use Venice context tier thresholds

fix generated tier change detection
align cost tiers with model schema
2026-05-12 17:39:40 -05:00 · 2026-05-12 17:01:35 -05:00 · 2026-05-12 16:01:49 -05:00 · 2026-05-11 17:07:19 -05:00 · 2026-05-11 16:43:32 -05:00 · 2026-05-11 16:31:07 -05:00
327 changed files with 2651 additions and 2483 deletions
@@ -17,6 +17,7 @@
  "scripts": {
    "validate": "bun ./packages/core/script/validate.ts",
    "compare:migrations": "bun ./packages/core/script/compare-model-migrations.ts",
+    "chutes:generate": "bun ./packages/core/script/generate-chutes.ts",
    "helicone:generate": "bun ./packages/core/script/generate-helicone.ts",
    "venice:generate": "bun ./packages/core/script/generate-venice.ts",
    "vercel:generate": "bun ./packages/core/script/generate-vercel.ts",
@@ -0,0 +1,589 @@
+#!/usr/bin/env bun
+
+/**
+ * Generates Chutes model TOML files from the Chutes LLM API.
+ *
+ * Flags:
+ * --dry-run: Preview changes without writing files
+ * --new-only: Only create new models, skip updating existing ones
+ * --keep-orphans: Don't delete TOML files for models no longer in the API
+ */
+
+import { z } from "zod";
+import path from "node:path";
+import { mkdir } from "node:fs/promises";
+import { ModelFamilyValues } from "../src/family.js";
+
+const API_ENDPOINT = "https://llm.chutes.ai/v1/models";
+
+enum SkipZeroFields {
+  LimitContext = "limit.context",
+  LimitOutput = "limit.output",
+}
+
+const Pricing = z.object({
+  prompt: z.number().optional(),
+  completion: z.number().optional(),
+  input_cache_read: z.number().optional(),
+}).passthrough();
+
+const ChutesModel = z.object({
+  id: z.string(),
+  created: z.number(),
+  pricing: Pricing.optional(),
+  context_length: z.number().optional(),
+  max_output_length: z.number().optional(),
+  max_model_len: z.number().optional(),
+  input_modalities: z.array(z.string()).optional(),
+  output_modalities: z.array(z.string()).optional(),
+  supported_features: z.array(z.string()).optional(),
+  supported_sampling_parameters: z.array(z.string()).optional(),
+  quantization: z.string().optional(),
+}).passthrough();
+
+const ChutesResponse = z.object({
+  data: z.array(ChutesModel),
+}).passthrough();
+
+interface ExistingModel {
+  name?: string;
+  family?: string;
+  attachment?: boolean;
+  reasoning?: boolean;
+  tool_call?: boolean;
+  structured_output?: boolean;
+  temperature?: boolean;
+  knowledge?: string;
+  release_date?: string;
+  last_updated?: string;
+  open_weights?: boolean;
+  interleaved?: boolean | { field: string };
+  status?: string;
+  cost?: {
+    input?: number;
+    output?: number;
+    cache_read?: number;
+  };
+  limit?: {
+    context?: number;
+    output?: number;
+  };
+  modalities?: {
+    input?: string[];
+    output?: string[];
+  };
+}
+
+interface MergedModel {
+  name: string;
+  family?: string;
+  attachment: boolean;
+  reasoning: boolean;
+  tool_call: boolean;
+  structured_output?: boolean;
+  temperature: boolean;
+  knowledge?: string;
+  release_date: string;
+  last_updated: string;
+  open_weights: boolean;
+  interleaved?: boolean | { field: string };
+  status?: string;
+  cost?: {
+    input: number;
+    output: number;
+    cache_read?: number;
+  };
+  limit: {
+    context: number;
+    output: number;
+  };
+  modalities: {
+    input: string[];
+    output: string[];
+  };
+}
+
+interface Changes {
+  field: string;
+  oldValue: string;
+  newValue: string;
+}
+
+// ── Utility functions ────────────────────────────────────────────────
+
+function timestampToDate(timestamp: number): string {
+  const date = new Date(timestamp * 1000);
+  return date.toISOString().slice(0, 10);
+}
+
+function getTodayDate(): string {
+  return new Date().toISOString().slice(0, 10);
+}
+
+function formatNumber(n: number): string {
+  if (n >= 1000) {
+    return n.toString().replace(/\B(?=(\d{3})+(?!\d))/g, "_");
+  }
+  return n.toString();
+}
+
+/**
+ * Humanize a model ID into a readable name.
+ * Strips the org prefix and replaces hyphens with spaces.
+ * e.g. "Qwen/Qwen3-32B-TEE" → "Qwen3 32B TEE"
+ */
+function humanizeModelName(modelId: string): string {
+  const parts = modelId.split("/");
+  const modelPart = parts[parts.length - 1];
+  return modelPart.replace(/-/g, " ");
+}
+
+// ── Family inference (same approach as generate-vercel.ts) ───────────
+
+function isSubstring(target: string, family: string): boolean {
+  return target.toLowerCase().includes(family.toLowerCase());
+}
+
+function matchesFamily(target: string, family: string): boolean {
+  const targetLower = target.toLowerCase();
+  const familyLower = family.toLowerCase();
+  let familyIdx = 0;
+
+  for (let i = 0; i < targetLower.length && familyIdx < familyLower.length; i++) {
+    if (targetLower[i] === familyLower[familyIdx]) {
+      familyIdx++;
+    }
+  }
+
+  return familyIdx === familyLower.length;
+}
+
+function inferFamily(modelId: string, modelName: string): string | undefined {
+  const sortedFamilies = [...ModelFamilyValues].sort((a, b) => b.length - a.length);
+
+  // First pass: try exact substring matches
+  for (const family of sortedFamilies) {
+    if (isSubstring(modelId, family)) {
+      return family;
+    }
+  }
+
+  for (const family of sortedFamilies) {
+    if (isSubstring(modelName, family)) {
+      return family;
+    }
+  }
+
+  // Second pass: fall back to subsequence matching
+  for (const family of sortedFamilies) {
+    if (matchesFamily(modelId, family)) {
+      return family;
+    }
+  }
+
+  for (const family of sortedFamilies) {
+    if (matchesFamily(modelName, family)) {
+      return family;
+    }
+  }
+
+  return undefined;
+}
+
+// ── Load existing TOML ───────────────────────────────────────────────
+
+async function loadExistingModel(filePath: string): Promise<ExistingModel | null> {
+  try {
+    const file = Bun.file(filePath);
+    if (!(await file.exists())) {
+      return null;
+    }
+    const toml = await import(filePath, { with: { type: "toml" } }).then(
+      (mod) => mod.default,
+    );
+    return toml as ExistingModel;
+  } catch (e) {
+    console.warn(`Warning: Failed to parse existing file ${filePath}:`, e);
+    return null;
+  }
+}
+
+// ── Merge API data with existing TOML ────────────────────────────────
+
+function mergeModel(
+  apiModel: z.infer<typeof ChutesModel>,
+  existing: ExistingModel | null,
+): MergedModel {
+  const features = new Set(apiModel.supported_features ?? []);
+  const samplingParams = new Set(apiModel.supported_sampling_parameters ?? []);
+  const inputMods = apiModel.input_modalities ?? ["text"];
+  const outputMods = apiModel.output_modalities ?? ["text"];
+
+  // Capabilities from API features
+  const hasAttachment = inputMods.some((m) =>
+    m === "image" || m === "video" || m === "pdf",
+  );
+  const hasReasoning = features.has("reasoning");
+  const hasToolCall = features.has("tools");
+  const hasStructuredOutput = features.has("structured_outputs");
+  const hasTemperature = samplingParams.size > 0
+    ? samplingParams.has("temperature")
+    : true; // default true if no sampling params info
+
+  // Preserve existing values when available (manually specified)
+  const modelName = existing?.name ?? humanizeModelName(apiModel.id);
+  const family = existing?.family ?? inferFamily(apiModel.id, modelName);
+  const knowledge = existing?.knowledge;
+  const interleaved = existing?.interleaved;
+  const status = existing?.status;
+
+  // Release date: existing > API created timestamp > today
+  const releaseDate = existing?.release_date
+    ?? timestampToDate(apiModel.created)
+    ?? getTodayDate();
+
+  // Context limit: prefer context_length, fallback to max_model_len
+  const apiContext = apiModel.context_length ?? apiModel.max_model_len ?? 0;
+  const contextLimit = apiContext > 0
+    ? apiContext
+    : (existing?.limit?.context ?? 0);
+
+  // Output limit: prefer max_output_length, fallback to existing
+  const apiOutput = apiModel.max_output_length ?? 0;
+  const outputLimit = apiOutput > 0
+    ? apiOutput
+    : (existing?.limit?.output ?? 0);
+
+  const merged: MergedModel = {
+    name: modelName,
+    family,
+    attachment: hasAttachment,
+    reasoning: hasReasoning,
+    tool_call: hasToolCall,
+    temperature: hasTemperature,
+    release_date: releaseDate,
+    last_updated: getTodayDate(),
+    open_weights: true, // Chutes hosts open-weight models
+    ...(hasStructuredOutput && { structured_output: hasStructuredOutput }),
+    ...(knowledge && { knowledge }),
+    ...(interleaved !== undefined && { interleaved }),
+    ...(status && { status }),
+    limit: {
+      context: contextLimit,
+      output: outputLimit,
+    },
+    modalities: {
+      input: inputMods,
+      output: outputMods,
+    },
+  };
+
+  // Cost: API values are already in USD per 1M tokens — use directly
+  if (apiModel.pricing) {
+    const inputPrice = apiModel.pricing.prompt;
+    const outputPrice = apiModel.pricing.completion;
+    const cacheReadPrice = apiModel.pricing.input_cache_read;
+
+    if (inputPrice !== undefined && outputPrice !== undefined) {
+      merged.cost = {
+        input: inputPrice,
+        output: outputPrice,
+        ...(cacheReadPrice !== undefined && { cache_read: cacheReadPrice }),
+      };
+    }
+  }
+
+  return merged;
+}
+
+// ── TOML formatting ──────────────────────────────────────────────────
+
+function formatToml(model: MergedModel): string {
+  const lines: string[] = [];
+
+  lines.push(`# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.`);
+  lines.push(`# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status`);
+  lines.push(`name = "${model.name.replace(/"/g, '\\"')}"`);
+  if (model.family) {
+    lines.push(`family = "${model.family}"`);
+  }
+  lines.push(`release_date = "${model.release_date}"`);
+  lines.push(`last_updated = "${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}`);
+  if (model.structured_output !== undefined) {
+    lines.push(`structured_output = ${model.structured_output}`);
+  }
+  lines.push(`open_weights = ${model.open_weights}`);
+  if (model.knowledge) {
+    lines.push(`knowledge = "${model.knowledge}"`);
+  }
+  if (model.status) {
+    lines.push(`status = "${model.status}"`);
+  }
+
+  if (model.cost) {
+    lines.push("");
+    lines.push(`[cost]`);
+    lines.push(`input = ${model.cost.input}`);
+    lines.push(`output = ${model.cost.output}`);
+    if (model.cost.cache_read !== undefined) {
+      lines.push(`cache_read = ${model.cost.cache_read}`);
+    }
+  }
+
+  lines.push("");
+  lines.push(`[limit]`);
+  lines.push(`context = ${formatNumber(model.limit.context)}`);
+  lines.push(`output = ${formatNumber(model.limit.output)}`);
+
+  lines.push("");
+  lines.push(`[modalities]`);
+  lines.push(`input = [${model.modalities.input.map((m) => `"${m}"`).join(", ")}]`);
+  lines.push(`output = [${model.modalities.output.map((m) => `"${m}"`).join(", ")}]`);
+
+  if (model.interleaved !== undefined) {
+    lines.push("");
+    if (model.interleaved === true) {
+      lines.push(`interleaved = true`);
+    } else if (typeof model.interleaved === "object") {
+      lines.push(`[interleaved]`);
+      lines.push(`field = "${model.interleaved.field}"`);
+    }
+  }
+
+  return lines.join("\n") + "\n";
+}
+
+// ── Change detection ─────────────────────────────────────────────────
+
+function detectChanges(
+  existing: ExistingModel | null,
+  merged: MergedModel,
+): Changes[] {
+  if (!existing) return [];
+
+  const changes: Changes[] = [];
+  const EPSILON = 0.001;
+
+  const shouldSkipZero = (field: string, oldVal: unknown, newVal: unknown): boolean => {
+    if (!Object.values(SkipZeroFields).includes(field as SkipZeroFields)) {
+      return false;
+    }
+    return (typeof oldVal === "number" && oldVal === 0) || (typeof newVal === "number" && newVal === 0);
+  };
+
+  const formatValue = (val: unknown): string => {
+    if (typeof val === "number") return formatNumber(val);
+    if (Array.isArray(val)) return `[${val.join(", ")}]`;
+    if (val === undefined) return "(none)";
+    return String(val);
+  };
+
+  const isMaterialPriceDiff = (oldPrice: unknown, newPrice: unknown): boolean => {
+    if (oldPrice === 0 && newPrice === undefined) return false;
+    if (oldPrice !== undefined && newPrice !== undefined) {
+      return Math.abs((oldPrice as number) - (newPrice as number)) > EPSILON;
+    }
+    return oldPrice !== newPrice;
+  };
+
+  const compare = (field: string, oldVal: unknown, newVal: unknown) => {
+    if (shouldSkipZero(field, oldVal, newVal)) return;
+
+    const isDiff = field.startsWith("cost.")
+      ? isMaterialPriceDiff(oldVal, newVal)
+      : JSON.stringify(oldVal) !== JSON.stringify(newVal);
+
+    if (isDiff) {
+      changes.push({
+        field,
+        oldValue: formatValue(oldVal),
+        newValue: formatValue(newVal),
+      });
+    }
+  };
+
+  compare("name", existing.name, merged.name);
+  compare("family", existing.family, merged.family);
+  compare("attachment", existing.attachment, merged.attachment);
+  compare("reasoning", existing.reasoning, merged.reasoning);
+  compare("tool_call", existing.tool_call, merged.tool_call);
+  compare("structured_output", existing.structured_output, merged.structured_output);
+  compare("open_weights", existing.open_weights, merged.open_weights);
+  compare("release_date", existing.release_date, merged.release_date);
+  compare("cost.input", existing.cost?.input, merged.cost?.input);
+  compare("cost.output", existing.cost?.output, merged.cost?.output);
+  compare("cost.cache_read", existing.cost?.cache_read, merged.cost?.cache_read);
+  compare("limit.context", existing.limit?.context, merged.limit.context);
+  compare("limit.output", existing.limit?.output, merged.limit.output);
+  compare("modalities.input", existing.modalities?.input, merged.modalities.input);
+  compare("modalities.output", existing.modalities?.output, merged.modalities.output);
+
+  return changes;
+}
+
+// ── Main ─────────────────────────────────────────────────────────────
+
+async function main() {
+  const args = process.argv.slice(2);
+  const dryRun = args.includes("--dry-run");
+  const newOnly = args.includes("--new-only");
+  const keepOrphans = args.includes("--keep-orphans");
+
+  const modelsDir = path.join(
+    import.meta.dirname,
+    "..",
+    "..",
+    "..",
+    "providers",
+    "chutes",
+    "models",
+  );
+
+  console.log(`${dryRun ? "[DRY RUN] " : ""}${newOnly ? "[NEW ONLY] " : ""}${keepOrphans ? "[KEEP ORPHANS] " : ""}Fetching Chutes models from API...`);
+
+  const res = await fetch(API_ENDPOINT);
+  if (!res.ok) {
+    console.error(`Failed to fetch API: ${res.status} ${res.statusText}`);
+    process.exit(1);
+  }
+
+  const json = await res.json();
+  const parsed = ChutesResponse.safeParse(json);
+  if (!parsed.success) {
+    console.error("Invalid API response:", parsed.error.errors);
+    process.exit(1);
+  }
+
+  const apiModels = parsed.data.data;
+
+  // Scan existing TOML files
+  const existingFiles = new Set<string>();
+  try {
+    for await (const file of new Bun.Glob("**/*.toml").scan({
+      cwd: modelsDir,
+      absolute: false,
+    })) {
+      existingFiles.add(file);
+    }
+  } catch {
+  }
+
+  console.log(`Found ${apiModels.length} models in API, ${existingFiles.size} existing files\n`);
+
+  const apiModelIds = new Set<string>();
+
+  let created = 0;
+  let updated = 0;
+  let unchanged = 0;
+
+  for (const apiModel of apiModels) {
+    const relativePath = `${apiModel.id}.toml`;
+    const filePath = path.join(modelsDir, relativePath);
+    const dirPath = path.dirname(filePath);
+
+    apiModelIds.add(relativePath);
+
+    const existing = await loadExistingModel(filePath);
+    const merged = mergeModel(apiModel, existing);
+    const tomlContent = formatToml(merged);
+
+    if (existing === null) {
+      created++;
+      if (dryRun) {
+        console.log(`[DRY RUN] Would create: ${relativePath}`);
+        console.log(`  name = "${merged.name}"`);
+        if (merged.family) {
+          console.log(`  family = "${merged.family}" (inferred)`);
+        }
+        console.log("");
+      } else {
+        await mkdir(dirPath, { recursive: true });
+        await Bun.write(filePath, tomlContent);
+        console.log(`Created: ${relativePath}`);
+      }
+    } else {
+      if (newOnly) {
+        unchanged++;
+        continue;
+      }
+
+      const changes = detectChanges(existing, merged);
+      const existingContent = await Bun.file(filePath).text();
+      const formatChanged = existingContent !== tomlContent;
+
+      if (changes.length > 0 || formatChanged) {
+        updated++;
+        if (dryRun) {
+          console.log(`[DRY RUN] Would update: ${relativePath}`);
+        } else {
+          await mkdir(dirPath, { recursive: true });
+          await Bun.write(filePath, tomlContent);
+          console.log(`Updated: ${relativePath}`);
+        }
+        for (const change of changes) {
+          console.log(`  ${change.field}: ${change.oldValue} → ${change.newValue}`);
+        }
+        if (changes.length === 0 && formatChanged) {
+          console.log(`  (format-only change)`);
+        }
+        console.log("");
+      } else {
+        unchanged++;
+      }
+    }
+  }
+
+  // Handle orphaned files (on disk but not in API)
+  const orphaned: string[] = [];
+  for (const file of existingFiles) {
+    if (!apiModelIds.has(file)) {
+      orphaned.push(file);
+      const orphanPath = path.join(modelsDir, file);
+      if (keepOrphans) {
+        console.log(`Orphaned (kept): ${file}`);
+      } else if (dryRun) {
+        console.log(`[DRY RUN] Would delete: ${file}`);
+      } else {
+        await Bun.file(orphanPath).delete();
+        console.log(`Deleted: ${file}`);
+
+        // Clean up empty parent directories
+        const parentDir = path.dirname(orphanPath);
+        try {
+          const remaining = [];
+          for await (const entry of new Bun.Glob("*").scan({ cwd: parentDir })) {
+            remaining.push(entry);
+          }
+          if (remaining.length === 0) {
+            const { rmdir } = await import("node:fs/promises");
+            await rmdir(parentDir);
+            console.log(`  Removed empty directory: ${path.basename(parentDir)}/`);
+          }
+        } catch {
+          // Directory not empty or other error, ignore
+        }
+      }
+    }
+  }
+
+  console.log("");
+  if (dryRun) {
+    console.log(
+      `Summary: ${created} would be created, ${updated} would be updated, ${unchanged} unchanged, ${orphaned.length} would be deleted`,
+    );
+  } else if (keepOrphans) {
+    console.log(
+      `Summary: ${created} created, ${updated} updated, ${unchanged} unchanged, ${orphaned.length} orphaned (kept)`,
+    );
+  } else {
+    console.log(
+      `Summary: ${created} created, ${updated} updated, ${unchanged} unchanged, ${orphaned.length} deleted`,
+    );
+  }
+}
+
+await main();
@@ -209,7 +209,18 @@ interface ExistingModel {
      output?: number;
      cache_read?: number;
      cache_write?: number;
+      context_min?: number;
    };
+    tiers?: Array<{
+      tier: {
+        type?: "context";
+        size: number;
+      };
+      input?: number;
+      output?: number;
+      cache_read?: number;
+      cache_write?: number;
+    }>;
  };
  limit?: {
    context?: number;
@@ -262,6 +273,7 @@ interface MergedModel {
      output: number;
      cache_read?: number;
      cache_write?: number;
+      context_min?: number;
    };
  };
  limit: {
@@ -310,6 +322,35 @@ function inferFamily(modelId: string, modelName: string): string | undefined {
  return undefined;
 }

+function getExistingLongContextCost(existing: ExistingModel | null) {
+  const tier = existing?.cost?.tiers?.find(
+    (tier) =>
+      (tier.tier.type === undefined || tier.tier.type === "context") &&
+      tier.tier.size >= 200_000,
+  );
+  if (tier) {
+    return {
+      ...tier,
+      context_min: tier.tier.size,
+    };
+  }
+
+  return existing?.cost?.context_over_200k === undefined
+    ? undefined
+    : {
+        ...existing.cost.context_over_200k,
+        context_min: 200_000,
+      };
+}
+
+function getLongContextMin(cost: { context_min?: number }) {
+  return cost.context_min ?? 200_000;
+}
+
+function formatInlineNumber(n: number): string {
+  return n >= 1000 ? n.toString().replace(/\B(?=(\d{3})+(?!\d))/g, "_") : n.toString();
+}
+
 // ---------------------------------------------------------------------------
 // Merge API data with existing TOML
 // ---------------------------------------------------------------------------
@@ -394,27 +435,30 @@ function mergeModel(
    };

    // Context-tiered pricing (>200k) from the static-content API
+    const existingLongContextCost = getExistingLongContextCost(existing);
    if (pricing?.inputOver200k !== undefined && pricing?.outputOver200k !== undefined) {
      merged.cost.context_over_200k = {
        input: pricing.inputOver200k,
        output: pricing.outputOver200k,
-        ...(existing?.cost?.context_over_200k?.cache_read !== undefined && {
-          cache_read: existing.cost.context_over_200k.cache_read,
+        context_min: existingLongContextCost?.context_min ?? 200_000,
+        ...(existingLongContextCost?.cache_read !== undefined && {
+          cache_read: existingLongContextCost.cache_read,
        }),
-        ...(existing?.cost?.context_over_200k?.cache_write !== undefined && {
-          cache_write: existing.cost.context_over_200k.cache_write,
+        ...(existingLongContextCost?.cache_write !== undefined && {
+          cache_write: existingLongContextCost.cache_write,
        }),
      };
-    } else if (existing?.cost?.context_over_200k) {
-      // Preserve manually-entered context_over_200k if API has no data
+    } else if (existingLongContextCost) {
+      // Preserve manually-entered tiered pricing if API has no data
      merged.cost.context_over_200k = {
-        input: existing.cost.context_over_200k.input ?? inputPrice,
-        output: existing.cost.context_over_200k.output ?? outputPrice,
-        ...(existing.cost.context_over_200k.cache_read !== undefined && {
-          cache_read: existing.cost.context_over_200k.cache_read,
+        input: existingLongContextCost.input ?? inputPrice,
+        output: existingLongContextCost.output ?? outputPrice,
+        context_min: existingLongContextCost.context_min,
+        ...(existingLongContextCost.cache_read !== undefined && {
+          cache_read: existingLongContextCost.cache_read,
        }),
-        ...(existing.cost.context_over_200k.cache_write !== undefined && {
-          cache_write: existing.cost.context_over_200k.cache_write,
+        ...(existingLongContextCost.cache_write !== undefined && {
+          cache_write: existingLongContextCost.cache_write,
        }),
      };
    }
@@ -463,7 +507,8 @@ function formatToml(model: MergedModel): string {

    if (model.cost.context_over_200k) {
      lines.push("");
-      lines.push(`[cost.context_over_200k]`);
+      lines.push(`[[cost.tiers]]`);
+      lines.push(`tier = { size = ${formatInlineNumber(getLongContextMin(model.cost.context_over_200k))} }`);
      lines.push(`input = ${model.cost.context_over_200k.input}`);
      lines.push(`output = ${model.cost.context_over_200k.output}`);
      if (model.cost.context_over_200k.cache_read !== undefined)
@@ -524,8 +569,9 @@ function detectChanges(existing: ExistingModel | null, merged: MergedModel): Cha
  compare("status", existing.status, merged.status);
  compare("cost.input", existing.cost?.input, merged.cost?.input);
  compare("cost.output", existing.cost?.output, merged.cost?.output);
-  compare("cost.context_over_200k.input", existing.cost?.context_over_200k?.input, merged.cost?.context_over_200k?.input);
-  compare("cost.context_over_200k.output", existing.cost?.context_over_200k?.output, merged.cost?.context_over_200k?.output);
+  const existingLongContextCost = getExistingLongContextCost(existing);
+  compare("cost.context_over_200k.input", existingLongContextCost?.input, merged.cost?.context_over_200k?.input);
+  compare("cost.context_over_200k.output", existingLongContextCost?.output, merged.cost?.context_over_200k?.output);
  compare("limit.context", existing.limit?.context, merged.limit.context);
  compare("limit.output", existing.limit?.output, merged.limit.output);
  compare("modalities.input", existing.modalities?.input, merged.modalities.input);
@@ -162,7 +162,18 @@ interface ExistingModel {
      output?: number;
      cache_read?: number;
      cache_write?: number;
+      context_min?: number;
    };
+    tiers?: Array<{
+      tier: {
+        type?: "context";
+        size: number;
+      };
+      input?: number;
+      output?: number;
+      cache_read?: number;
+      cache_write?: number;
+    }>;
  };
  limit?: {
    context?: number;
@@ -195,6 +206,30 @@ async function loadExistingModel(filePath: string): Promise<ExistingModel | null
  }
 }

+function getExistingLongContextMin(existing: ExistingModel | null) {
+  return (
+    existing?.cost?.tiers?.find(
+      (tier) =>
+        (tier.tier.type === undefined || tier.tier.type === "context") &&
+        tier.tier.size >= 200_000,
+    )?.tier.size ?? 200_000
+  );
+}
+
+function getExistingLongContextCost(existing: ExistingModel | null) {
+  return (
+    existing?.cost?.tiers?.find(
+      (tier) =>
+        (tier.tier.type === undefined || tier.tier.type === "context") &&
+        tier.tier.size >= 200_000,
+    ) ?? existing?.cost?.context_over_200k
+  );
+}
+
+function getLongContextMin(cost: { context_min?: number }) {
+  return cost.context_min ?? 200_000;
+}
+
 interface MergedModel {
  name: string;
  family?: string;
@@ -219,6 +254,7 @@ interface MergedModel {
      output: number;
      cache_read?: number;
      cache_write?: number;
+      context_min?: number;
    };
  };
  limit: {
@@ -292,6 +328,7 @@ function mergeModel(
      merged.cost.context_over_200k = {
        input: spec.pricing.extended.input.usd,
        output: spec.pricing.extended.output.usd,
+        context_min: spec.pricing.extended.context_token_threshold,
        ...(spec.pricing.extended.cache_input && { cache_read: spec.pricing.extended.cache_input.usd }),
        ...(spec.pricing.extended.cache_write && { cache_write: spec.pricing.extended.cache_write.usd }),
      };
@@ -366,7 +403,8 @@ function formatToml(model: MergedModel): string {

    if (model.cost.context_over_200k) {
      lines.push("");
-      lines.push(`[cost.context_over_200k]`);
+      lines.push(`[[cost.tiers]]`);
+      lines.push(`tier = { size = ${formatNumber(getLongContextMin(model.cost.context_over_200k))} }`);
      lines.push(`input = ${model.cost.context_over_200k.input}`);
      lines.push(`output = ${model.cost.context_over_200k.output}`);
      if (model.cost.context_over_200k.cache_read !== undefined) {
@@ -438,10 +476,11 @@ function detectChanges(
  compare("cost.output", existing.cost?.output, merged.cost?.output);
  compare("cost.cache_read", existing.cost?.cache_read, merged.cost?.cache_read);
  compare("cost.cache_write", existing.cost?.cache_write, merged.cost?.cache_write);
-  compare("cost.context_over_200k.input", existing.cost?.context_over_200k?.input, merged.cost?.context_over_200k?.input);
-  compare("cost.context_over_200k.output", existing.cost?.context_over_200k?.output, merged.cost?.context_over_200k?.output);
-  compare("cost.context_over_200k.cache_read", existing.cost?.context_over_200k?.cache_read, merged.cost?.context_over_200k?.cache_read);
-  compare("cost.context_over_200k.cache_write", existing.cost?.context_over_200k?.cache_write, merged.cost?.context_over_200k?.cache_write);
+  const existingLongContextCost = getExistingLongContextCost(existing);
+  compare("cost.context_over_200k.input", existingLongContextCost?.input, merged.cost?.context_over_200k?.input);
+  compare("cost.context_over_200k.output", existingLongContextCost?.output, merged.cost?.context_over_200k?.output);
+  compare("cost.context_over_200k.cache_read", existingLongContextCost?.cache_read, merged.cost?.context_over_200k?.cache_read);
+  compare("cost.context_over_200k.cache_write", existingLongContextCost?.cache_write, merged.cost?.context_over_200k?.cache_write);
  compare("limit.context", existing.limit?.context, merged.limit.context);
  compare("limit.output", existing.limit?.output, merged.limit.output);
  compare("modalities.input", existing.modalities?.input, merged.modalities.input);
@@ -55,6 +55,7 @@ export const ModelFamilyValues = [
  "deepseek",
  "deepseek-thinking",
  "deepseek-flash",
+  "deepseek-flash-free",
  "deepseek-flash-think",

  // Microsoft Phi
@@ -289,12 +290,14 @@ export const ModelFamilyValues = [
  "rnj",

  // Tecent Hy
+  "hy3",
  "hy3-free",

  // Ling & Ring (InclusionAI)
  "ling",
  "ling-flash-free",
  "ring",
+  "ring-1t-free",

  // Kat Coder
  "kat-coder",
@@ -2,9 +2,9 @@ import path from "path";
 import { mergeDeep } from "remeda";
 import { z } from "zod";

-import { Provider, Model } from "./schema.js";
+import { Provider, Model, AuthoredModel, AuthoredModelShape } from "./schema.js";

-const ExtendsModel = Model.sourceType()
+const ExtendsModel = AuthoredModelShape
  .partial()
  .extend({
    extends: z
@@ -71,12 +71,12 @@ export async function generate(directory: string) {
        });
        continue;
      }
-      const model = Model.safeParse(toml);
+      const model = AuthoredModel.safeParse(toml);
      if (!model.success) {
        model.error.cause = { modelPath, toml };
        throw model.error;
      }
-      provider.data.models[modelID] = model.data;
+      provider.data.models[modelID] = normalizeModelCost(model.data);
    }
    result[providerID] = provider.data;
  }
@@ -144,7 +144,7 @@ export async function generate(directory: string) {
      }
    }

-    const model = Model.safeParse(merged);
+    const model = Model.safeParse(normalizeCost(merged));
    if (!model.success) {
      model.error.cause = { modelPath: pendingModel.modelPath, toml: merged };
      throw model.error;
@@ -155,3 +155,51 @@ export async function generate(directory: string) {

  return result;
 }
+
+function normalizeModelCost(model: z.infer<typeof AuthoredModel>): Model {
+  return normalizeCost(model) as Model;
+}
+
+function normalizeCost(model: Record<string, unknown>) {
+  const cost = model.cost;
+  if (cost === undefined || cost === null || typeof cost !== "object" || Array.isArray(cost)) {
+    return model;
+  }
+
+  const tiers = (cost as { tiers?: unknown }).tiers;
+  if (!Array.isArray(tiers)) {
+    return model;
+  }
+
+  if (tiers.length !== 1) {
+    return model;
+  }
+
+  const contextOver200k = tiers.find((tier) => {
+    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 type = (tierConfig as { type?: unknown }).type;
+    const size = (tierConfig as { size?: unknown }).size;
+    // context_over_200k is a legacy compatibility field. It intentionally
+    // includes higher thresholds; cost.tiers carries the exact threshold.
+    return (
+      (type === undefined || type === "context") &&
+      typeof size === "number" &&
+      size >= 200_000
+    );
+  });
+
+  if (contextOver200k === undefined) {
+    return model;
+  }
+
+  const { tier: _tier, ...legacyCost } = contextOver200k as Record<string, unknown>;
+  return {
+    ...model,
+    cost: {
+      ...(cost as Record<string, unknown>),
+      context_over_200k: legacyCost,
+    },
+  };
+}
@@ -21,110 +21,164 @@ const JsonValue: z.ZodType<JsonValue> = z.lazy(() =>
  ]),
 );

-const Cost = z.object({
-  input: z.number().min(0, "Input price cannot be negative"),
-  output: z.number().min(0, "Output price cannot be negative"),
-  reasoning: z.number().min(0, "Input price cannot be negative").optional(),
-  cache_read: z
-    .number()
-    .min(0, "Cache read price cannot be negative")
+const Cost = z
+  .object({
+    input: z.number().min(0, "Input price cannot be negative"),
+    output: z.number().min(0, "Output price cannot be negative"),
+    reasoning: z
+      .number()
+      .min(0, "Reasoning price cannot be negative")
+      .optional(),
+    cache_read: z
+      .number()
+      .min(0, "Cache read price cannot be negative")
+      .optional(),
+    cache_write: z
+      .number()
+      .min(0, "Cache write price cannot be negative")
+      .optional(),
+    input_audio: z
+      .number()
+      .min(0, "Audio input price cannot be negative")
+      .optional(),
+    output_audio: z
+      .number()
+      .min(0, "Audio output price cannot be negative")
+      .optional(),
+  });
+
+const CostTier = Cost.extend({
+  tier: z
+    .object({
+      type: z.literal("context").default("context"),
+      size: z.number().int().min(0, "Context tier size cannot be negative"),
+    })
+    .strict(),
+}).strict();
+
+const AuthoredCost = Cost.extend({
+  context_over_200k: z.never().optional(),
+  tiers: z.array(CostTier).optional(),
+});
+
+const OutputCost = Cost.extend({
+  context_over_200k: Cost.optional(),
+  tiers: z.array(CostTier).optional(),
+});
+
+const ModelBase = z.object({
+  id: z.string(),
+  name: z.string().min(1, "Model name cannot be empty"),
+  family: ModelFamily.optional(),
+  attachment: z.boolean(),
+  reasoning: z.boolean(),
+  tool_call: z.boolean(),
+  interleaved: z
+    .union([
+      z.literal(true),
+      z
+        .object({
+          field: z.enum(["reasoning_content", "reasoning_details"]),
+        })
+        .strict(),
+    ])
    .optional(),
-  cache_write: z
-    .number()
-    .min(0, "Cache write price cannot be negative")
+  structured_output: z.boolean().optional(),
+  temperature: z.boolean().optional(),
+  knowledge: z
+    .string()
+    .regex(/^\d{4}-\d{2}(-\d{2})?$/, {
+      message: "Must be in YYYY-MM or YYYY-MM-DD format",
+    })
    .optional(),
-  input_audio: z
-    .number()
-    .min(0, "Audio input price cannot be negative")
+  release_date: z.string().regex(/^\d{4}-\d{2}(-\d{2})?$/, {
+    message: "Must be in YYYY-MM or YYYY-MM-DD format",
+  }),
+  last_updated: z.string().regex(/^\d{4}-\d{2}(-\d{2})?$/, {
+    message: "Must be in YYYY-MM or YYYY-MM-DD format",
+  }),
+  modalities: z.object({
+    input: z.array(z.enum(["text", "audio", "image", "video", "pdf"])),
+    output: z.array(z.enum(["text", "audio", "image", "video", "pdf"])),
+  }),
+  open_weights: z.boolean(),
+  limit: z.object({
+    context: z.number().min(0, "Context window must be positive"),
+    input: z.number().min(0, "Input tokens must be positive").optional(),
+    output: z.number().min(0, "Output tokens must be positive"),
+  }),
+  status: z.enum(["alpha", "beta", "deprecated"]).optional(),
+  experimental: z
+    .object({
+      modes: z
+        .record(
+          z.object({
+            cost: Cost.optional(),
+            provider: z
+              .object({
+                body: z.record(JsonValue).optional(),
+                headers: z.record(z.string()).optional(),
+              })
+              .optional(),
+          }),
+        )
+        .optional(),
+    })
    .optional(),
-  output_audio: z
-    .number()
-    .min(0, "Audio output price cannot be negative")
+  provider: z
+    .object({
+      npm: z.string().optional(),
+      api: z.string().optional(),
+      shape: z.enum(["responses", "completions"]).optional(),
+      body: z.record(JsonValue).optional(),
+      headers: z.record(z.string()).optional(),
+    })
    .optional(),
 });
-export const Model = z
+
+function refineModel<T extends z.ZodTypeAny>(schema: T) {
+  return schema
+    .refine(
+      (data) => {
+        return !(data.reasoning === false && data.cost?.reasoning !== undefined);
+      },
+      {
+        message: "Cannot set cost.reasoning when reasoning is false",
+        path: ["cost", "reasoning"],
+      },
+    )
+    .refine(
+      (data) => {
+        const tiers = data.cost?.tiers;
+        if (tiers === undefined) return true;
+
+        const sizes = tiers.map((tier: { tier: { size: number } }) => tier.tier.size);
+        return new Set(sizes).size === sizes.length;
+      },
+      {
+        message: "Cost context tiers must not have duplicate sizes",
+        path: ["cost", "tiers"],
+      },
+    );
+}
+
+export const ModelShape = z
  .object({
-    id: z.string(),
-    name: z.string().min(1, "Model name cannot be empty"),
-    family: ModelFamily.optional(),
-    attachment: z.boolean(),
-    reasoning: z.boolean(),
-    tool_call: z.boolean(),
-    interleaved: z
-      .union([
-        z.literal(true),
-        z
-          .object({
-            field: z.enum(["reasoning_content", "reasoning_details"]),
-          })
-          .strict(),
-      ])
-      .optional(),
-    structured_output: z.boolean().optional(),
-    temperature: z.boolean().optional(),
-    knowledge: z
-      .string()
-      .regex(/^\d{4}-\d{2}(-\d{2})?$/, {
-        message: "Must be in YYYY-MM or YYYY-MM-DD format",
-      })
-      .optional(),
-    release_date: z.string().regex(/^\d{4}-\d{2}(-\d{2})?$/, {
-      message: "Must be in YYYY-MM or YYYY-MM-DD format",
-    }),
-    last_updated: z.string().regex(/^\d{4}-\d{2}(-\d{2})?$/, {
-      message: "Must be in YYYY-MM or YYYY-MM-DD format",
-    }),
-    modalities: z.object({
-      input: z.array(z.enum(["text", "audio", "image", "video", "pdf"])),
-      output: z.array(z.enum(["text", "audio", "image", "video", "pdf"])),
-    }),
-    open_weights: z.boolean(),
-    cost: Cost.extend({
-      context_over_200k: Cost.optional(),
-    }).optional(),
-    limit: z.object({
-      context: z.number().min(0, "Context window must be positive"),
-      input: z.number().min(0, "Input tokens must be positive").optional(),
-      output: z.number().min(0, "Output tokens must be positive"),
-    }),
-    status: z.enum(["alpha", "beta", "deprecated"]).optional(),
-    experimental: z
-      .object({
-        modes: z
-          .record(
-            z.object({
-              cost: Cost.optional(),
-              provider: z
-                .object({
-                  body: z.record(JsonValue).optional(),
-                  headers: z.record(z.string()).optional(),
-                })
-                .optional(),
-            }),
-          )
-          .optional(),
-      })
-      .optional(),
-    provider: z
-      .object({
-        npm: z.string().optional(),
-        api: z.string().optional(),
-        shape: z.enum(["responses", "completions"]).optional(),
-        body: z.record(JsonValue).optional(),
-        headers: z.record(z.string()).optional(),
-      })
-      .optional(),
+    ...ModelBase.shape,
+    cost: OutputCost.optional(),
  })
-  .strict()
-  .refine(
-    (data) => {
-      return !(data.reasoning === false && data.cost?.reasoning !== undefined);
-    },
-    {
-      message: "Cannot set cost.reasoning when reasoning is false",
-      path: ["cost", "reasoning"],
-    },
-  );
+  .strict();
+
+export const AuthoredModelShape = z
+  .object({
+    ...ModelBase.shape,
+    cost: AuthoredCost.optional(),
+  })
+  .strict();
+
+export const Model = refineModel(ModelShape);
+
+export const AuthoredModel = refineModel(AuthoredModelShape);

 export type Model = z.infer<typeof Model>;

@@ -15,7 +15,8 @@ output = 25.000
 cache_read = 0.500
 cache_write = 6.250

-[cost.context_over_200k]
+[[cost.tiers]]
+tier = { size = 200_000 }
 input = 10.000
 output = 37.500
 cache_read = 1.000
@@ -16,7 +16,8 @@ output = 180.000
 cache_read = 0
 cache_write = 0

-[cost.context_over_200k]
+[[cost.tiers]]
+tier = { size = 272_000 }
 input = 60.000
 output = 270.000

@@ -16,7 +16,8 @@ output = 15.000
 cache_read = 0.250
 cache_write = 0

-[cost.context_over_200k]
+[[cost.tiers]]
+tier = { size = 272_000 }
 input = 5.000
 output = 22.500

@@ -0,0 +1,25 @@
+name = "Claude Opus 4.7 Thinking"
+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
+
+[cost]
+input = 5
+output = 25
+cache_read = 0.5
+cache_write = 6.25
+
+[limit]
+context = 200_000
+output = 32_000
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
@@ -15,7 +15,8 @@ output = 15.00
 cache_read = 0.30
 cache_write = 3.75

-[cost.context_over_200k]
+[[cost.tiers]]
+tier = { size = 200_000 }
 input = 6.00
 output = 22.50
 cache_read = 0.60
@@ -15,7 +15,8 @@ output = 15.00
 cache_read = 0.30
 cache_write = 3.75

-[cost.context_over_200k]
+[[cost.tiers]]
+tier = { size = 200_000 }
 input = 6.00
 output = 22.50
 cache_read = 0.60
@@ -1,24 +0,0 @@
-name = "Coding-GLM-5-Free"
-family = "glm"
-release_date = "2026-02-11"
-last_updated = "2026-02-11"
-attachment = false
-reasoning = true
-temperature = true
-tool_call = true
-open_weights = false
-
-[interleaved]
-field = "reasoning_content"
-
-[cost]
-input = 0.0
-output = 0.0
-
-[limit]
-context = 204800
-output = 131072
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,24 +1,25 @@
-name = "GLM5"
+name = "Coding GLM 5.1 (free)"
 family = "glm"
-release_date = "2026-02-12"
-last_updated = "2026-02-12"
+release_date = "2026-04-11"
+last_updated = "2026-04-11"
 attachment = false
 reasoning = true
 temperature = true
 tool_call = true
 structured_output = true
-open_weights = true
+open_weights = false

 [interleaved]
 field = "reasoning_content"

 [cost]
-input = 0.0
-output = 0.0
+input = 0
+output = 0
+cache_read = 0

 [limit]
-context = 202752
-output = 131000
+context = 204_800
+output = 128_000

 [modalities]
 input = ["text"]
@@ -1,23 +1,21 @@
-name = "MiniMax-M2.1"
+name = "Coding MiniMax M2.7 Highspeed"
 family = "minimax"
-release_date = "2025-12-23"
-last_updated = "2025-12-23"
+release_date = "2026-03-18"
+last_updated = "2026-03-18"
 attachment = false
 reasoning = true
 temperature = true
 tool_call = true
+structured_output = true
 open_weights = true

-[interleaved]
-field = "reasoning_details"
-
 [cost]
-input = 0.288
-output = 1.152
+input = 0.2
+output = 0.2

 [limit]
 context = 204_800
-output = 192_000
+output = 13_100

 [modalities]
 input = ["text"]
@@ -1,20 +1,21 @@
-name = "MiniMax-M2.1"
+name = "Coding MiniMax M2.7"
 family = "minimax"
-release_date = "2025-12-23"
-last_updated = "2025-12-23"
+release_date = "2026-03-18"
+last_updated = "2026-03-18"
 attachment = false
 reasoning = true
 temperature = true
 tool_call = true
+structured_output = true
 open_weights = true

 [cost]
-input = 0.0
-output = 0.0
+input = 0.2
+output = 0.2

 [limit]
 context = 204_800
-output = 131_072
+output = 13_100

 [modalities]
 input = ["text"]
@@ -1,4 +1,4 @@
-name = "Gemini 3.1 Flash Lite Preview"
+name = "Gemini 3.1 Flash Lite"
 family = "gemini-flash-lite"
 release_date = "2026-03-03"
 last_updated = "2026-03-03"
@@ -0,0 +1,30 @@
+name = "Gemini 3.1 Pro Preview Custom Tools"
+family = "gemini-pro"
+release_date = "2026-02-19"
+last_updated = "2026-02-19"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+knowledge = "2025-01"
+open_weights = false
+
+[cost]
+input = 2
+output = 12
+cache_read = 0.2
+
+[[cost.tiers]]
+tier = { size = 200_000 }
+input = 4
+output = 18
+cache_read = 0.4
+
+[limit]
+context = 1_048_576
+output = 65_536
+
+[modalities]
+input = ["text", "image", "audio", "video", "pdf"]
+output = ["text"]
@@ -1,25 +0,0 @@
-name = "GLM-5"
-family = "glm"
-release_date = "2026-02-11"
-last_updated = "2026-02-11"
-attachment = false
-reasoning = true
-temperature = true
-tool_call = true
-open_weights = true
-
-[interleaved]
-field = "reasoning_content"
-
-[cost]
-input = 0.88
-output = 2.816
-cache_read = 0.176
-
-[limit]
-context = 202_752
-output = 0
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -0,0 +1,26 @@
+name = "GLM 5 Vision Turbo"
+family = "glm"
+release_date = "2026-05-09"
+last_updated = "2026-05-09"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = false
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 0.7042
+output = 3.09848
+cache_read = 0.169008
+
+[limit]
+context = 200_000
+output = 128_000
+
+[modalities]
+input = ["text", "image", "video"]
+output = ["text"]
@@ -0,0 +1,29 @@
+name = "Grok 4.3"
+family = "grok"
+release_date = "2026-05-01"
+last_updated = "2026-05-01"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = false
+
+[cost]
+input = 1.25
+output = 2.5
+cache_read = 0.2
+
+[[cost.tiers]]
+tier = { size = 200_000 }
+input = 2.5
+output = 5.0
+cache_read = 0.4
+
+[limit]
+context = 1_000_000
+output = 1_000_000
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
@@ -1,24 +0,0 @@
-name = "Qwen3 Coder Plus"
-family = "qwen"
-release_date = "2025-07-23"
-last_updated = "2025-07-23"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = true
-knowledge = "2025-04"
-open_weights = true
-
-[cost]
-input = 0.137
-output = 0.548
-cache_read = 0.137
-
-[limit]
-context = 2_000_000
-output = 64_000
-input = 262_144
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,23 +0,0 @@
-name = "Qwen3 Max"
-family = "qwen"
-release_date = "2025-09-23"
-last_updated = "2025-09-23"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = true
-knowledge = "2025-04"
-open_weights = false
-
-[cost]
-input = 0.34246
-output = 1.36984
-cache_read = 0.34246
-
-[limit]
-context = 252_000
-output = 32_000
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,11 +1,12 @@
-name = "Qwen3.6 Plus"
+name = "Qwen3.6 Flash"
 family = "qwen"
 release_date = "2026-04-02"
 last_updated = "2026-04-02"
-attachment = false
+attachment = true
 reasoning = true
 temperature = true
 tool_call = true
+structured_output = true
 knowledge = "2025-04"
 open_weights = false

@@ -16,8 +17,8 @@ cache_read = 0.0169
 cache_write = 0.21125

 [limit]
-context = 1_000_000
-output = 65_536
+context = 991_000
+output = 64_000

 [modalities]
 input = ["text", "image", "video"]
@@ -0,0 +1,25 @@
+name = "Qwen3.6 Max Preview"
+family = "qwen"
+release_date = "2026-05-09"
+last_updated = "2026-05-09"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+knowledge = "2025-04"
+open_weights = false
+
+[cost]
+input = 1.268
+output = 7.608
+cache_read = 0.1268
+cache_write = 1.585
+
+[limit]
+context = 240_000
+output = 64_000
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -1,19 +1,20 @@
-name = "Qwen3.5 Plus"
+name = "Qwen3.6 Plus"
 family = "qwen"
-release_date = "2026-02-16"
-last_updated = "2026-02-16"
-attachment = false
+release_date = "2026-05-09"
+last_updated = "2026-05-09"
+attachment = true
 reasoning = true
 temperature = true
 tool_call = true
+structured_output = true
 knowledge = "2025-04"
 open_weights = false

 [cost]
-input = 0.1096
-output = 0.6576
-cache_read = 0.01096
-cache_write = 0.137
+input = 0.282
+output = 1.692
+cache_read = 0.0282
+cache_write = 0.3525

 [limit]
 context = 991_000
@@ -10,10 +10,17 @@ tool_call = true
 open_weights = false

 [cost]
-input = 0.276
-output = 1.651
-cache_read = 0.028
-cache_write = 0.344
+input = 0.50
+output = 3.00
+cache_read = 0.05
+cache_write = 0.625
+
+[[cost.tiers]]
+tier = { size = 256_000 }
+input = 2.00
+output = 6.00
+cache_read = 0.20
+cache_write = 2.50

 [limit]
 context = 1_000_000
@@ -10,10 +10,17 @@ tool_call = true
 open_weights = false

 [cost]
-input = 0.276
-output = 1.651
-cache_read = 0.028
-cache_write = 0.344
+input = 0.50
+output = 3.00
+cache_read = 0.05
+cache_write = 0.625
+
+[[cost.tiers]]
+tier = { size = 256_000 }
+input = 2.00
+output = 6.00
+cache_read = 0.20
+cache_write = 2.50

 [limit]
 context = 1_000_000
@@ -1,22 +0,0 @@
-name = "Nova Premier"
-family = "nova"
-release_date = "2024-12-03"
-last_updated = "2024-12-03"
-attachment = true
-reasoning = true
-temperature = true
-knowledge = "2024-10"
-tool_call = true
-open_weights = false
-
-[cost]
-input = 2.50
-output = 12.50
-
-[limit]
-context = 1_000_000
-output = 16_384
-
-[modalities]
-input = ["text", "image", "video"]
-output = ["text"]
@@ -1,2 +0,0 @@
-[extends]
-from = "anthropic/claude-3-5-haiku-20241022"
@@ -1,2 +0,0 @@
-[extends]
-from = "anthropic/claude-3-5-sonnet-20240620"
@@ -1,24 +0,0 @@
-name = "Claude Sonnet 3.5 v2"
-family = "claude-sonnet"
-release_date = "2024-10-22"
-last_updated = "2024-10-22"
-attachment = true
-reasoning = false
-temperature = true
-knowledge = "2024-04"
-tool_call = true
-open_weights = false
-
-[cost]
-input = 3.00
-output = 15.00
-cache_read = 0.30
-cache_write = 3.75
-
-[limit]
-context = 200_000
-output = 8_192
-
-[modalities]
-input = ["text", "image", "pdf"]
-output = ["text"]
@@ -1,24 +0,0 @@
-name = "Claude Sonnet 3.7"
-family = "claude-sonnet"
-release_date = "2025-02-19"
-last_updated = "2025-02-19"
-attachment = true
-reasoning = false
-temperature = true
-knowledge = "2024-04"
-tool_call = true
-open_weights = false
-
-[cost]
-input = 3.00
-output = 15.00
-cache_read = 0.30
-cache_write = 3.75
-
-[limit]
-context = 200_000
-output = 8_192
-
-[modalities]
-input = ["text", "image", "pdf"]
-output = ["text"]
@@ -1,22 +0,0 @@
-name = "Claude Haiku 3"
-family = "claude-haiku"
-release_date = "2024-03-13"
-last_updated = "2024-03-13"
-attachment = true
-reasoning = false
-temperature = true
-knowledge = "2024-02"
-tool_call = true
-open_weights = false
-
-[cost]
-input = 0.25
-output = 1.25
-
-[limit]
-context = 200_000
-output = 4_096
-
-[modalities]
-input = ["text", "image", "pdf"]
-output = ["text"]
@@ -1,2 +0,0 @@
-[extends]
-from = "anthropic/claude-opus-4-20250514"
@@ -1,2 +0,0 @@
-[extends]
-from = "anthropic/claude-sonnet-4-20250514"
@@ -0,0 +1,4 @@
+name = "Claude Haiku 4.5 (AU)"
+
+[extends]
+from = "anthropic/claude-haiku-4-5-20251001"
@@ -0,0 +1,4 @@
+name = "Claude Sonnet 4.5 (AU)"
+
+[extends]
+from = "anthropic/claude-sonnet-4-5"
@@ -1,4 +0,0 @@
-name = "Claude Sonnet 4 (EU)"
-
-[extends]
-from = "anthropic/claude-sonnet-4-20250514"
@@ -1,4 +0,0 @@
-name = "Claude Sonnet 4 (Global)"
-
-[extends]
-from = "anthropic/claude-sonnet-4-20250514"
@@ -0,0 +1,4 @@
+name = "Claude Opus 4.7 (JP)"
+
+[extends]
+from = "anthropic/claude-opus-4-7"
@@ -0,0 +1,4 @@
+name = "Claude Sonnet 4.5 (JP)"
+
+[extends]
+from = "anthropic/claude-sonnet-4-5"
@@ -0,0 +1,4 @@
+name = "Claude Sonnet 4.6 (JP)"
+
+[extends]
+from = "anthropic/claude-sonnet-4-6"
@@ -1,4 +0,0 @@
-name = "Claude Opus 4 (US)"
-
-[extends]
-from = "anthropic/claude-opus-4-20250514"
@@ -1,4 +0,0 @@
-name = "Claude Sonnet 4 (US)"
-
-[extends]
-from = "anthropic/claude-sonnet-4-20250514"
@@ -0,0 +1,4 @@
+name = "DeepSeek-R1 (US)"
+
+[extends]
+from = "amazon-bedrock/deepseek.r1-v1:0"
@@ -0,0 +1,4 @@
+name = "Llama 4 Maverick 17B Instruct (US)"
+
+[extends]
+from = "amazon-bedrock/meta.llama4-maverick-17b-instruct-v1:0"
@@ -0,0 +1,4 @@
+name = "Llama 4 Scout 17B Instruct (US)"
+
+[extends]
+from = "amazon-bedrock/meta.llama4-scout-17b-instruct-v1:0"
@@ -15,7 +15,8 @@ output = 25.00
 cache_read = 0.50
 cache_write = 6.25

-[cost.context_over_200k]
+[[cost.tiers]]
+tier = { size = 200_000 }
 input = 10.00
 output = 37.50
 cache_read = 1.00
@@ -15,7 +15,8 @@ output = 25.00
 cache_read = 0.50
 cache_write = 6.25

-[cost.context_over_200k]
+[[cost.tiers]]
+tier = { size = 200_000 }
 input = 10.00
 output = 37.50
 cache_read = 1.00
@@ -1,25 +0,0 @@
-name = "MiniMax M2.1 TEE"
-family = "minimax"
-release_date = "2025-12-29"
-last_updated = "2026-01-27"
-attachment = false
-reasoning = true
-temperature = true
-tool_call = true
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.27
-output = 1.12
-
-[limit]
-context = 196_608
-output = 65_536
-
-[modalities]
-input = ["text"]
-output = ["text"]
-
-[interleaved]
-field = "reasoning_content"
@@ -1,7 +1,9 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
 name = "MiniMax M2.5 TEE"
 family = "minimax"
 release_date = "2026-02-15"
-last_updated = "2026-02-15"
+last_updated = "2026-04-25"
 attachment = false
 reasoning = true
 temperature = true
@@ -10,9 +12,9 @@ structured_output = true
 open_weights = true

 [cost]
-input = 0.30
-output = 1.10
-cache_read = 0.15
+input = 0.15
+output = 1.2
+cache_read = 0.075

 [limit]
 context = 196_608
@@ -1,7 +1,9 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
 name = "DeepHermes 3 Mistral 24B Preview"
 family = "nousresearch"
 release_date = "2025-12-29"
-last_updated = "2026-01-10"
+last_updated = "2026-04-25"
 attachment = false
 reasoning = false
 temperature = true
@@ -10,8 +12,9 @@ structured_output = true
 open_weights = true

 [cost]
-input = 0.02
-output = 0.10
+input = 0.0245
+output = 0.0978
+cache_read = 0.01225

 [limit]
 context = 32_768
@@ -1,7 +1,9 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
 name = "Hermes 4 14B"
 family = "nousresearch"
 release_date = "2025-12-29"
-last_updated = "2026-01-10"
+last_updated = "2026-04-25"
 attachment = false
 reasoning = true
 temperature = true
@@ -10,8 +12,9 @@ structured_output = true
 open_weights = true

 [cost]
-input = 0.01
-output = 0.05
+input = 0.0136
+output = 0.0543
+cache_read = 0.0068

 [limit]
 context = 40_960
@@ -1,25 +0,0 @@
-name = "Hermes 4 405B FP8 TEE"
-family = "nousresearch"
-release_date = "2025-12-29"
-last_updated = "2026-01-10"
-attachment = false
-reasoning = true
-temperature = true
-tool_call = true
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.30
-output = 1.20
-
-[limit]
-context = 131_072
-output = 65_536
-
-[modalities]
-input = ["text"]
-output = ["text"]
-
-[interleaved]
-field = "reasoning_content"
@@ -1,22 +0,0 @@
-name = "InternVL3 78B TEE"
-family = "opengvlab"
-release_date = "2025-01-06"
-last_updated = "2026-01-10"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = false
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.10
-output = 0.39
-
-[limit]
-context = 32_768
-output = 32_768
-
-[modalities]
-input = ["text", "image"]
-output = ["text"]
@@ -1,7 +1,9 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
 name = "Qwen2.5 72B Instruct"
 family = "qwen"
 release_date = "2025-12-29"
-last_updated = "2026-01-10"
+last_updated = "2026-04-25"
 attachment = false
 reasoning = false
 temperature = true
@@ -10,8 +12,9 @@ structured_output = true
 open_weights = true

 [cost]
-input = 0.13
-output = 0.52
+input = 0.2989
+output = 1.1957
+cache_read = 0.14945

 [limit]
 context = 32_768
@@ -1,7 +1,9 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
 name = "Qwen2.5 Coder 32B Instruct"
 family = "qwen"
 release_date = "2025-12-29"
-last_updated = "2026-01-10"
+last_updated = "2026-04-25"
 attachment = false
 reasoning = false
 temperature = true
@@ -10,8 +12,9 @@ structured_output = true
 open_weights = true

 [cost]
-input = 0.03
-output = 0.11
+input = 0.0272
+output = 0.1087
+cache_read = 0.0136

 [limit]
 context = 32_768
@@ -1,8 +1,10 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
 name = "Qwen2.5 VL 32B Instruct"
 family = "qwen"
 release_date = "2025-12-29"
-last_updated = "2026-01-10"
-attachment = false
+last_updated = "2026-04-25"
+attachment = true
 reasoning = false
 temperature = true
 tool_call = false
@@ -10,8 +12,9 @@ structured_output = true
 open_weights = true

 [cost]
-input = 0.05
-output = 0.22
+input = 0.0543
+output = 0.2174
+cache_read = 0.02715

 [limit]
 context = 16_384
@@ -1,22 +0,0 @@
-name = "Qwen2.5 VL 72B Instruct TEE"
-family = "qwen"
-release_date = "2025-12-29"
-last_updated = "2026-01-10"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = false
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.15
-output = 0.60
-
-[limit]
-context = 32_768
-output = 32_768
-
-[modalities]
-input = ["text", "image"]
-output = ["text"]
@@ -1,25 +0,0 @@
-name = "Qwen3 14B"
-family = "qwen"
-release_date = "2025-12-29"
-last_updated = "2026-01-10"
-attachment = false
-reasoning = true
-temperature = true
-tool_call = true
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.05
-output = 0.22
-
-[limit]
-context = 40_960
-output = 40_960
-
-[modalities]
-input = ["text"]
-output = ["text"]
-
-[interleaved]
-field = "reasoning_content"
@@ -1,7 +1,9 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
 name = "Qwen3 235B A22B Instruct 2507 TEE"
 family = "qwen"
 release_date = "2025-12-29"
-last_updated = "2026-01-10"
+last_updated = "2026-04-25"
 attachment = false
 reasoning = false
 temperature = true
@@ -10,9 +12,9 @@ structured_output = true
 open_weights = true

 [cost]
-input = 0.08
-output = 0.55
-cache_read = 0.04
+input = 0.1
+output = 0.6
+cache_read = 0.05

 [limit]
 context = 262_144
@@ -1,7 +1,9 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
 name = "Qwen3 235B A22B Thinking 2507"
 family = "qwen"
 release_date = "2025-12-29"
-last_updated = "2026-01-10"
+last_updated = "2026-04-25"
 attachment = false
 reasoning = true
 temperature = true
@@ -11,7 +13,8 @@ open_weights = true

 [cost]
 input = 0.11
-output = 0.60
+output = 0.6
+cache_read = 0.055

 [limit]
 context = 262_144
@@ -1,25 +0,0 @@
-name = "Qwen3 235B A22B"
-family = "qwen"
-release_date = "2025-12-29"
-last_updated = "2026-01-10"
-attachment = false
-reasoning = true
-temperature = true
-tool_call = true
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.30
-output = 1.20
-
-[limit]
-context = 40_960
-output = 40_960
-
-[modalities]
-input = ["text"]
-output = ["text"]
-
-[interleaved]
-field = "reasoning_content"
@@ -1,22 +0,0 @@
-name = "Qwen3 30B A3B Instruct 2507"
-family = "qwen"
-release_date = "2025-12-29"
-last_updated = "2026-01-10"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = true
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.08
-output = 0.33
-
-[limit]
-context = 262_144
-output = 262_144
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,7 +1,9 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
 name = "Qwen3 30B A3B"
 family = "qwen"
 release_date = "2025-12-29"
-last_updated = "2026-01-10"
+last_updated = "2026-04-25"
 attachment = false
 reasoning = true
 temperature = true
@@ -12,6 +14,7 @@ open_weights = true
 [cost]
 input = 0.06
 output = 0.22
+cache_read = 0.03

 [limit]
 context = 40_960
@@ -1,7 +1,9 @@
-name = "Qwen3 32B"
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
+name = "Qwen3 32B TEE"
 family = "qwen"
-release_date = "2025-12-29"
-last_updated = "2026-01-10"
+release_date = "2026-04-25"
+last_updated = "2026-04-25"
 attachment = false
 reasoning = true
 temperature = true
@@ -21,6 +23,3 @@ output = 40_960
 [modalities]
 input = ["text"]
 output = ["text"]
-
-[interleaved]
-field = "reasoning_content"
@@ -1,23 +0,0 @@
-name = "Qwen3 Coder 480B A35B Instruct FP8 TEE"
-family = "qwen"
-release_date = "2025-12-29"
-last_updated = "2026-01-10"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = true
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.22
-output = 0.95
-cache_read = 0.11
-
-[limit]
-context = 262_144
-output = 262_144
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -0,0 +1,25 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
+name = "Qwen3 Coder Next TEE"
+family = "qwen"
+release_date = "2026-04-25"
+last_updated = "2026-04-25"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = true
+
+[cost]
+input = 0.12
+output = 0.75
+cache_read = 0.06
+
+[limit]
+context = 262_144
+output = 65_536
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -1,22 +0,0 @@
-name = "Qwen3 Coder Next"
-family = "qwen"
-release_date = "2026-02-05"
-last_updated = "2026-02-05"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = true
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.07
-output = 0.30
-
-[limit]
-context = 262_144
-output = 65_536
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,7 +1,9 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
 name = "Qwen3 Next 80B A3B Instruct"
 family = "qwen"
 release_date = "2025-12-29"
-last_updated = "2026-01-10"
+last_updated = "2026-04-25"
 attachment = false
 reasoning = false
 temperature = true
@@ -10,8 +12,9 @@ structured_output = true
 open_weights = true

 [cost]
-input = 0.10
-output = 0.80
+input = 0.1
+output = 0.8
+cache_read = 0.05

 [limit]
 context = 262_144
@@ -1,22 +0,0 @@
-name = "Qwen3 VL 235B A22B Instruct"
-family = "qwen"
-release_date = "2025-12-29"
-last_updated = "2026-01-10"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = true
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.30
-output = 1.20
-
-[limit]
-context = 262_144
-output = 262_144
-
-[modalities]
-input = ["text", "image"]
-output = ["text"]
@@ -1,7 +1,9 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
 name = "Qwen3.5 397B A17B TEE"
 family = "qwen"
 release_date = "2026-02-18"
-last_updated = "2026-02-18"
+last_updated = "2026-04-25"
 attachment = true
 reasoning = true
 temperature = true
@@ -0,0 +1,25 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
+name = "Qwen3.6 27B TEE"
+family = "qwen"
+release_date = "2026-04-25"
+last_updated = "2026-04-25"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = true
+
+[cost]
+input = 0.195
+output = 1.56
+cache_read = 0.0975
+
+[limit]
+context = 262_144
+output = 65_536
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
@@ -1,17 +1,18 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
 name = "Qwen3Guard Gen 0.6B"
 family = "qwen"
 release_date = "2025-12-29"
-last_updated = "2026-01-10"
+last_updated = "2026-04-25"
 attachment = false
 reasoning = false
 temperature = true
 tool_call = false
-structured_output = false
 open_weights = true

 [cost]
 input = 0.01
-output = 0.01
+output = 0.0109
 cache_read = 0.005

 [limit]
@@ -0,0 +1,25 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
+name = "MiMo V2 Flash TEE"
+family = "mimo"
+release_date = "2026-04-25"
+last_updated = "2026-04-25"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = true
+
+[cost]
+input = 0.09
+output = 0.29
+cache_read = 0.045
+
+[limit]
+context = 262_144
+output = 65_536
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -1,22 +0,0 @@
-name = "MiMo V2 Flash"
-family = "mimo"
-release_date = "2025-12-29"
-last_updated = "2026-01-27"
-attachment = false
-reasoning = true
-temperature = true
-tool_call = true
-structured_output = false
-open_weights = true
-
-[cost]
-input = 0.09
-output = 0.29
-
-[limit]
-context = 262_144
-output = 32_000
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,23 +0,0 @@
-name = "Mistral Small 3.1 24B Instruct 2503"
-family = "chutesai"
-release_date = "2025-12-29"
-last_updated = "2026-01-10"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = true
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.03
-output = 0.11
-cache_read = 0.015
-
-[limit]
-context = 131_072
-output = 131_072
-
-[modalities]
-input = ["text", "image"]
-output = ["text"]
@@ -1,22 +0,0 @@
-name = "Mistral Small 3.2 24B Instruct 2506"
-family = "chutesai"
-release_date = "2025-12-29"
-last_updated = "2026-01-10"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = true
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.06
-output = 0.18
-
-[limit]
-context = 131_072
-output = 131_072
-
-[modalities]
-input = ["text", "image"]
-output = ["text"]
@@ -1,7 +1,9 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
 name = "DeepSeek R1 0528 TEE"
 family = "deepseek-thinking"
 release_date = "2025-12-29"
-last_updated = "2026-01-10"
+last_updated = "2026-04-25"
 attachment = false
 reasoning = true
 temperature = true
@@ -10,8 +12,9 @@ structured_output = true
 open_weights = true

 [cost]
-input = 0.40
-output = 1.75
+input = 0.45
+output = 2.15
+cache_read = 0.225

 [limit]
 context = 163_840
@@ -1,7 +1,9 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
 name = "DeepSeek R1 Distill Llama 70B"
 family = "deepseek-thinking"
 release_date = "2025-12-29"
-last_updated = "2026-01-10"
+last_updated = "2026-04-25"
 attachment = false
 reasoning = true
 temperature = true
@@ -10,8 +12,9 @@ structured_output = true
 open_weights = true

 [cost]
-input = 0.03
-output = 0.11
+input = 0.0272
+output = 0.1087
+cache_read = 0.0136

 [limit]
 context = 131_072
@@ -1,25 +0,0 @@
-name = "DeepSeek R1 TEE"
-family = "deepseek-thinking"
-release_date = "2025-12-29"
-last_updated = "2026-01-10"
-attachment = false
-reasoning = true
-temperature = true
-tool_call = false
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.30
-output = 1.20
-
-[limit]
-context = 163_840
-output = 163_840
-
-[modalities]
-input = ["text"]
-output = ["text"]
-
-[interleaved]
-field = "reasoning_content"
@@ -1,7 +1,9 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
 name = "DeepSeek V3 0324 TEE"
 family = "deepseek"
 release_date = "2025-12-29"
-last_updated = "2026-01-10"
+last_updated = "2026-04-25"
 attachment = false
 reasoning = false
 temperature = true
@@ -10,9 +12,9 @@ structured_output = true
 open_weights = true

 [cost]
-input = 0.19
-output = 0.87
-cache_read = 0.095
+input = 0.25
+output = 1
+cache_read = 0.125

 [limit]
 context = 163_840
@@ -1,7 +1,9 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
 name = "DeepSeek V3.1 TEE"
 family = "deepseek"
 release_date = "2025-12-29"
-last_updated = "2026-01-10"
+last_updated = "2026-04-25"
 attachment = false
 reasoning = true
 temperature = true
@@ -10,8 +12,9 @@ structured_output = true
 open_weights = true

 [cost]
-input = 0.20
-output = 0.80
+input = 0.27
+output = 1
+cache_read = 0.135

 [limit]
 context = 163_840
@@ -1,25 +0,0 @@
-name = "DeepSeek V3.1 Terminus TEE"
-family = "deepseek"
-release_date = "2025-12-29"
-last_updated = "2026-01-10"
-attachment = false
-reasoning = true
-temperature = true
-tool_call = true
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.23
-output = 0.90
-
-[limit]
-context = 163_840
-output = 65_536
-
-[modalities]
-input = ["text"]
-output = ["text"]
-
-[interleaved]
-field = "reasoning_content"
@@ -1,25 +0,0 @@
-name = "DeepSeek V3.2 Speciale TEE"
-family = "deepseek"
-release_date = "2025-12-29"
-last_updated = "2026-01-10"
-attachment = false
-reasoning = true
-temperature = true
-tool_call = false
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.27
-output = 0.41
-
-[limit]
-context = 163_840
-output = 65_536
-
-[modalities]
-input = ["text"]
-output = ["text"]
-
-[interleaved]
-field = "reasoning_content"
@@ -1,7 +1,9 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
 name = "DeepSeek V3.2 TEE"
 family = "deepseek"
 release_date = "2025-12-29"
-last_updated = "2026-01-10"
+last_updated = "2026-04-25"
 attachment = false
 reasoning = true
 temperature = true
@@ -1,22 +0,0 @@
-name = "DeepSeek V3"
-family = "deepseek"
-release_date = "2025-12-29"
-last_updated = "2026-01-10"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = false
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.30
-output = 1.20
-
-[limit]
-context = 163_840
-output = 163_840
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -0,0 +1,25 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
+name = "gemma 4 31B turbo TEE"
+family = "gemma"
+release_date = "2026-04-25"
+last_updated = "2026-04-25"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = true
+
+[cost]
+input = 0.13
+output = 0.38
+cache_read = 0.065
+
+[limit]
+context = 131_072
+output = 65_536
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
@@ -1,22 +0,0 @@
-name = "MiroThinker V1.5 235B"
-release_date = "2026-01-10"
-last_updated = "2026-01-10"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = false
-structured_output = false
-open_weights = true
-
-[cost]
-input = 0.30
-output = 1.20
-cache_read = 0.15
-
-[limit]
-context = 262_144
-output = 8_192
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,23 +0,0 @@
-name = "Kimi K2 Instruct 0905"
-family = "kimi"
-release_date = "2025-12-29"
-last_updated = "2026-01-10"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = true
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.39
-output = 1.90
-cache_read = 0.195
-
-[limit]
-context = 262_144
-output = 262_144
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,21 +1,21 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
 name = "Kimi K2.5 TEE"
 family = "kimi"
 release_date = "2026-01-27"
-last_updated = "2026-01-27"
-attachment = false
+last_updated = "2026-04-25"
+attachment = true
 reasoning = true
 temperature = true
 tool_call = true
-knowledge = "2024-10"
-open_weights = true
 structured_output = true
-
-[interleaved]
-field = "reasoning_content"
+open_weights = true
+knowledge = "2024-10"

 [cost]
-input = 0.60
-output = 3.00
+input = 0.44
+output = 2
+cache_read = 0.22

 [limit]
 context = 262_144
@@ -24,3 +24,6 @@ output = 65_535
 [modalities]
 input = ["text", "image", "video"]
 output = ["text"]
+
+[interleaved]
+field = "reasoning_content"
@@ -1,26 +1,29 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
 name = "Kimi K2.6 TEE"
 family = "kimi"
 release_date = "2026-04-20"
-last_updated = "2026-04-23"
+last_updated = "2026-04-25"
 attachment = true
 reasoning = true
 temperature = true
 tool_call = true
-knowledge = "2025-12"
-open_weights = true
 structured_output = true
-
-[interleaved]
-field = "reasoning_content"
+open_weights = true
+knowledge = "2025-12"

 [cost]
-input = 0.44
-output = 2.00
+input = 0.95
+output = 4
+cache_read = 0.475

 [limit]
 context = 262_144
-output = 262_144
+output = 65_535

 [modalities]
 input = ["text", "image", "video"]
 output = ["text"]
+
+[interleaved]
+field = "reasoning_content"
@@ -1,22 +0,0 @@
-name = "NVIDIA Nemotron 3 Nano 30B A3B BF16"
-family = "nemotron"
-release_date = "2025-12-29"
-last_updated = "2026-01-10"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = true
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.06
-output = 0.24
-
-[limit]
-context = 262_144
-output = 262_144
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,7 +1,9 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
 name = "gpt oss 120b TEE"
 family = "gpt-oss"
 release_date = "2025-12-29"
-last_updated = "2026-01-10"
+last_updated = "2026-04-25"
 attachment = false
 reasoning = true
 temperature = true
@@ -10,8 +12,9 @@ structured_output = true
 open_weights = true

 [cost]
-input = 0.04
-output = 0.18
+input = 0.09
+output = 0.36
+cache_read = 0.045

 [limit]
 context = 131_072
@@ -1,8 +1,10 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
 name = "dots.ocr"
 family = "rednote"
 release_date = "2025-12-29"
-last_updated = "2026-01-10"
-attachment = false
+last_updated = "2026-04-25"
+attachment = true
 reasoning = false
 temperature = true
 tool_call = false
@@ -11,7 +13,7 @@ open_weights = true

 [cost]
 input = 0.01
-output = 0.01
+output = 0.0109
 cache_read = 0.005

 [limit]
@@ -1,25 +0,0 @@
-name = "DeepSeek R1T Chimera"
-family = "tngtech"
-release_date = "2025-12-29"
-last_updated = "2026-01-10"
-attachment = false
-reasoning = true
-temperature = true
-tool_call = false
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.30
-output = 1.20
-
-[limit]
-context = 163_840
-output = 163_840
-
-[modalities]
-input = ["text"]
-output = ["text"]
-
-[interleaved]
-field = "reasoning_content"
@@ -0,0 +1,25 @@
+# Auto-generated by generate-chutes.ts — do not edit pricing, limits, or capabilities.
+# Manual overrides preserved on re-run: name, family, knowledge, interleaved, status
+name = "DeepSeek TNG R1T2 Chimera TEE"
+family = "deepseek"
+release_date = "2026-04-25"
+last_updated = "2026-04-25"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = true
+
+[cost]
+input = 0.3
+output = 1.1
+cache_read = 0.15
+
+[limit]
+context = 163_840
+output = 163_840
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -1,25 +0,0 @@
-name = "DeepSeek TNG R1T2 Chimera"
-family = "tngtech"
-release_date = "2025-12-29"
-last_updated = "2026-01-10"
-attachment = false
-reasoning = true
-temperature = true
-tool_call = true
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.25
-output = 0.85
-
-[limit]
-context = 163_840
-output = 163_840
-
-[modalities]
-input = ["text"]
-output = ["text"]
-
-[interleaved]
-field = "reasoning_content"
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Aiden Cline	458b7f4d1a	use Venice context tier thresholds	2026-05-12 17:39:40 -05:00
Aiden Cline	8f9adc7567	fix generated tier change detection	2026-05-12 17:01:35 -05:00
Aiden Cline	a671cc05d5	align cost tiers with model schema	2026-05-12 16:01:49 -05:00
Aiden Cline	2e015de42d	preserve generated tier thresholds	2026-05-11 17:07:19 -05:00
Aiden Cline	151e9c9071	fix tiered cost generation	2026-05-11 16:43:32 -05:00
Aiden Cline	b96b074a3b	fix long-context cost tier omissions	2026-05-11 16:31:07 -05:00
Aiden Cline	2593e131a1	wip	2026-05-11 16:11:19 -05:00
Frank	bc95b42ccd	update zen models	2026-05-11 11:26:08 -04:00
Aiden Cline	6139fb8c69	Merge pull request #1598 from mugnimaestra/feat/chutes-generate-script feat(chutes): add API-driven model generator script	2026-05-11 09:30:38 -05:00
Aiden Cline	3070758007	Merge pull request #1678 from 5kahoisaac/chore/nvidia-models Sync NVIDIA endpoint model catalog	2026-05-11 09:29:54 -05:00
Frank	5525e83de4	update zen models	2026-05-11 10:00:09 -04:00
Frank	359fd879b8	update zen models	2026-05-10 03:54:03 -04:00
Frank	01b5a1a656	update zen models	2026-05-10 02:52:44 -04:00
Frank	b1958be099	update zen models	2026-05-10 02:42:44 -04:00
Aiden Cline	08aa068523	Temporarily remove kiro provider and models	2026-05-10 01:19:47 -05:00
Aiden Cline	f31ad0b02f	Merge pull request #1738 from mattiacerutti/chore/remove-gh-copilot-deprecated chore(copilot): mark deprecated models	2026-05-09 15:42:19 -05:00
Aiden Cline	585aa7fa1b	Merge pull request #1741 from EriDeLee/dev Update aihubmix models	2026-05-09 15:42:03 -05:00
Aiden Cline	c42a327b3e	Merge pull request #1745 from mads-digitial-solutions/patch-1 Update Google provider docs url from pricing page to models page	2026-05-09 15:41:51 -05:00
mads-digitial-solutions	92ebbfb5c4	Update provider.toml Update Google provider docs URL from the pricing page to the models page	2026-05-09 19:49:19 +01:00
Aiden Cline	535fe8c971	Merge pull request #1744 from OpeOginni/fix/bedrock-model-ids chore(bedrock): Getting rid of legacy Amazon Bedrock model offerings	2026-05-09 13:48:52 -05:00
OpeOginni	a3b4bfc16c	fix(bedrock): remove uneeded model configurations	2026-05-09 20:35:34 +02:00
OpeOginni	d0fcd6f11f	fix(bedrock): align models with current docs	2026-05-09 20:29:11 +02:00
OpeOginni	e55cd54218	fix(bedrock): remove legacy model entries	2026-05-09 20:21:33 +02:00
OpeOginni	0d73b82b9f	fix(bedrock): restore regional model IDs	2026-05-09 20:18:59 +02:00
Aiden Cline	83c7e2b63f	Merge pull request #1742 from Adam8234/add-firepass-provider feat: add Fireworks (Firepass) provider	2026-05-09 12:45:09 -05:00
Adam	83ae4cf813	feat: add Fireworks (Firepass) provider Adds the Fireworks AI Firepass subscription provider. - Provider uses a dedicated FIREPASS_API_KEY - Uses @ai-sdk/openai-compatible SDK - Includes Kimi K2.6 Turbo (accounts/fireworks/routers/kimi-k2p6-turbo) - Zero per-token cost since covered by subscription	2026-05-09 12:22:13 -05:00
EriDeLee	77eae6eef7	Update aihubmix models	2026-05-09 20:53:57 +08:00
Aiden Cline	8cbf6ed10e	Merge pull request #1736 from vercel/update-vercel-models-1778258030 Update Vercel models	2026-05-08 21:58:40 -05:00
Mattia Cerutti	06d87e4411	chore(copilot): remove deprecated models	2026-05-09 00:19:50 +02:00
Frank	2cb3832618	update zen models	2026-05-08 17:11:05 -04:00
github-actions[bot]	df960d1a90	chore(vercel): update Vercel model definitions Auto-generated by weekly workflow from Vercel AI Gateway API. Co-Authored-By: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>	2026-05-08 16:33:52 +00:00
Aiden Cline	8f2f83ef61	Merge pull request #1735 from slpdy/dev Create DeepSeek-V4-Pro.toml	2026-05-08 10:59:33 -05:00
Aiden Cline	b133426465	Merge pull request #1734 from oskarkocol/chore/update-novita-deepseek-prices chore: update novita deepseek-v4-pro prices	2026-05-08 10:59:25 -05:00
Aiden Cline	dafff5a770	Merge pull request #1730 from smakosh/add-llmgateway-models Add new LLM Gateway text models (gpt-5.5, grok-4-3, gemini-3.1-flash-lite, qwen3.6, MiMo v2)	2026-05-08 10:58:20 -05:00
smakosh	dd894f077f	Merge remote-tracking branch 'upstream/dev' into add-llmgateway-models # Conflicts: # providers/google/models/gemini-3.1-flash-lite.toml	2026-05-08 17:46:52 +02:00
smakosh	91590874e7	Revert "fix(models): use canonical entries for qwen3.6-max-preview and grok-4.3" This reverts commit `70ac6fccda`.	2026-05-08 17:43:07 +02:00
Jj	a436236146	Create DeepSeek-V4-Pro.toml Added DeepSeek-v4-Pro model to Nebius provider	2026-05-08 11:03:16 +01:00
oskar	1415b4be97	chore: update novita deepseek prices	2026-05-08 14:21:27 +07:00
smakosh	70ac6fccda	fix(models): use canonical entries for qwen3.6-max-preview and grok-4.3 Apply the canonical TOML provided by the LLM Gateway team for the Qwen3.6 Max Preview and Grok 4.3 parent definitions, replacing the upstream-merged variants whose dates and pricing did not match. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-07 22:34:01 +02:00
smakosh	dc3283417d	Merge remote-tracking branch 'upstream/dev' into add-llmgateway-models # Conflicts: # providers/alibaba/models/qwen3.6-max-preview.toml # providers/llmgateway/models/qwen3.6-max-preview.toml	2026-05-07 22:28:23 +02:00
smakosh	34fd6673e5	chore(llmgateway): add new text models from llmgateway catalog Add gemini-3.1-flash-lite, grok-4-3, gpt-5.5, gpt-5.5-pro, qwen3.6 and MiMo v2 models that exist in llmgateway.io but were missing from models.dev. Adds parent definitions for grok-4-3, gemini-3.1-flash-lite, and qwen3.6-max-preview where they did not already exist. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-05-07 22:20:52 +02:00
Isaac Ng	c5fbcc2c9b	📦 CHORE: remove senera	2026-05-03 16:17:37 +08:00
Isaac Ng	ab2eb51b4e	chore(nvidia): align Nemotron endpoint slugs Replace stale NVIDIA Nemotron entries with the live Build catalog slugs so the local provider catalog matches current free and partner endpoints.	2026-05-03 15:47:19 +08:00
Isaac Ng	8e19ec580c	📦 CHORE: sync latest nvidia model	2026-05-03 15:19:17 +08:00
Isaac Ng	3aecc94c46	chore(nvidia): sync endpoint model catalog Update NVIDIA model TOMLs to match the live Build endpoint list by removing stale entries and adding missing ones. This keeps the provider catalog aligned with the current free and partner endpoint inventory.	2026-05-03 14:47:30 +08:00
Muhammad Mugni Hadi	dbe92646c3	chore(chutes): add header comments to generated TOML files Each generated TOML now includes a comment noting which fields are auto-managed vs manually overridable on re-run.	2026-04-26 06:52:21 +07:00
Muhammad Mugni Hadi	4717c67054	feat(chutes): add API-driven model generator script Add generate-chutes.ts that fetches models from https://llm.chutes.ai/v1/models and generates/updates TOML files, following the same pattern as generate-vercel.ts. Supports --dry-run, --new-only, and --keep-orphans flags. Auto-deletes TOML files for models no longer in the API (with empty directory cleanup). Preserves manually-set fields (family, knowledge, interleaved, status) when merging with API data. Also syncs current models from the API.	2026-04-26 06:51:16 +07:00