex/models/gemini-2.5-flash-preview-05-20.toml

@@ -1,3 +0,0 @@
-[extends]
-from = "google/gemini-2.5-flash-preview-05-20"
-omit = ["structured_output"]

providers/google-vertex/models/gemini-2.5-pro-preview-05-06.toml

@@ -1,3 +0,0 @@
-[extends]
-from = "google/gemini-2.5-pro-preview-05-06"
-omit = ["structured_output"]

providers/google-vertex/models/gemini-2.5-pro-preview-06-05.toml

@@ -1,3 +0,0 @@
-[extends]
-from = "google/gemini-2.5-pro-preview-06-05"
-omit = ["structured_output"]

providers/google/models/gemini-1.5-flash-8b.toml

@@ -1,23 +0,0 @@
-name = "Gemini 1.5 Flash-8B"
-family = "gemini-flash"
-release_date = "2024-10-03"
-last_updated = "2024-10-03"
-attachment = true
-reasoning = false
-temperature = true
-knowledge = "2024-04"
-tool_call = true
-open_weights = false
-
-[cost]
-input = 0.0375
-output = 0.15
-cache_read = 0.01
-
-[limit]
-context = 1_000_000
-output = 8_192
-
-[modalities]
-input = ["text", "image", "audio", "video"]
-output = ["text"]

providers/google/models/gemini-1.5-flash.toml

@@ -1,23 +0,0 @@
-name = "Gemini 1.5 Flash"
-family = "gemini-flash"
-release_date = "2024-05-14"
-last_updated = "2024-05-14"
-attachment = true
-reasoning = false
-temperature = true
-knowledge = "2024-04"
-tool_call = true
-open_weights = false
-
-[cost]
-input = 0.075
-output = 0.30
-cache_read = 0.01875
-
-[limit]
-context = 1_000_000
-output = 8_192
-
-[modalities]
-input = ["text", "image", "audio", "video"]
-output = ["text"]

providers/google/models/gemini-1.5-pro.toml

@@ -1,23 +0,0 @@
-name = "Gemini 1.5 Pro"
-family = "gemini-pro"
-release_date = "2024-02-15"
-last_updated = "2024-02-15"
-attachment = true
-reasoning = false
-temperature = true
-knowledge = "2024-04"
-tool_call = true
-open_weights = false
-
-[cost]
-input = 1.25
-output = 5.00
-cache_read = 0.3125
-
-[limit]
-context = 1_000_000
-output = 8_192
-
-[modalities]
-input = ["text", "image", "audio", "video"]
-output = ["text"]

providers/google/models/gemini-2.0-flash-lite.toml

@@ -1,18 +1,18 @@
-name = "Gemini 2.0 Flash Lite"
+name = "Gemini 2.0 Flash-Lite"
 family = "gemini-flash-lite"
 release_date = "2024-12-11"
 last_updated = "2024-12-11"
 attachment = true
 reasoning = false
 temperature = true
-knowledge = "2024-06"
 tool_call = true
 structured_output = true
+knowledge = "2024-06"
 open_weights = false
 
 [cost]
 input = 0.075
-output = 0.30
+output = 0.3
 
 [limit]
 context = 1_048_576

providers/google/models/gemini-2.5-flash-image-preview.toml

@@ -1,23 +0,0 @@
-name = "Gemini 2.5 Flash Image (Preview)"
-family = "gemini-flash"
-release_date = "2025-08-26"
-last_updated = "2025-08-26"
-attachment = true
-reasoning = true
-temperature = true
-knowledge = "2025-06"
-tool_call = false
-open_weights = false
-
-[cost]
-input = 0.30
-output = 30
-cache_read = 0.075
-
-[limit]
-context = 32_768
-output = 32_768
-
-[modalities]
-input = ["text", "image"]
-output = ["text", "image"]

providers/google/models/gemini-2.5-flash-image.toml

@@ -1,16 +1,16 @@
-name = "Gemini 2.5 Flash Image"
+name = "Nano Banana"
 family = "gemini-flash"
 release_date = "2025-08-26"
 last_updated = "2025-08-26"
 attachment = true
 reasoning = true
 temperature = true
-knowledge = "2025-06"
 tool_call = false
+knowledge = "2025-06"
 open_weights = false
 
 [cost]
-input = 0.30
+input = 0.3
 output = 30
 cache_read = 0.075
 
@@ -20,4 +20,4 @@ output = 32_768
 
 [modalities]
 input = ["text", "image"]
-output = ["text", "image"]
+output = ["text", "image"]

providers/google/models/gemini-2.5-flash-lite-preview-06-17.toml

@@ -1,24 +0,0 @@
-name = "Gemini 2.5 Flash Lite Preview 06-17"
-family = "gemini-flash-lite"
-release_date = "2025-06-17"
-last_updated = "2025-06-17"
-attachment = true
-reasoning = true
-temperature = true
-knowledge = "2025-01"
-tool_call = true
-open_weights = false
-
-[cost]
-input = 0.10
-output = 0.40
-cache_read = 0.025
-input_audio = 0.30
-
-[limit]
-context = 1_048_576
-output = 65_536
-
-[modalities]
-input = ["text", "image", "audio", "video", "pdf"]
-output = ["text"]

providers/google/models/gemini-2.5-flash-lite-preview-09-2025.toml

@@ -1,24 +0,0 @@
-name = "Gemini 2.5 Flash Lite Preview 09-25"
-family = "gemini-flash-lite"
-release_date = "2025-09-25"
-last_updated = "2025-09-25"
-attachment = true
-reasoning = true
-temperature = true
-knowledge = "2025-01"
-tool_call = true
-structured_output = true
-open_weights = false
-
-[cost]
-input = 0.10
-output = 0.40
-cache_read = 0.025
-
-[limit]
-context = 1_048_576
-output = 65_536
-
-[modalities]
-input = ["text", "image", "audio", "video", "pdf"]
-output = ["text"]

providers/google/models/gemini-2.5-flash-lite.toml

@@ -1,20 +1,20 @@
-name = "Gemini 2.5 Flash Lite"
+name = "Gemini 2.5 Flash-Lite"
 family = "gemini-flash-lite"
 release_date = "2025-06-17"
 last_updated = "2025-06-17"
 attachment = true
 reasoning = true
 temperature = true
-knowledge = "2025-01"
 tool_call = true
 structured_output = true
+knowledge = "2025-01"
 open_weights = false
 
 [cost]
-input = 0.10
-output = 0.40
+input = 0.1
+output = 0.4
 cache_read = 0.01
-input_audio = 0.30
+input_audio = 0.3
 
 [limit]
 context = 1_048_576

providers/google/models/gemini-2.5-flash-preview-04-17.toml

@@ -1,23 +0,0 @@
-name = "Gemini 2.5 Flash Preview 04-17"
-family = "gemini-flash"
-release_date = "2025-04-17"
-last_updated = "2025-04-17"
-attachment = true
-reasoning = true
-temperature = true
-knowledge = "2025-01"
-tool_call = true
-open_weights = false
-
-[cost]
-input = 0.15
-output = 0.60
-cache_read = 0.0375
-
-[limit]
-context = 1_048_576
-output = 65_536
-
-[modalities]
-input = ["text", "image", "audio", "video", "pdf"]
-output = ["text"]

providers/google/models/gemini-2.5-flash-preview-09-2025.toml

@@ -1,25 +0,0 @@
-name = "Gemini 2.5 Flash Preview 09-25"
-family = "gemini-flash"
-release_date = "2025-09-25"
-last_updated = "2025-09-25"
-attachment = true
-reasoning = true
-temperature = true
-knowledge = "2025-01"
-tool_call = true
-structured_output = true
-open_weights = false
-
-[cost]
-input = 0.30
-output = 2.50
-cache_read = 0.075
-input_audio = 1.00
-
-[limit]
-context = 1_048_576
-output = 65_536
-
-[modalities]
-input = ["text", "image", "audio", "video", "pdf"]
-output = ["text"]

providers/google/models/gemini-2.5-flash-preview-tts.toml

@@ -4,19 +4,19 @@ release_date = "2025-05-01"
 last_updated = "2025-05-01"
 attachment = false
 reasoning = false
-temperature = false
-knowledge = "2025-01"
+temperature = true
 tool_call = false
+knowledge = "2025-01"
 open_weights = false
 
 [cost]
-input = 0.50
-output = 10.00
+input = 0.5
+output = 10
 
 [limit]
-context = 8_000
-output = 16_000
+context = 8_192
+output = 16_384
 
 [modalities]
 input = ["text"]
-output = ["audio"]
+output = ["audio"]

providers/google/models/gemini-2.5-pro-preview-05-06.toml

@@ -1,24 +0,0 @@
-name = "Gemini 2.5 Pro Preview 05-06"
-family = "gemini-pro"
-release_date = "2025-05-06"
-last_updated = "2025-05-06"
-attachment = true
-reasoning = true
-temperature = true
-knowledge = "2025-01"
-tool_call = true
-structured_output = true
-open_weights = false
-
-[cost]
-input = 1.25
-output = 10.00
-cache_read = 0.31
-
-[limit]
-context = 1_048_576
-output = 65_536
-
-[modalities]
-input = ["text", "image", "audio", "video", "pdf"]
-output = ["text"]

providers/google/models/gemini-2.5-pro-preview-06-05.toml

@@ -1,24 +0,0 @@
-name = "Gemini 2.5 Pro Preview 06-05"
-family = "gemini-pro"
-release_date = "2025-06-05"
-last_updated = "2025-06-05"
-attachment = true
-reasoning = true
-temperature = true
-knowledge = "2025-01"
-tool_call = true
-structured_output = true
-open_weights = false
-
-[cost]
-input = 1.25
-output = 10.00
-cache_read = 0.31
-
-[limit]
-context = 1_048_576
-output = 65_536
-
-[modalities]
-input = ["text", "image", "audio", "video", "pdf"]
-output = ["text"]

providers/google/models/gemini-2.5-pro-preview-tts.toml

@@ -4,19 +4,19 @@ release_date = "2025-05-01"
 last_updated = "2025-05-01"
 attachment = false
 reasoning = false
-temperature = false
-knowledge = "2025-01"
+temperature = true
 tool_call = false
+knowledge = "2025-01"
 open_weights = false
 
 [cost]
-input = 1.00
-output = 20.00
+input = 1
+output = 20
 
 [limit]
-context = 8_000
-output = 16_000
+context = 8_192
+output = 16_384
 
 [modalities]
 input = ["text"]
-output = ["audio"]
+output = ["audio"]

providers/google/models/gemini-3-pro-preview.toml

@@ -11,19 +11,19 @@ knowledge = "2025-01"
 open_weights = false
 
 [cost]
-input = 2.00
-output = 12.00
-cache_read = 0.20
+input = 2
+output = 12
+cache_read = 0.2
 
 [[cost.tiers]]
 tier = { size = 200_000 }
-input = 4.00
-output = 18.00
-cache_read = 0.40
+input = 4
+output = 18
+cache_read = 0.4
 
 [limit]
-context = 1000000
-output = 64000
+context = 1_048_576
+output = 65_536
 
 [modalities]
 input = ["text", "image", "video", "audio", "pdf"]

providers/google/models/gemini-3.1-flash-image-preview.toml

@@ -1,21 +1,21 @@
-name = "Gemini 3.1 Flash Image (Preview)"
+name = "Nano Banana 2"
 family = "gemini-flash"
 release_date = "2026-02-26"
 last_updated = "2026-02-26"
 attachment = true
 reasoning = true
 temperature = true
-knowledge = "2025-01"
 tool_call = false
+knowledge = "2025-01"
 open_weights = false
 
 [cost]
-input = 0.50
-output = 60.00
+input = 0.5
+output = 60
 
 [limit]
-context = 131_072
-output = 32_768
+context = 65_536
+output = 65_536
 
 [modalities]
 input = ["text", "image", "pdf"]

providers/google/models/gemini-3.5-flash.toml

@@ -1,24 +1,25 @@
-name = "Gemini 2.5 Flash Preview 05-20"
+name = "Gemini 3.5 Flash"
 family = "gemini-flash"
-release_date = "2025-05-20"
-last_updated = "2025-05-20"
+release_date = "2026-05-19"
+last_updated = "2026-05-19"
 attachment = true
 reasoning = true
 temperature = true
-knowledge = "2025-01"
 tool_call = true
 structured_output = true
+knowledge = "2025-01"
 open_weights = false
 
 [cost]
-input = 0.15
-output = 0.60
-cache_read = 0.0375
+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", "audio", "video", "pdf"]
+input = ["text", "image", "video", "audio", "pdf"]
 output = ["text"]

providers/google/models/gemini-embedding-001.toml

@@ -5,17 +5,17 @@ last_updated = "2025-05-20"
 attachment = false
 reasoning = false
 temperature = false
-knowledge = "2025-05"
 tool_call = false
+knowledge = "2025-05"
 open_weights = false
 
 [cost]
 input = 0.15
-output = 0.00
+output = 0
 
 [limit]
 context = 2_048
-output = 3_072
+output = 1
 
 [modalities]
 input = ["text"]

providers/google/models/gemini-live-2.5-flash-preview-native-audio.toml

@@ -1,24 +0,0 @@
-name = "Gemini Live 2.5 Flash Preview Native Audio"
-family = "gemini-flash"
-release_date = "2025-06-17"
-last_updated = "2025-09-18"
-attachment = false
-reasoning = true
-temperature = false
-knowledge = "2025-01"
-tool_call = true
-open_weights = false
-
-[cost]
-input = 0.50
-output = 2.00
-input_audio = 3.00
-output_audio = 12.00
-
-[limit]
-context = 131_072
-output = 65_536
-
-[modalities]
-input = ["text", "audio", "video"]
-output = ["text", "audio"]

providers/google/models/gemini-live-2.5-flash.toml

@@ -1,24 +0,0 @@
-name = "Gemini Live 2.5 Flash"
-family = "gemini-flash"
-release_date = "2025-09-01"
-last_updated = "2025-09-01"
-attachment = true
-reasoning = true
-temperature = true
-knowledge = "2025-01"
-tool_call = true
-open_weights = false
-
-[cost]
-input = 0.50
-output = 2.00
-input_audio = 3.00
-output_audio = 12.00
-
-[limit]
-context = 128_000
-output = 8_000
-
-[modalities]
-input = ["text", "image", "audio", "video"]
-output = ["text", "audio"]

providers/google/models/gemma-3-12b-it.toml

@@ -1,23 +0,0 @@
-name = "Gemma 3 12B"
-family = "gemma"
-release_date = "2025-03-13"
-last_updated = "2025-03-13"
-attachment = true
-reasoning = false
-temperature = true
-knowledge = "2024-10"
-tool_call = false
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0
-output = 0
-
-[limit]
-context = 32_768
-output = 8_192
-
-[modalities]
-input = ["text", "image"]
-output = ["text"]

providers/google/models/gemma-3-27b-it.toml

@@ -1,23 +0,0 @@
-name = "Gemma 3 27B"
-family = "gemma"
-release_date = "2025-03-12"
-last_updated = "2025-03-12"
-attachment = true
-reasoning = false
-temperature = true
-knowledge = "2024-10"
-tool_call = true
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0
-output = 0
-
-[limit]
-context = 131_072
-output = 8_192
-
-[modalities]
-input = ["text", "image"]
-output = ["text"]

providers/google/models/gemma-3-4b-it.toml

@@ -1,22 +0,0 @@
-name = "Gemma 3 4B"
-family = "gemma"
-release_date = "2025-03-13"
-last_updated = "2025-03-13"
-attachment = true
-reasoning = false
-temperature = true
-knowledge = "2024-10"
-tool_call = false
-open_weights = true
-
-[cost]
-input = 0
-output = 0
-
-[limit]
-context = 32_768
-output = 8_192
-
-[modalities]
-input = ["text", "image"]
-output = ["text"]

providers/google/models/gemma-3n-e2b-it.toml

@@ -1,22 +0,0 @@
-name = "Gemma 3n 2B"
-family = "gemma"
-release_date = "2025-07-09"
-last_updated = "2025-07-09"
-attachment = true
-reasoning = false
-temperature = true
-knowledge = "2024-10"
-tool_call = false
-open_weights = true
-
-[cost]
-input = 0
-output = 0
-
-[limit]
-context = 8_192
-output = 2_000
-
-[modalities]
-input = ["text"]
-output = ["text"]

providers/google/models/gemma-3n-e4b-it.toml

@@ -1,22 +0,0 @@
-name = "Gemma 3n 4B"
-family = "gemma"
-release_date = "2025-05-20"
-last_updated = "2025-05-20"
-attachment = true
-reasoning = false
-temperature = true
-knowledge = "2024-10"
-tool_call = false
-open_weights = true
-
-[cost]  
-input = 0
-output = 0
-
-[limit]
-context = 8_192
-output = 2_000
-
-[modalities]
-input = ["text"]
-output = ["text"]

providers/google/models/gemma-4-26b-a4b-it.toml

@@ -1,8 +1,8 @@
-name = "Gemma 4 26B"
+name = "Gemma 4 26B A4B IT"
 family = "gemma"
 release_date = "2026-04-02"
 last_updated = "2026-04-02"
-attachment = false
+attachment = true
 reasoning = true
 temperature = true
 tool_call = true
@@ -10,8 +10,8 @@ structured_output = true
 open_weights = true
 
 [limit]
-context = 256000
-output = 8192
+context = 262_144
+output = 32_768
 
 [modalities]
 input = ["text", "image"]

providers/google/models/gemma-4-31b-it.toml

@@ -1,8 +1,8 @@
-name = "Gemma 4 31B"
+name = "Gemma 4 31B IT"
 family = "gemma"
 release_date = "2026-04-02"
 last_updated = "2026-04-02"
-attachment = false
+attachment = true
 reasoning = true
 temperature = true
 tool_call = true
@@ -10,8 +10,8 @@ structured_output = true
 open_weights = true
 
 [limit]
-context = 256000
-output = 8192
+context = 262_144
+output = 32_768
 
 [modalities]
 input = ["text", "image"]

providers/google/provider.toml

@@ -1,4 +1,4 @@
 name = "Google"
-env = ["GOOGLE_GENERATIVE_AI_API_KEY", "GEMINI_API_KEY"]
+env = ["GOOGLE_API_KEY", "GOOGLE_GENERATIVE_AI_API_KEY", "GEMINI_API_KEY"]
 npm = "@ai-sdk/google"
 doc = "https://ai.google.dev/gemini-api/docs/models"

providers/kilo/models/alpindale/goliath-120b.toml

@@ -1,20 +0,0 @@
-name = "Goliath 120B"
-release_date = "2023-11-10"
-last_updated = "2026-03-15"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = false
-open_weights = true
-
-[cost]
-input = 3.75
-output = 7.5
-
-[limit]
-context = 6144
-output = 1024
-
-[modalities]
-input = ["text"]
-output = ["text"]

providers/kilo/models/anthropic/claude-3.7-sonnet.toml

@@ -1,22 +0,0 @@
-name = "Anthropic: Claude 3.7 Sonnet"
-release_date = "2025-02-19"
-last_updated = "2026-03-15"
-attachment = true
-reasoning = true
-temperature = true
-tool_call = true
-open_weights = false
-
-[cost]
-input = 3
-output = 15
-cache_read = 0.3
-cache_write = 3.75
-
-[limit]
-context = 200000
-output = 64000
-
-[modalities]
-input = ["image", "pdf", "text"]
-output = ["text"]

providers/kilo/models/anthropic/claude-3.7-sonnet:thinking.toml

@@ -1,22 +0,0 @@
-name = "Anthropic: Claude 3.7 Sonnet (thinking)"
-release_date = "2025-02-19"
-last_updated = "2026-03-15"
-attachment = true
-reasoning = true
-temperature = true
-tool_call = true
-open_weights = false
-
-[cost]
-input = 3
-output = 15
-cache_read = 0.3
-cache_write = 3.75
-
-[limit]
-context = 200000
-output = 64000
-
-[modalities]
-input = ["image", "pdf", "text"]
-output = ["text"]

providers/kilo/models/anthropic/claude-opus-4.7-fast.toml

@@ -0,0 +1,22 @@
+name = "Anthropic: Claude Opus 4.7 (Fast)"
+release_date = "2026-05-12"
+last_updated = "2026-05-16"
+attachment = true
+reasoning = true
+tool_call = true
+temperature = false
+open_weights = false
+
+[cost]
+input = 30.0
+output = 150.0
+cache_read = 3.0
+cache_write = 37.5
+
+[limit]
+context = 1000000
+output = 128000
+
+[modalities]
+input = ["image", "pdf", "text"]
+output = ["text"]

providers/kilo/models/baidu/qianfan-ocr-fast.toml

@@ -1,6 +1,6 @@
-name = "Baidu: Qianfan-OCR-Fast (free)"
+name = "Baidu: Qianfan-OCR-Fast"
 release_date = "2026-04-20"
-last_updated = "2026-05-01"
+last_updated = "2026-05-16"
 attachment = true
 reasoning = true
 tool_call = false
@@ -8,8 +8,8 @@ temperature = true
 open_weights = false
 
 [cost]
-input = 0.0
-output = 0.0
+input = 0.68
+output = 2.81
 
 [limit]
 context = 65536

providers/kilo/models/deepseek/deepseek-v4-flash:free.toml

@@ -0,0 +1,20 @@
+name = "DeepSeek: DeepSeek V4 Flash (free)"
+release_date = "2026-04-24"
+last_updated = "2026-05-16"
+attachment = false
+reasoning = true
+tool_call = true
+temperature = false
+open_weights = false
+
+[cost]
+input = 0
+output = 0
+
+[limit]
+context = 1048576
+output = 384000
+
+[modalities]
+input = ["text"]
+output = ["text"]

providers/kilo/models/google/gemini-3.1-flash-lite.toml

@@ -0,0 +1,23 @@
+name = "Google: Gemini 3.1 Flash Lite"
+release_date = "2026-05-07"
+last_updated = "2026-05-16"
+attachment = true
+reasoning = true
+tool_call = true
+temperature = true
+open_weights = false
+
+[cost]
+input = 0.25
+output = 1.5
+reasoning = 1.5
+cache_read = 0.025
+cache_write = 0.08333
+
+[limit]
+context = 1048576
+output = 65536
+
+[modalities]
+input = ["audio", "image", "pdf", "text", "video"]
+output = ["text"]

providers/kilo/models/inclusionai/ling-2.6-1t.toml

@@ -1,6 +1,6 @@
-name = "inclusionAI: Ling-2.6-1T (free)"
+name = "inclusionAI: Ling-2.6-1T"
 release_date = "2026-04-23"
-last_updated = "2026-05-01"
+last_updated = "2026-05-16"
 attachment = false
 reasoning = false
 tool_call = true
@@ -8,8 +8,9 @@ temperature = true
 open_weights = false
 
 [cost]
-input = 0.0
-output = 0.0
+input = 0.3
+output = 2.5
+cache_read = 0.06
 
 [limit]
 context = 262144

providers/kilo/models/inclusionai/ring-2.6-1t.toml

@@ -0,0 +1,21 @@
+name = "inclusionAI: Ring-2.6-1T"
+release_date = "2026-05-08"
+last_updated = "2026-05-16"
+attachment = false
+reasoning = true
+tool_call = true
+temperature = true
+open_weights = false
+
+[cost]
+input = 0.075
+output = 0.625
+cache_read = 0.015
+
+[limit]
+context = 262144
+output = 65536
+
+[modalities]
+input = ["text"]
+output = ["text"]

providers/kilo/models/mistralai/mixtral-8x7b-instruct.toml

@@ -1,20 +0,0 @@
-name = "Mistral: Mixtral 8x7B Instruct"
-release_date = "2023-12-10"
-last_updated = "2026-03-15"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = true
-open_weights = true
-
-[cost]
-input = 0.54
-output = 0.54
-
-[limit]
-context = 32768
-output = 16384
-
-[modalities]
-input = ["text"]
-output = ["text"]

providers/kilo/models/nvidia/llama-3.1-nemotron-70b-instruct.toml

@@ -1,20 +0,0 @@
-name = "NVIDIA: Llama 3.1 Nemotron 70B Instruct"
-release_date = "2024-10-12"
-last_updated = "2024-10-12"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = true
-open_weights = false
-
-[cost]
-input = 1.2
-output = 1.2
-
-[limit]
-context = 131072
-output = 16384
-
-[modalities]
-input = ["text"]
-output = ["text"]

providers/kilo/models/perceptron/perceptron-mk1.toml

@@ -0,0 +1,20 @@
+name = "Perceptron: Perceptron Mk1"
+release_date = "2026-05-12"
+last_updated = "2026-05-16"
+attachment = true
+reasoning = true
+tool_call = false
+temperature = true
+open_weights = false
+
+[cost]
+input = 0.15
+output = 1.5
+
+[limit]
+context = 32768
+output = 8192
+
+[modalities]
+input = ["image", "text", "video"]
+output = ["text"]

providers/kilo/models/qwen/qwen-max.toml

@@ -1,21 +0,0 @@
-name = "Qwen: Qwen-Max "
-release_date = "2024-04-03"
-last_updated = "2026-03-15"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = true
-open_weights = false
-
-[cost]
-input = 1.04
-output = 4.16
-cache_read = 0.32
-
-[limit]
-context = 32768
-output = 8192
-
-[modalities]
-input = ["text"]
-output = ["text"]

providers/kilo/models/qwen/qwen-turbo.toml

@@ -1,21 +0,0 @@
-name = "Qwen: Qwen-Turbo"
-release_date = "2024-11-01"
-last_updated = "2026-03-15"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = true
-open_weights = false
-
-[cost]
-input = 0.0325
-output = 0.13
-cache_read = 0.01
-
-[limit]
-context = 131072
-output = 8192
-
-[modalities]
-input = ["text"]
-output = ["text"]

providers/kilo/models/qwen/qwen-vl-max.toml

@@ -1,20 +0,0 @@
-name = "Qwen: Qwen VL Max"
-release_date = "2024-04-08"
-last_updated = "2025-08-13"
-attachment = true
-reasoning = false
-temperature = true
-tool_call = true
-open_weights = false
-
-[cost]
-input = 0.8
-output = 3.2
-
-[limit]
-context = 131072
-output = 32768
-
-[modalities]
-input = ["text", "image"]
-output = ["text"]

providers/kilo/models/qwen/qwen-vl-plus.toml

@@ -1,21 +0,0 @@
-name = "Qwen: Qwen VL Plus"
-release_date = "2024-01-25"
-last_updated = "2026-03-15"
-attachment = true
-reasoning = false
-temperature = true
-tool_call = false
-open_weights = false
-
-[cost]
-input = 0.1365
-output = 0.4095
-cache_read = 0.042
-
-[limit]
-context = 131072
-output = 8192
-
-[modalities]
-input = ["image", "text"]
-output = ["text"]

providers/kilo/models/tencent/hy3-preview.toml

@@ -1,6 +1,6 @@
-name = "Tencent: Hy3 Preview (free)"
+name = "Tencent: Hy3 Preview"
 release_date = "2026-04-22"
-last_updated = "2026-05-01"
+last_updated = "2026-05-16"
 attachment = false
 reasoning = true
 tool_call = true
@@ -8,8 +8,9 @@ temperature = true
 open_weights = false
 
 [cost]
-input = 0.0
-output = 0.0
+input = 0.066
+output = 0.26
+cache_read = 0.029
 
 [limit]
 context = 262144

providers/kilo/models/tngtech/deepseek-r1t2-chimera.toml

@@ -1,21 +0,0 @@
-name = "TNG: DeepSeek R1T2 Chimera"
-release_date = "2025-07-08"
-last_updated = "2025-07-08"
-attachment = false
-reasoning = true
-temperature = true
-tool_call = true
-open_weights = true
-
-[cost]
-input = 0.25
-output = 0.85
-cache_read = 0.125
-
-[limit]
-context = 163840
-output = 163840
-
-[modalities]
-input = ["text"]
-output = ["text"]

providers/kilo/models/x-ai/grok-3-beta.toml

@@ -1,21 +0,0 @@
-name = "xAI: Grok 3 Beta"
-release_date = "2025-02-17"
-last_updated = "2025-02-17"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = true
-open_weights = false
-
-[cost]
-input = 3.0
-output = 15.0
-cache_read = 0.75
-
-[limit]
-context = 131072
-output = 26215
-
-[modalities]
-input = ["text"]
-output = ["text"]

providers/kilo/models/x-ai/grok-3-mini-beta.toml

@@ -1,21 +0,0 @@
-name = "xAI: Grok 3 Mini Beta"
-release_date = "2025-02-17"
-last_updated = "2025-02-17"
-attachment = false
-reasoning = true
-temperature = true
-tool_call = true
-open_weights = false
-
-[cost]
-input = 0.3
-output = 0.5
-cache_read = 0.075
-
-[limit]
-context = 131072
-output = 26215
-
-[modalities]
-input = ["text"]
-output = ["text"]

providers/kilo/models/x-ai/grok-3-mini.toml

@@ -1,21 +0,0 @@
-name = "xAI: Grok 3 Mini"
-release_date = "2025-02-17"
-last_updated = "2025-02-17"
-attachment = false
-reasoning = true
-temperature = true
-tool_call = true
-open_weights = false
-
-[cost]
-input = 0.3
-output = 0.5
-cache_read = 0.075
-
-[limit]
-context = 131072
-output = 26215
-
-[modalities]
-input = ["text"]
-output = ["text"]

providers/kilo/models/x-ai/grok-3.toml

@@ -1,21 +0,0 @@
-name = "xAI: Grok 3"
-release_date = "2025-02-17"
-last_updated = "2025-02-17"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = true
-open_weights = false
-
-[cost]
-input = 3.0
-output = 15.0
-cache_read = 0.75
-
-[limit]
-context = 131072
-output = 26215
-
-[modalities]
-input = ["text"]
-output = ["text"]

providers/kilo/models/x-ai/grok-4-fast.toml

@@ -1,21 +0,0 @@
-name = "xAI: Grok 4 Fast"
-release_date = "2025-08-19"
-last_updated = "2025-08-19"
-attachment = true
-reasoning = true
-temperature = true
-tool_call = true
-open_weights = false
-
-[cost]
-input = 0.2
-output = 0.5
-cache_read = 0.05
-
-[limit]
-context = 2000000
-output = 30000
-
-[modalities]
-input = ["text", "image"]
-output = ["text"]

providers/kilo/models/x-ai/grok-4.1-fast.toml

@@ -1,21 +0,0 @@
-name = "xAI: Grok 4.1 Fast"
-release_date = "2025-11-19"
-last_updated = "2025-11-19"
-attachment = true
-reasoning = true
-temperature = true
-tool_call = true
-open_weights = false
-
-[cost]
-input = 0.2
-output = 0.5
-cache_read = 0.05
-
-[limit]
-context = 2000000
-output = 30000
-
-[modalities]
-input = ["text", "image"]
-output = ["text"]

providers/kilo/models/x-ai/grok-4.toml

@@ -1,21 +0,0 @@
-name = "xAI: Grok 4"
-release_date = "2025-07-09"
-last_updated = "2025-07-09"
-attachment = true
-reasoning = true
-temperature = true
-tool_call = true
-open_weights = false
-
-[cost]
-input = 3.0
-output = 15.0
-cache_read = 0.75
-
-[limit]
-context = 256000
-output = 51200
-
-[modalities]
-input = ["image", "text"]
-output = ["text"]

providers/kilo/models/x-ai/grok-code-fast-1.toml

@@ -1,21 +0,0 @@
-name = "xAI: Grok Code Fast 1"
-release_date = "2025-08-26"
-last_updated = "2025-08-26"
-attachment = false
-reasoning = true
-temperature = true
-tool_call = true
-open_weights = false
-
-[cost]
-input = 0.2
-output = 1.5
-cache_read = 0.02
-
-[limit]
-context = 256000
-output = 10000
-
-[modalities]
-input = ["text"]
-output = ["text"]

providers/lilac/logo.svg

@@ -0,0 +1,6 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 510 510" fill="currentColor">
+  <path d="M316.18,14.51l-20.5,34.9c-8.66,14.74-24.47,23.79-41.56,23.79h0c-17.09,0-32.91-9.05-41.56-23.79l-20.5-34.9c-15.98-27.2-53.34-31.98-75.65-9.67l-7.48,7.48c-18.82,18.82-18.82,49.35,0,68.17l81.74,81.74c9.04,9.04,21.3,14.12,34.09,14.12h58.73c12.78,0,25.05-5.08,34.09-14.12l81.74-81.74c18.82-18.82,18.82-49.35,0-68.17l-7.48-7.48c-22.31-22.31-59.67-17.53-75.65,9.67Z"/>
+  <path d="M192.06,495.81l20.5-34.9c8.66-14.74,24.47-23.79,41.56-23.79h0c17.09,0,32.91,9.05,41.56,23.79l20.5,34.9c15.98,27.2,53.34,31.98,75.65,9.67l7.48-7.48c18.82-18.82,18.82-49.35,0-68.17l-81.74-81.74c-9.04-9.04-21.3-14.12-34.09-14.12h-58.73c-12.78,0-25.05,5.08-34.09,14.12l-81.74,81.74c-18.82,18.82-18.82,49.35,0,68.17l7.48,7.48c22.31,22.31,59.67,17.53,75.65-9.67Z"/>
+  <path d="M494.77,317.23l-34.9-20.5c-14.74-8.66-23.79-24.47-23.79-41.56h0c0-17.09,9.05-32.91,23.79-41.56l34.9-20.5c27.2-15.98,31.98-53.34,9.67-75.65l-7.48-7.48c-18.82-18.82-49.35-18.82-68.17,0l-81.74,81.74c-9.04,9.04-14.12,21.3-14.12,34.09v58.73c0,12.78,5.08,25.05,14.12,34.09l81.74,81.74c18.82,18.82,49.35,18.82,68.17,0l7.48-7.48c22.31-22.31,17.53-59.67-9.67-75.65Z"/>
+  <path d="M13.47,193.1l34.9,20.5c14.74,8.66,23.79,24.47,23.79,41.56h0c0,17.09-9.05,32.91-23.79,41.56l-34.9,20.5c-27.2,15.98-31.98,53.34-9.67,75.65l7.48,7.48c18.82,18.82,49.35,18.82,68.17,0l81.74-81.74c9.04-9.04,14.12-21.3,14.12-34.09v-58.73c0-12.78-5.08-25.05-14.12-34.09l-81.74-81.74c-18.82-18.82-49.35-18.82-68.17,0l-7.48,7.48c-22.31,22.31-17.53,59.67,9.67,75.65Z"/>
+</svg>

providers/lilac/models/google/gemma-4-31b-it.toml

@@ -0,0 +1,21 @@
+name = "Gemma 4 31B IT"
+attachment = true
+knowledge = "2025-01"
+
+[extends]
+from = "google/gemma-4-31b-it"
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 0.11
+output = 0.35
+
+[limit]
+context = 262_100
+output = 262_100
+
+[modalities]
+input = ["text", "image", "video"]
+output = ["text"]

providers/lilac/models/minimaxai/minimax-m2.7.toml

@@ -0,0 +1,23 @@
+name = "MiniMax M2.7"
+structured_output = true
+knowledge = "2025-01"
+
+[extends]
+from = "minimax/MiniMax-M2.7"
+omit = ["cost.cache_write"]
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 0.30
+output = 1.20
+cache_read = 0.055
+
+[limit]
+context = 204_800
+output = 204_800
+
+[modalities]
+input = ["text"]
+output = ["text"]

providers/lilac/models/moonshotai/kimi-k2.6.toml

@@ -0,0 +1,11 @@
+[extends]
+from = "moonshotai/kimi-k2.6"
+
+[cost]
+input = 0.70
+output = 3.50
+cache_read = 0.20
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]

providers/lilac/models/zai-org/glm-5.1.toml

@@ -0,0 +1,20 @@
+name = "GLM 5.1"
+knowledge = "2025-04"
+open_weights = true
+
+[extends]
+from = "zai/glm-5.1"
+omit = ["cost.cache_write"]
+
+[cost]
+input = 0.90
+output = 3.00
+cache_read = 0.27
+
+[limit]
+context = 202_800
+output = 131_072
+
+[modalities]
+input = ["text"]
+output = ["text"]

providers/lilac/provider.toml

@@ -0,0 +1,5 @@
+name = "Lilac"
+env = ["LILAC_API_KEY"]
+npm = "@ai-sdk/openai-compatible"
+api = "https://api.getlilac.com/v1"
+doc = "https://docs.getlilac.com/inference/models"

providers/llmgateway/models/gemini-2.5-flash-lite-preview-09-2025.toml

@@ -1,2 +0,0 @@
-[extends]
-from = "google/gemini-2.5-flash-lite-preview-09-2025"

providers/llmgateway/models/gemma-3-12b-it.toml

@@ -1,2 +0,0 @@
-[extends]
-from = "google/gemma-3-12b-it"

providers/llmgateway/models/gemma-3-4b-it.toml

@@ -1,2 +0,0 @@
-[extends]
-from = "google/gemma-3-4b-it"

providers/llmgateway/models/gemma-3n-e2b-it.toml

@@ -1,2 +0,0 @@
-[extends]
-from = "google/gemma-3n-e2b-it"

providers/llmgateway/models/gemma-3n-e4b-it.toml

@@ -1,2 +0,0 @@
-[extends]
-from = "google/gemma-3n-e4b-it"

providers/llmgateway/models/grok-3.toml

@@ -1,2 +0,0 @@
-[extends]
-from = "xai/grok-3"

providers/llmgateway/models/grok-4-1-fast-non-reasoning.toml

@@ -1,2 +0,0 @@
-[extends]
-from = "xai/grok-4-1-fast-non-reasoning"

providers/llmgateway/models/grok-4-1-fast.toml

@@ -1,2 +0,0 @@
-[extends]
-from = "xai/grok-4-1-fast"

providers/llmgateway/models/grok-4-fast-non-reasoning.toml

@@ -1,2 +0,0 @@
-[extends]
-from = "xai/grok-4-fast-non-reasoning"

providers/llmgateway/models/grok-4-fast.toml

@@ -1,2 +0,0 @@
-[extends]
-from = "xai/grok-4-fast"

providers/llmgateway/models/grok-4.toml

@@ -1,2 +0,0 @@
-[extends]
-from = "xai/grok-4"

providers/llmgateway/models/grok-code-fast-1.toml

@@ -1,2 +0,0 @@
-[extends]
-from = "xai/grok-code-fast-1"

providers/neuralwatt/README.md

@@ -14,7 +14,6 @@ Reasoning Models (with interleaved thinking):
 - zai-org/GLM-5.1-FP8 — GLM 5.1 FP8, reasoning enabled
 - moonshotai/Kimi-K2.5 — Kimi K2.5, reasoning + image input
 - moonshotai/Kimi-K2.6 — Kimi K2.6, reasoning + image input
-- kimi-k2.6-fast — Kimi K2.6 Fast, reasoning + image input
 - MiniMaxAI/MiniMax-M2.5 — MiniMax M2.5, reasoning enabled
 - Qwen/Qwen3.5-397B-A17B-FP8 — Qwen3.5 397B, reasoning enabled
 - Qwen/Qwen3.6-35B-A3B — Qwen3.6 35B A3B, reasoning enabled
@@ -24,6 +23,7 @@ Fast Variants (optimized for speed, non-reasoning):
 - glm-5-fast — GLM 5 Fast
 - glm-5.1-fast — GLM 5.1 Fast
 - kimi-k2.5-fast — Kimi K2.5 Fast, image input
+- kimi-k2.6-fast — Kimi K2.6 Fast, image input
 - qwen3.5-397b-fast — Qwen3.5 397B Fast
 - qwen3.6-35b-fast — Qwen3.6 35B Fast
 
@@ -31,7 +31,7 @@ Other:
 - mistralai/Devstral-Small-2-24B-Instruct-2512 — Devstral Small 2, code-focused + image input
 
 Notes
-- Model IDs and pricing sourced directly from the Neuralwatt API (now fully accurate)
+- Model IDs, pricing, and limits sourced directly from the Neuralwatt API
 - Neuralwatt provides real-time energy consumption data (Joules/kWh) per request
-- "Fast" variants are optimized for lower latency; some fast variants also support reasoning (kimi-k2.6-fast)
+- "Fast" variants are optimized for lower latency without reasoning
 - Vision models support image input via OpenAI-compatible API

providers/neuralwatt/models/MiniMaxAI/MiniMax-M2.5.toml

@@ -13,8 +13,8 @@ input = 0.35
 output = 1.38
 
 [limit]
-context = 196_608
-output = 196_608
+context = 196_592
+output = 196_592
 
 [modalities]
 input = ["text"]

providers/neuralwatt/models/Qwen/Qwen3.5-397B-A17B-FP8.toml

@@ -13,8 +13,8 @@ input = 0.69
 output = 4.14
 
 [limit]
-context = 262_144
-output = 262_144
+context = 262_128
+output = 262_128
 
 [modalities]
 input = ["text"]

providers/neuralwatt/models/Qwen/Qwen3.6-35B-A3B.toml

@@ -13,8 +13,8 @@ input = 0.05
 output = 0.10
 
 [limit]
-context = 131_072
-output = 131_072
+context = 131_056
+output = 131_056
 
 [modalities]
 input = ["text", "image"]

providers/neuralwatt/models/glm-5-fast.toml

@@ -13,8 +13,8 @@ input = 1.1
 output = 3.6
 
 [limit]
-context = 200_000
-output = 200_000
+context = 202_736
+output = 202_736
 
 [modalities]
 input = ["text"]