feat: add llmgateway.io model sync provider

Add a sync provider for the LLM Gateway (llmgateway.io) aggregator, mirroring its public /v1/models catalog into providers/llmgateway. The gateway exposes an OpenRouter-shaped response, but its supported_parameters and modality data are noisy (it omits "tools" for flagship models yet lists "temperature" for ones marked temperature=false). So the gateway is treated as authoritative only for the volatile, gateway-specific data — cost and served limits — while capability and modality fields stay curated (preserved from the existing entry, which a factored model inherits from its base). Only text-output models are synced. - packages/core/src/sync/providers/llmgateway.ts: new provider - packages/core/src/sync/index.ts: register in providers + aggregators - package.json: add llmgateway:sync script - .github/workflows/sync-models.yml: optional LLMGATEWAY_API_KEY Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-24 17:01:30 +01:00
parent 2a005dcfd5
commit 9629a104d5
101 changed files with 561 additions and 480 deletions
@@ -67,6 +67,7 @@ jobs:
          HF_TOKEN: ${{ secrets.HF_TOKEN }}
          OPENROUTER_API_KEY: ${{ secrets.OPENROUTER_API_KEY }}
          VENICE_API_KEY: ${{ secrets.VENICE_API_KEY }}
+          LLMGATEWAY_API_KEY: ${{ secrets.LLMGATEWAY_API_KEY }}
          GOOGLE_API_KEY: ${{ secrets.GOOGLE_API_KEY }}
          GEMINI_API_KEY: ${{ secrets.GEMINI_API_KEY }}
          GOOGLE_GENERATIVE_AI_API_KEY: ${{ secrets.GOOGLE_GENERATIVE_AI_API_KEY }}
@@ -24,6 +24,7 @@
    "databricks:generate": "bun ./packages/core/script/generate-databricks.ts",
    "helicone:generate": "bun ./packages/core/script/generate-helicone.ts",
    "huggingface:sync": "bun ./packages/core/script/sync-models.ts huggingface",
+    "llmgateway:sync": "bun ./packages/core/script/sync-models.ts llmgateway",
    "venice:sync": "bun ./packages/core/script/sync-models.ts venice",
    "vercel:generate": "bun ./packages/core/script/sync-models.ts vercel",
    "wandb:generate": "bun ./packages/core/script/generate-wandb.ts",
@@ -8,6 +8,7 @@ import { baseten } from "./providers/baseten.js";
 import { cloudflareWorkersAi } from "./providers/cloudflare-workers-ai.js";
 import { google } from "./providers/google.js";
 import { huggingface } from "./providers/huggingface.js";
+import { llmgateway } from "./providers/llmgateway.js";
 import { openrouter } from "./providers/openrouter.js";
 import { ovhcloud } from "./providers/ovhcloud.js";
 import { vercel } from "./providers/vercel.js";
@@ -82,6 +83,7 @@ export const providers: {
  "cloudflare-workers-ai": SyncProvider<any>;
  google: SyncProvider<any>;
  huggingface: SyncProvider<any>;
+  llmgateway: SyncProvider<any>;
  openrouter: SyncProvider<any>;
  ovhcloud: SyncProvider<any>;
  vercel: SyncProvider<any>;
@@ -92,6 +94,7 @@ export const providers: {
  "cloudflare-workers-ai": cloudflareWorkersAi,
  google,
  huggingface,
+  llmgateway,
  openrouter,
  ovhcloud,
  vercel,
@@ -100,7 +103,7 @@ export const providers: {
 };

 export const groups = {
-  aggregators: ["huggingface", "openrouter", "vercel"],
+  aggregators: ["huggingface", "llmgateway", "openrouter", "vercel"],
  cloudflare: ["cloudflare-workers-ai"],
  direct: ["baseten", "google", "ovhcloud", "venice", "xai"],
 } as const;
@@ -0,0 +1,217 @@
+import { z } from "zod";
+
+import { inferKimiFamily, ModelFamilyValues } from "../../family.js";
+import type { ExistingModel, SyncProvider, SyncedFullModel, SyncedModel } from "../index.js";
+import { factorBaseModel } from "./openrouter.js";
+
+const API_ENDPOINT = "https://api.llmgateway.io/v1/models";
+
+const Pricing = z.object({
+  prompt: z.string().optional(),
+  completion: z.string().optional(),
+  internal_reasoning: z.string().optional(),
+  input_cache_read: z.string().optional(),
+  input_cache_write: z.string().optional(),
+});
+
+export const LLMGatewayModel = z.object({
+  id: z.string(),
+  name: z.string(),
+  created: z.number(),
+  family: z.string().optional(),
+  architecture: z.object({
+    input_modalities: z.array(z.string()),
+    output_modalities: z.array(z.string()),
+  }),
+  pricing: Pricing,
+  context_length: z.number(),
+  supported_parameters: z.array(z.string()),
+  structured_outputs: z.boolean().optional(),
+}).passthrough();
+
+export const LLMGatewayResponse = z.object({
+  data: z.array(LLMGatewayModel),
+}).passthrough();
+
+export type LLMGatewayModel = z.infer<typeof LLMGatewayModel>;
+
+export const llmgateway = {
+  id: "llmgateway",
+  name: "LLM Gateway",
+  modelsDir: "providers/llmgateway/models",
+  async fetchModels() {
+    const headers = process.env.LLMGATEWAY_API_KEY
+      ? { Authorization: `Bearer ${process.env.LLMGATEWAY_API_KEY}` }
+      : undefined;
+    const response = await fetch(API_ENDPOINT, { headers });
+    if (!response.ok) {
+      throw new Error(`LLM Gateway request failed: ${response.status} ${response.statusText}`);
+    }
+    return response.json();
+  },
+  parseModels(raw) {
+    return LLMGatewayResponse.parse(raw).data.filter((model) => {
+      const output = model.architecture.output_modalities;
+      return output.length === 1 && output[0] === "text";
+    });
+  },
+  translateModel(model, context) {
+    return {
+      id: model.id,
+      model: buildLLMGatewayModel(model, context.existing(model.id)),
+    };
+  },
+} satisfies SyncProvider<LLMGatewayModel>;
+
+function dateFromTimestamp(timestamp: number) {
+  return new Date(timestamp * 1000).toISOString().slice(0, 10);
+}
+
+function price(value: string | undefined) {
+  if (value === undefined) return undefined;
+  const number = Number(value);
+  return Number.isFinite(number) && number >= 0
+    ? Math.round(number * 1_000_000_000_000) / 1_000_000
+    : undefined;
+}
+
+// Cache/reasoning prices are reported as "0" when the gateway has no data; treat
+// those as unknown so we never downgrade a hand-authored value to zero.
+function nonZeroPrice(value: string | undefined) {
+  const result = price(value);
+  return result !== undefined && result > 0 ? result : undefined;
+}
+
+type Modality = "text" | "audio" | "image" | "video" | "pdf";
+
+function modalities(values: string[], fallback: Modality[]): Modality[] {
+  const allowed = new Set<Modality>(["text", "audio", "image", "video", "pdf"]);
+  const result = values
+    .map((value) => value.toLowerCase())
+    .map((value) => (value === "file" ? "pdf" : value))
+    .filter((value): value is Modality => allowed.has(value as Modality));
+  return [...new Set(result.length > 0 ? result : fallback)];
+}
+
+function inferFamily(model: LLMGatewayModel, name: string) {
+  const kimiFamily = inferKimiFamily(model.id, name);
+  if (kimiFamily !== undefined) return kimiFamily;
+
+  const target = `${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 buildLLMGatewayModel(
+  model: LLMGatewayModel,
+  existing: ExistingModel | undefined,
+): SyncedModel {
+  const prompt = price(model.pricing.prompt);
+  const completion = price(model.pricing.completion);
+  const reasoning = model.supported_parameters.includes("reasoning")
+    || model.supported_parameters.includes("include_reasoning");
+  const context = model.context_length > 0
+    ? model.context_length
+    : existing?.limit?.context ?? model.context_length;
+
+  // The gateway is authoritative for the volatile, gateway-specific data — cost
+  // and served limits. Its supported_parameters / modalities are too noisy to
+  // drive capability fields (it omits "tools" for flagship models yet lists
+  // "temperature" for ones the catalog deliberately marks temperature=false),
+  // so those stay curated: preserved from the existing entry (which, for a
+  // factored model, inherits its base when the field is absent).
+  const cost = prompt !== undefined && completion !== undefined
+    ? {
+        input: prompt,
+        output: completion,
+        reasoning: reasoning ? nonZeroPrice(model.pricing.internal_reasoning) ?? existing?.cost?.reasoning : existing?.cost?.reasoning,
+        cache_read: nonZeroPrice(model.pricing.input_cache_read) ?? existing?.cost?.cache_read,
+        cache_write: nonZeroPrice(model.pricing.input_cache_write) ?? existing?.cost?.cache_write,
+        tiers: existing?.cost?.tiers,
+      }
+    : existing?.cost;
+  const limit = {
+    context,
+    input: existing?.limit?.input,
+    output: existing?.limit?.output ?? context,
+  };
+
+  // Existing factored model: refresh cost + limit, keep every authored override
+  // as-is (undefined fields keep inheriting the base model).
+  if (existing?.base_model !== undefined) {
+    return factorBaseModel(
+      existing.base_model,
+      {
+        attachment: existing.attachment,
+        reasoning: existing.reasoning,
+        temperature: existing.temperature,
+        tool_call: existing.tool_call,
+        structured_output: existing.structured_output,
+        status: existing.status,
+        interleaved: existing.interleaved,
+        knowledge: existing.knowledge,
+        modalities: existing.modalities,
+        limit,
+        cost,
+      },
+      limit,
+      existing.base_model_omit,
+    );
+  }
+
+  // Existing full model: refresh cost + limit, preserve curated metadata.
+  if (existing !== undefined) {
+    return {
+      name: existing.name ?? model.name,
+      family: existing.family,
+      release_date: existing.release_date ?? dateFromTimestamp(model.created),
+      last_updated: existing.last_updated ?? dateFromTimestamp(model.created),
+      attachment: existing.attachment ?? false,
+      reasoning: existing.reasoning ?? false,
+      temperature: existing.temperature ?? false,
+      tool_call: existing.tool_call ?? false,
+      structured_output: existing.structured_output,
+      knowledge: existing.knowledge,
+      open_weights: existing.open_weights ?? false,
+      status: existing.status,
+      interleaved: existing.interleaved,
+      cost,
+      limit,
+      modalities: existing.modalities ?? defaultModalities(model),
+    } satisfies SyncedFullModel;
+  }
+
+  // Brand-new model: best-effort translation from the gateway. Capability and
+  // modality data are unreliable here and should be hand-reviewed.
+  const { input, output } = defaultModalities(model);
+  return {
+    name: model.name,
+    family: inferFamily(model, model.name),
+    release_date: dateFromTimestamp(model.created),
+    last_updated: dateFromTimestamp(model.created),
+    attachment: input.some((value) => value !== "text"),
+    reasoning,
+    temperature: model.supported_parameters.includes("temperature"),
+    tool_call: model.supported_parameters.includes("tools")
+      || model.supported_parameters.includes("tool_choice"),
+    structured_output: model.structured_outputs ?? false,
+    open_weights: false,
+    cost,
+    limit,
+    modalities: { input, output },
+  } satisfies SyncedFullModel;
+}
+
+function defaultModalities(model: LLMGatewayModel) {
+  return {
+    input: modalities(model.architecture.input_modalities, ["text"]),
+    output: modalities(model.architecture.output_modalities, ["text"]),
+  };
+}
@@ -1,24 +0,0 @@
-name = "Claude 3.5 Haiku"
-family = "claude"
-release_date = "2024-10-22"
-last_updated = "2024-10-22"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = true
-structured_output = false
-open_weights = false
-status = "deprecated"
-
-[cost]
-input = 0.80
-output = 4.00
-cache_read = 0.08
-
-[limit]
-context = 200_000
-output = 8_192
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,7 +0,0 @@
-base_model = "anthropic/claude-3-5-sonnet-20241022"
-
-[cost]
-input = 3
-output = 15
-cache_read = 0.3
-cache_write = 3.75
@@ -10,9 +10,10 @@ structured_output = false
 open_weights = false

 [cost]
-input = 3.00
-output = 15.00
-cache_read = 0.30
+input = 3
+output = 15
+cache_read = 0.3
+cache_write = 3.75

 [limit]
 context = 200_000
@@ -20,4 +21,4 @@ output = 8_192

 [modalities]
 input = ["text"]
-output = ["text"]
+output = ["text"]
@@ -10,9 +10,10 @@ structured_output = false
 open_weights = false

 [cost]
-input = 15.00
-output = 75.00
-cache_read = 1.50
+input = 15
+output = 75
+cache_read = 1.5
+cache_write = 18.75

 [limit]
 context = 200_000
@@ -20,4 +21,4 @@ output = 4_096

 [modalities]
 input = ["text", "image"]
-output = ["text"]
+output = ["text"]
@@ -1,7 +0,0 @@
-base_model = "anthropic/claude-opus-4-20250514"
-
-[cost]
-input = 15
-output = 75
-cache_read = 1.5
-cache_write = 18.75
@@ -1,7 +0,0 @@
-base_model = "anthropic/claude-sonnet-4-20250514"
-
-[cost]
-input = 3
-output = 15
-cache_read = 0.3
-cache_write = 3.75
@@ -1,23 +0,0 @@
-name = "DeepSeek R1 (0528)"
-family = "deepseek"
-release_date = "2025-05-28"
-last_updated = "2025-05-28"
-attachment = false
-reasoning = true
-temperature = true
-tool_call = false
-structured_output = false
-open_weights = true
-status = "beta"
-
-[cost]
-input = 0.55
-output = 2.19
-
-[limit]
-context = 64_000
-output = 16_384
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -12,7 +12,7 @@ open_weights = true
 [cost]
 input = 0.56
 output = 1.68
-cache_read = 0.07
+cache_read = 0.112

 [limit]
 context = 128_000
@@ -12,7 +12,7 @@ open_weights = true
 [cost]
 input = 0.28
 output = 0.42
-cache_read = 0.056
+cache_read = 0.028

 [limit]
 context = 163_840
@@ -7,3 +7,6 @@ field = "reasoning_content"
 input = 0.14
 output = 0.28
 cache_read = 0.0028
+
+[limit]
+context = 1_050_000
@@ -7,3 +7,6 @@ field = "reasoning_content"
 input = 0.435
 output = 0.87
 cache_read = 0.003625
+
+[limit]
+context = 1_050_000
@@ -4,3 +4,6 @@ status = "deprecated"
 [cost]
 input = 0.1
 output = 0.3
+
+[limit]
+context = 131_072
@@ -0,0 +1,22 @@
+name = "Fugu Ultra"
+release_date = "2026-06-22"
+last_updated = "2026-06-22"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = false
+open_weights = false
+
+[cost]
+input = 5
+output = 30
+cache_read = 0.5
+
+[limit]
+context = 1_000_000
+output = 1_000_000
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
@@ -1,5 +0,0 @@
-base_model = "google/gemini-2.0-flash-lite"
-
-[cost]
-input = 0.075
-output = 0.3
@@ -1,6 +0,0 @@
-base_model = "google/gemini-2.0-flash"
-
-[cost]
-input = 0.1
-output = 0.4
-cache_read = 0.025
@@ -0,0 +1,23 @@
+name = "Gemini 2.5 Flash Lite Preview (09-2025)"
+family = "gemini"
+release_date = "2025-09-25"
+last_updated = "2025-09-25"
+attachment = true
+reasoning = false
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = false
+
+[cost]
+input = 0.1
+output = 0.4
+cache_read = 0.01
+
+[limit]
+context = 1_048_576
+output = 1_048_576
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
@@ -4,4 +4,3 @@ base_model = "google/gemini-2.5-flash-lite"
 input = 0.1
 output = 0.4
 cache_read = 0.01
-input_audio = 0.3
@@ -4,4 +4,3 @@ base_model = "google/gemini-2.5-flash"
 input = 0.3
 output = 2.5
 cache_read = 0.03
-input_audio = 1
@@ -4,4 +4,3 @@ base_model = "google/gemini-3-flash-preview"
 input = 0.5
 output = 3
 cache_read = 0.05
-input_audio = 1
@@ -1,7 +0,0 @@
-base_model = "google/gemini-3.1-flash-lite-preview"
-
-[cost]
-input = 0.25
-output = 1.5
-cache_read = 0.025
-input_audio = 0.5
@@ -4,4 +4,4 @@ base_model = "google/gemini-3.1-flash-lite"
 input = 0.25
 output = 1.5
 cache_read = 0.025
-input_audio = 0.5
+cache_write = 0.08333
@@ -4,4 +4,4 @@ base_model = "google/gemini-3.5-flash"
 input = 1.5
 output = 9
 cache_read = 0.15
-input_audio = 1.5
+cache_write = 0.08333
@@ -1,22 +0,0 @@
-name = "Gemma 2 27B IT"
-family = "gemma"
-release_date = "2024-06-27"
-last_updated = "2024-06-27"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = false
-structured_output = false
-open_weights = true
-
-[cost]
-input = 0.08
-output = 0.08
-
-[limit]
-context = 8_192
-output = 16_384
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,22 +0,0 @@
-name = "Gemma 3 1B IT"
-family = "gemma"
-release_date = "2025-03-12"
-last_updated = "2025-03-12"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = false
-structured_output = false
-open_weights = true
-
-[cost]
-input = 0.08
-output = 0.30
-
-[limit]
-context = 1_000_000
-output = 16_384
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,22 +0,0 @@
-name = "Gemma 3 27B"
-family = "gemma"
-release_date = "2025-03-12"
-last_updated = "2025-03-12"
-attachment = true
-reasoning = false
-temperature = true
-tool_call = false
-structured_output = false
-open_weights = true
-
-[cost]
-input = 0.27
-output = 0.27
-
-[limit]
-context = 128_000
-output = 16_384
-
-[modalities]
-input = ["text", "image"]
-output = ["text"]
@@ -5,3 +5,6 @@ input = 0.2
 output = 1.1
 cache_read = 0.03
 cache_write = 0
+
+[limit]
+context = 131_000
@@ -5,3 +5,6 @@ input = 0
 output = 0
 cache_read = 0
 cache_write = 0
+
+[limit]
+context = 128_000
@@ -5,3 +5,6 @@ input = 0.6
 output = 2.2
 cache_read = 0.11
 cache_write = 0
+
+[limit]
+context = 131_000
@@ -3,3 +3,7 @@ base_model = "zhipuai/glm-4.5v"
 [cost]
 input = 0.6
 output = 1.8
+cache_read = 0.11
+
+[limit]
+context = 128_000
@@ -3,3 +3,7 @@ base_model = "zhipuai/glm-4.6v"
 [cost]
 input = 0.3
 output = 0.9
+cache_read = 0.05
+
+[limit]
+context = 131_072
@@ -0,0 +1,22 @@
+name = "GLM-4.7 Flash (Free)"
+family = "glm"
+release_date = "2025-12-22"
+last_updated = "2025-12-22"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = false
+open_weights = false
+
+[cost]
+input = 0
+output = 0
+
+[limit]
+context = 200_000
+output = 200_000
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -1,7 +1,7 @@
 base_model = "zhipuai/glm-4.7-flash"

 [cost]
-input = 0
-output = 0
-cache_read = 0
+input = 0.06
+output = 0.4
+cache_read = 0.01
 cache_write = 0
@@ -4,7 +4,10 @@ base_model = "zhipuai/glm-5.1"
 field = "reasoning_content"

 [cost]
-input = 6
-output = 24
-cache_read = 1.3
+input = 1.4
+output = 4.4
+cache_read = 0.26
 cache_write = 0
+
+[limit]
+context = 204_800
@@ -8,3 +8,6 @@ input = 1
 output = 3.2
 cache_read = 0.2
 cache_write = 0
+
+[limit]
+context = 203_000
@@ -4,3 +4,6 @@ base_model = "openai/gpt-4.1-mini"
 input = 0.4
 output = 1.6
 cache_read = 0.1
+
+[limit]
+context = 1_000_000
@@ -4,3 +4,6 @@ base_model = "openai/gpt-4.1-nano"
 input = 0.1
 output = 0.4
 cache_read = 0.025
+
+[limit]
+context = 1_000_000
@@ -4,3 +4,6 @@ base_model = "openai/gpt-4.1"
 input = 2
 output = 8
 cache_read = 0.5
+
+[limit]
+context = 1_000_000
@@ -10,7 +10,3 @@ tier = { type = "context", size = 272_000 }
 input = 10
 output = 45
 cache_read = 1
-
-[experimental.modes.fast]
-cost = { input = 12.5, output = 75, cache_read = 1.25 }
-provider = { body = { service_tier = "priority" } }
@@ -10,8 +10,8 @@ structured_output = true
 open_weights = false

 [cost]
-input = 0.05
-output = 0.25
+input = 0.15
+output = 0.75

 [limit]
 context = 131_072
@@ -10,8 +10,8 @@ structured_output = true
 open_weights = false

 [cost]
-input = 0.04
-output = 0.15
+input = 0.1
+output = 0.5

 [limit]
 context = 131_072
@@ -0,0 +1,23 @@
+name = "Grok 4.1 Fast Non-Reasoning"
+family = "grok"
+release_date = "2025-11-19"
+last_updated = "2025-11-19"
+attachment = true
+reasoning = false
+temperature = true
+tool_call = true
+structured_output = false
+open_weights = false
+
+[cost]
+input = 0.2
+output = 0.5
+cache_read = 0.05
+
+[limit]
+context = 2_000_000
+output = 2_000_000
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
@@ -1,8 +1,8 @@
 base_model = "xai/grok-4.20-0309-non-reasoning"

 [cost]
-input = 1.25
-output = 2.5
+input = 2
+output = 6
 cache_read = 0.2

 [[cost.tiers]]
@@ -10,3 +10,6 @@ tier = { type = "context", size = 200_000 }
 input = 2.5
 output = 5
 cache_read = 0.4
+
+[limit]
+context = 2_000_000
@@ -1,9 +1,12 @@
 base_model = "xai/grok-4.20-0309-reasoning"
-reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }]
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high"]

 [cost]
-input = 1.25
-output = 2.5
+input = 2
+output = 6
 cache_read = 0.2

 [[cost.tiers]]
@@ -11,3 +14,6 @@ tier = { type = "context", size = 200_000 }
 input = 2.5
 output = 5
 cache_read = 0.4
+
+[limit]
+context = 2_000_000
@@ -1,8 +1,8 @@
 base_model = "xai/grok-4.20-0309-non-reasoning"

 [cost]
-input = 1.25
-output = 2.5
+input = 2
+output = 6
 cache_read = 0.2

 [[cost.tiers]]
@@ -10,3 +10,6 @@ tier = { type = "context", size = 200_000 }
 input = 2.5
 output = 5
 cache_read = 0.4
+
+[limit]
+context = 2_000_000
@@ -1,9 +1,12 @@
 base_model = "xai/grok-4.20-0309-reasoning"
-reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }]
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high"]

 [cost]
-input = 1.25
-output = 2.5
+input = 2
+output = 6
 cache_read = 0.2

 [[cost.tiers]]
@@ -11,3 +14,6 @@ tier = { type = "context", size = 200_000 }
 input = 2.5
 output = 5
 cache_read = 0.4
+
+[limit]
+context = 2_000_000
@@ -1,10 +1,13 @@
 base_model = "xai/grok-4.3"
-reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }]
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high"]

 [cost]
 input = 1.25
 output = 2.5
-cache_read = 0.2
+cache_read = 0.3125

 [[cost.tiers]]
 tier = { type = "context", size = 200_000 }
@@ -1,24 +0,0 @@
-name = "Grok 4 Fast Reasoning"
-family = "grok"
-release_date = "2025-07-09"
-last_updated = "2025-07-09"
-attachment = true
-reasoning = true
-reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }]
-temperature = true
-tool_call = true
-structured_output = true
-open_weights = false
-
-[cost]
-input = 0.20
-output = 0.50
-cache_read = 0.05
-
-[limit]
-context = 2_000_000
-output = 30_000
-
-[modalities]
-input = ["text", "image"]
-output = ["text"]
@@ -1,22 +1,23 @@
-name = "Grok 4 (0709)"
+name = "Grok 4"
 family = "grok"
 release_date = "2025-07-09"
 last_updated = "2025-07-09"
-attachment = false
+attachment = true
 reasoning = false
 temperature = true
 tool_call = true
-structured_output = true
+structured_output = false
 open_weights = false

 [cost]
-input = 3.00
-output = 15.00
+input = 3
+output = 15
+cache_read = 0.75

 [limit]
 context = 256_000
 output = 256_000

 [modalities]
-input = ["text"]
-output = ["text"]
+input = ["text", "image"]
+output = ["text"]
@@ -1,22 +0,0 @@
-name = "Hermes 2 Pro Llama 3 8B"
-family = "hermes"
-release_date = "2024-05-27"
-last_updated = "2024-05-27"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = false
-structured_output = false
-open_weights = true
-
-[cost]
-input = 0.14
-output = 0.14
-
-[limit]
-context = 8_192
-output = 8_192
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -10,12 +10,12 @@ structured_output = true
 open_weights = true

 [cost]
-input = 0.60
-output = 2.50
-cache_read = 0.12
+input = 1
+output = 3
+cache_read = 0.5

 [limit]
-context = 131_072
+context = 256_000
 output = 16_384

 [modalities]
@@ -1,23 +0,0 @@
-name = "Llama 3.1 8B Instruct"
-family = "llama"
-release_date = "2024-07-23"
-last_updated = "2024-07-23"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = true
-structured_output = true
-open_weights = true
-status = "beta"
-
-[cost]
-input = 0.22
-output = 0.22
-
-[limit]
-context = 128_000
-output = 2_048
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,23 +1,8 @@
-name = "Llama 3.1 Nemotron Ultra 253B"
 base_model = "nvidia/llama-3.1-nemotron-ultra-253b"
-family = "llama"
-release_date = "2025-04-07"
-last_updated = "2025-04-07"
-attachment = false
 reasoning = false
-temperature = true
 tool_call = false
 structured_output = true
-open_weights = true

 [cost]
-input = 0.60
-output = 1.80
-
-[limit]
-context = 128_000
-output = 8_192
-
-[modalities]
-input = ["text"]
-output = ["text"]
+input = 0.6
+output = 1.8
@@ -1,5 +1,8 @@
 base_model = "meta/llama-3.3-70b-instruct"

 [cost]
-input = 0
-output = 0
+input = 0.13
+output = 0.4
+
+[limit]
+context = 131_072
@@ -14,7 +14,7 @@ input = 0.24
 output = 0.97

 [limit]
-context = 8_192
+context = 1_048_576
 output = 2_048

 [modalities]
@@ -14,7 +14,7 @@ input = 0.17
 output = 0.66

 [limit]
-context = 8_192
+context = 131_072
 output = 2_048

 [modalities]
@@ -1,23 +0,0 @@
-name = "Llama 4 Scout"
-family = "llama"
-release_date = "2025-04-05"
-last_updated = "2025-04-05"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = true
-structured_output = false
-open_weights = true
-status = "beta"
-
-[cost]
-input = 0.18
-output = 0.59
-
-[limit]
-context = 32_768
-output = 16_384
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,10 +0,0 @@
-base_model = "xiaomi/mimo-v2-flash"
-reasoning_options = [{ type = "toggle" }]
-
-[interleaved]
-field = "reasoning_content"
-
-[cost]
-input = 0.1
-output = 0.3
-cache_read = 0.01
@@ -8,3 +8,6 @@ field = "reasoning_content"
 input = 0.4
 output = 2
 cache_read = 0.08
+
+[limit]
+context = 256_000
@@ -1,5 +1,7 @@
 base_model = "xiaomi/mimo-v2-pro"
-reasoning_options = [{ type = "toggle" }]
+
+[[reasoning_options]]
+type = "toggle"

 [interleaved]
 field = "reasoning_content"
@@ -14,3 +16,6 @@ tier = { type = "context", size = 256_000 }
 input = 2
 output = 6
 cache_read = 0.4
+
+[limit]
+context = 1_000_000
@@ -1,16 +1,21 @@
 base_model = "xiaomi/mimo-v2.5-pro"
-reasoning_options = [{ type = "toggle" }]
+
+[[reasoning_options]]
+type = "toggle"

 [interleaved]
 field = "reasoning_content"

 [cost]
-input = 1
-output = 3
-cache_read = 0.2
+input = 0.435
+output = 0.87
+cache_read = 0.0036

 [[cost.tiers]]
 tier = { type = "context", size = 256_000 }
 input = 2
 output = 6
 cache_read = 0.4
+
+[limit]
+context = 1_000_000
@@ -1,16 +1,21 @@
 base_model = "xiaomi/mimo-v2.5"
-reasoning_options = [{ type = "toggle" }]
+
+[[reasoning_options]]
+type = "toggle"

 [interleaved]
 field = "reasoning_content"

 [cost]
-input = 0.4
-output = 2
-cache_read = 0.08
+input = 0.14
+output = 0.28
+cache_read = 0.0028

 [[cost.tiers]]
 tier = { type = "context", size = 256_000 }
 input = 0.8
 output = 4
 cache_read = 0.16
+
+[limit]
+context = 1_000_000
@@ -1,5 +1,5 @@
 base_model = "minimax/MiniMax-M2.1"

 [cost]
-input = 0.3
-output = 1.2
+input = 0.27
+output = 1.1
@@ -3,5 +3,5 @@ base_model = "minimax/MiniMax-M2.5-highspeed"
 [cost]
 input = 0.6
 output = 2.4
-cache_read = 0.06
+cache_read = 0.03
 cache_write = 0.375
@@ -5,3 +5,6 @@ input = 0.3
 output = 1.2
 cache_read = 0.03
 cache_write = 0.375
+
+[limit]
+context = 228_700
@@ -1,5 +1,6 @@
 base_model = "minimax/MiniMax-M2"

 [cost]
-input = 0.3
-output = 1.2
+input = 0.2
+output = 1
+cache_read = 0.03
@@ -1,5 +1,8 @@
 base_model = "mistral/mistral-large-latest"

 [cost]
-input = 0.5
-output = 1.5
+input = 4
+output = 12
+
+[limit]
+context = 128_000
@@ -1,7 +1,12 @@
 base_model = "nvidia/nemotron-3-ultra-550b-a55b"
-reasoning_options = [{ type = "toggle" }]
+
+[[reasoning_options]]
+type = "toggle"

 [cost]
 input = 0.5
 output = 2.5
 cache_read = 0.15
+
+[limit]
+context = 262_144
@@ -1,5 +1,5 @@
 base_model = "mistral/pixtral-large-latest"

 [cost]
-input = 2
-output = 6
+input = 4
+output = 12
@@ -3,3 +3,5 @@ base_model = "alibaba/qwen-flash"
 [cost]
 input = 0.05
 output = 0.4
+cache_read = 0.01
+cache_write = 0.0625
@@ -10,8 +10,8 @@ structured_output = true
 open_weights = false

 [cost]
-input = 0.345
-output = 1.377
+input = 1.6
+output = 6.4

 [limit]
 context = 32_768
@@ -1,7 +1,5 @@
 base_model = "alibaba/qwen-omni-turbo"

 [cost]
-input = 0.07
-output = 0.27
-input_audio = 4.44
-output_audio = 8.89
+input = 0.2
+output = 0.8
@@ -10,11 +10,13 @@ structured_output = true
 open_weights = false

 [cost]
-input = 0.115
-output = 0.287
+input = 0.4
+output = 1.2
+cache_read = 0.08
+cache_write = 0.5

 [limit]
-context = 131_072
+context = 1_000_000
 output = 8_192

 [modalities]
@@ -4,3 +4,8 @@ base_model = "alibaba/qwen-plus"
 input = 0.4
 output = 1.2
 reasoning = 4
+cache_read = 0.08
+cache_write = 0.5
+
+[limit]
+context = 131_072
@@ -2,4 +2,4 @@ base_model = "alibaba/qwen-vl-plus"

 [cost]
 input = 0.21
-output = 0.63
+output = 0.64
@@ -1,5 +1,8 @@
 base_model = "alibaba/qwen2-5-vl-72b-instruct"

 [cost]
-input = 2.8
-output = 8.4
+input = 0.13
+output = 0.4
+
+[limit]
+context = 32_768
@@ -1,22 +0,0 @@
-name = "Qwen2.5 Coder 7B"
-family = "qwen"
-release_date = "2024-09-19"
-last_updated = "2024-09-19"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = false
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.05
-output = 0.05
-
-[limit]
-context = 131_072
-output = 8_192
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -10,11 +10,11 @@ structured_output = true
 open_weights = true

 [cost]
-input = 0.20
-output = 0.80
+input = 0.2
+output = 0.8

 [limit]
-context = 131_072
+context = 40_960
 output = 8_192

 [modalities]
@@ -10,11 +10,11 @@ structured_output = true
 open_weights = true

 [cost]
-input = 0.09
-output = 0.58
+input = 0.2
+output = 0.6

 [limit]
-context = 131_072
+context = 262_000
 output = 8_192

 [modalities]
@@ -10,11 +10,11 @@ structured_output = true
 open_weights = true

 [cost]
-input = 0.20
-output = 0.60
+input = 0.2
+output = 0.6

 [limit]
-context = 131_072
+context = 262_000
 output = 8_192

 [modalities]
@@ -1,22 +0,0 @@
-name = "Qwen3 30B A3B FP8"
-family = "qwen"
-release_date = "2025-04-28"
-last_updated = "2025-04-28"
-attachment = false
-reasoning = true
-temperature = true
-tool_call = true
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.10
-output = 0.10
-
-[limit]
-context = 131_072
-output = 8_192
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -10,11 +10,11 @@ structured_output = true
 open_weights = true

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

 [limit]
-context = 131_072
+context = 262_000
 output = 8_192

 [modalities]
@@ -1,22 +0,0 @@
-name = "Qwen3 30B A3B Thinking (2507)"
-family = "qwen"
-release_date = "2025-07-08"
-last_updated = "2025-07-08"
-attachment = false
-reasoning = true
-temperature = true
-tool_call = true
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.10
-output = 0.10
-
-[limit]
-context = 131_072
-output = 8_192
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,22 +0,0 @@
-name = "Qwen3 32B FP8"
-family = "qwen"
-release_date = "2025-04-28"
-last_updated = "2025-04-28"
-attachment = false
-reasoning = true
-temperature = true
-tool_call = true
-structured_output = true
-open_weights = true
-
-[cost]
-input = 0.10
-output = 0.10
-
-[limit]
-context = 131_072
-output = 8_192
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,6 +1,9 @@
 base_model = "alibaba/qwen3-32b"

 [cost]
-input = 0.7
-output = 2.8
+input = 0.1
+output = 0.3
 reasoning = 8.4
+
+[limit]
+context = 32_768
@@ -14,7 +14,7 @@ input = 0.03
 output = 0.03

 [limit]
-context = 131_072
+context = 128_000
 output = 8_192

 [modalities]
@@ -1,5 +1,8 @@
 base_model = "alibaba/qwen3-coder-30b-a3b-instruct"

 [cost]
-input = 0.45
-output = 2.25
+input = 0.1
+output = 0.3
+
+[limit]
+context = 262_000
@@ -1,5 +1,5 @@
 base_model = "alibaba/qwen3-coder-480b-a35b-instruct"

 [cost]
-input = 1.5
-output = 7.5
+input = 0.4
+output = 1.8
@@ -3,3 +3,5 @@ base_model = "alibaba/qwen3-coder-flash"
 [cost]
 input = 0.3
 output = 1.5
+cache_read = 0.06
+cache_write = 0.375
@@ -12,6 +12,7 @@ open_weights = false
 [cost]
 input = 0.108
 output = 0.675
+cache_read = 0.06

 [limit]
 context = 262_144
@@ -1,5 +1,10 @@
 base_model = "alibaba/qwen3-coder-plus"

 [cost]
-input = 1
-output = 5
+input = 6
+output = 60
+cache_read = 1.2
+cache_write = 7.5
+
+[limit]
+context = 1_000_000
@@ -10,12 +10,13 @@ structured_output = true
 open_weights = false

 [cost]
-input = 0.359
-output = 1.434
-cache_read = 0.072
+input = 1.2
+output = 6
+cache_read = 0.24
+cache_write = 1.5

 [limit]
-context = 256_000
+context = 262_144
 output = 32_800

 [modalities]
@@ -1,5 +1,7 @@
 base_model = "alibaba/qwen3-max"

 [cost]
-input = 1.2
-output = 6
+input = 3
+output = 15
+cache_read = 0.6
+cache_write = 3.75
@@ -10,8 +10,8 @@ structured_output = true
 open_weights = true

 [cost]
-input = 0.30
-output = 1.50
+input = 0.5
+output = 2

 [limit]
 context = 131_072
@@ -10,12 +10,12 @@ structured_output = true
 open_weights = false

 [cost]
-input = 0.022
-output = 0.215
-cache_read = 0.0044
+input = 0.05
+output = 0.4
+cache_read = 0.01

 [limit]
-context = 1_000_000
+context = 262_144
 output = 32_000

 [modalities]
@@ -4,3 +4,5 @@ base_model = "alibaba/qwen3-vl-plus"
 input = 0.2
 output = 1.6
 reasoning = 4.8
+cache_read = 0.04
+cache_write = 0.25
@@ -12,3 +12,6 @@ input = 2
 output = 6
 cache_read = 0.2
 cache_write = 2.5
+
+[limit]
+context = 262_144
--- a/Show More
+++ b/Show More