chore(sync): update DigitalOcean model catalog

feat(synthetic): Add Kimi K3 model (#3794 )
* Add Synthetic's Kimi K3 offering * Update cache read price * fix(synthetic): declare effort-only reasoning for Kimi K3 per Synthetic API docs Synthetic's OpenAI-compatible chat completions API documents reasoning_effort with values low | medium | high and no reasoning on/off toggle, so drop the toggle option and align effort values with the documented surface (matching the existing Synthetic Kimi K2.6 / K2.7-Code entries). https://dev.synthetic.new/docs/openai/chat-completions Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * Remove unnecessary comment * Retrigger transient actions failure --------- Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
2026-07-28 10:09:45 +00:00 · 2026-07-27 23:36:15 -05:00 · 2026-07-27 23:36:01 -05:00 · 2026-07-27 23:35:53 -05:00 · 2026-07-27 23:35:42 -05:00 · 2026-07-27 23:27:01 -05:00
259 changed files with 1940 additions and 1459 deletions
@@ -38,6 +38,8 @@ For model catalog changes, enforce these review rules:
 - Treat a missing compliant logo for a new provider as a merge blocker. The SVG must use `currentColor`, have no fixed size or hardcoded color, and preferably use a square `viewBox`.
 - Treat duplicated provider-agnostic metadata as a merge blocker when a matching `models/<provider>/<model>.toml` exists; the provider entry must use `base_model` and retain only provider-specific fields and overrides.
 - Treat missing `reasoning_options` on `reasoning = true` provider models as a merge blocker. Options describe controls exposed by that inference provider, not merely by the upstream model. An empty array is correct when reasoning exists but no caller control is verified.
+- Before reporting a `reasoning_options` problem, compare the proposed model with existing entries for the same underlying model that use a comparable request surface. Determine that surface from the effective `npm`, provider API shape, and any model-level provider override—not from the model family alone. Prefer native-provider examples when the target uses the native SDK (for example, an Anthropic model through `@ai-sdk/anthropic` should be compared with the Anthropic provider). Prefer established OpenAI-compatible gateway examples when the target uses an OpenAI-compatible chat-completions surface (for example, Cloudflare AI Gateway may be usefully compared with OpenRouter). Do not compare a native Anthropic route with an OpenAI-compatible gateway as though their controls were interchangeable.
+- Use those peer entries as required review context, not as values to copy mechanically or as standalone proof. Consistent same-model, same-surface examples make a proposed option more plausible and help identify likely omissions or contradictions; target-provider documentation, endpoint metadata, adapter behavior, or reproduced requests still override peer precedent. A lack of bespoke provider documentation is not by itself an action item when the target surface and strong peer examples support the proposal and the diff contains no concrete contradictory evidence. Conversely, do not accept or reject `reasoning_options = []` mechanically: explain the specific mismatch with the target API shape or comparable providers before requesting a change.
 - Do not treat absence of a sync module as a blocker. Recommend one only when a context-rich provider API can authoritatively populate model data or delete models no longer served.
 - Data-changing PRs should cite direct provider pricing, model documentation, or API references in the PR body. Missing citations are not by themselves a merge blocker, but should be reported as a low-severity request for evidence when material factual changes otherwise cannot be reviewed. Prefer first-party sources and require each citation to state what it supports.
 - You cannot fetch citation URLs. Assess whether citations are present, direct, and mapped to claims, but never claim you opened a URL or verified its contents. A URL or PR assertion alone does not prove a disputed value.
@@ -15,6 +15,7 @@ import { deepinfra } from "./providers/deepinfra.js";
 import { digitalocean } from "./providers/digitalocean.js";
 import { empiriolabs } from "./providers/empiriolabs.js";
 import { google } from "./providers/google.js";
+import { hyper } from "./providers/hyper.js";
 import { huggingface } from "./providers/huggingface.js";
 import { kilo } from "./providers/kilo.js";
 import { llmgateway } from "./providers/llmgateway.js";
@@ -114,8 +115,9 @@ export const providers: {
  digitalocean: SyncProvider<any>;
  empiriolabs: SyncProvider<any>;
  google: SyncProvider<any>;
-  kilo: SyncProvider<any>;
+  hyper: SyncProvider<any>;
  huggingface: SyncProvider<any>;
+  kilo: SyncProvider<any>;
  llmgateway: SyncProvider<any>;
  openai: SyncProvider<any>;
  openrouter: SyncProvider<any>;
@@ -136,8 +138,9 @@ export const providers: {
  digitalocean,
  empiriolabs,
  google,
-  kilo,
+  hyper,
  huggingface,
+  kilo,
  llmgateway,
  openai,
  openrouter,
@@ -152,7 +155,7 @@ export const providers: {
 export const groups = {
  aggregators: ["crossmodel", "empiriolabs", "huggingface", "kilo", "llmgateway", "openrouter", "vercel"],
  cloudflare: ["cloudflare-workers-ai"],
-  direct: ["ambient", "anthropic", "baseten", "chutes", "deepinfra", "digitalocean", "google", "openai", "ovhcloud", "pioneer", "venice", "wandb", "xai"],
+  direct: ["ambient", "anthropic", "baseten", "chutes", "deepinfra", "digitalocean", "google", "hyper", "openai", "ovhcloud", "pioneer", "venice", "wandb", "xai"],
 } as const;

 type ProviderID = keyof typeof providers;
@@ -0,0 +1,156 @@
+import { existsSync } from "node:fs";
+import path from "node:path";
+import { z } from "zod";
+
+import type { ExistingModel, SyncProvider, SyncedFullModel, SyncedModel } from "../index.js";
+import { factorBaseModel } from "./openrouter.js";
+
+const API_ENDPOINT = "https://hyper.charm.land/v1/models";
+const MODELS_DIR = path.join(import.meta.dirname, "..", "..", "..", "..", "..", "models");
+
+function baseModelExists(modelID: string) {
+  return existsSync(path.join(MODELS_DIR, `${modelID}.toml`));
+}
+
+const ReasoningEffort = z.enum([
+  "default",
+  "max",
+  "low",
+  "high",
+  "none",
+  "medium",
+  "minimal",
+  "xhigh",
+]);
+
+export const HyperModel = z.object({
+  id: z.string(),
+  created: z.number(),
+  display_name: z.string(),
+  context_window: z.number(),
+  max_output_tokens: z.number(),
+  capabilities: z.object({
+    vision: z.boolean().optional(),
+  }).optional(),
+  reasoning: z.object({
+    effort_levels: z.array(z.object({
+      value: z.string(),
+      display: z.string().optional(),
+    })).optional(),
+  }).optional(),
+  pricing: z.object({
+    input: z.number().optional(),
+    output: z.number().optional(),
+    cache_hit: z.number().optional(),
+    cache_create: z.number().optional(),
+  }).optional(),
+}).passthrough();
+
+const HyperResponse = z.object({
+  data: z.array(HyperModel),
+}).passthrough();
+
+export type HyperModel = z.infer<typeof HyperModel>;
+
+export const hyper = {
+  id: "hyper",
+  name: "Charm Hyper",
+  modelsDir: "providers/hyper/models",
+  preserveBaseModels: false,
+  async fetchModels() {
+    const key = process.env.HYPER_API_KEY;
+    const response = await fetch(API_ENDPOINT, key
+      ? { headers: { Authorization: `Bearer ${key}` } }
+      : undefined);
+    if (!response.ok) {
+      throw new Error(`Hyper models request failed: ${response.status} ${response.statusText}`);
+    }
+    return response.json();
+  },
+  parseModels(raw) {
+    return HyperResponse.parse(raw).data;
+  },
+  translateModel(model, context) {
+    const existing = context.existing(model.id);
+    const baseModel = existing?.base_model;
+    if (baseModel === undefined || !baseModelExists(baseModel)) return undefined;
+    return {
+      id: model.id,
+      model: buildHyperModel(model, existing, baseModel),
+    };
+  },
+} satisfies SyncProvider<HyperModel>;
+
+function dateFromTimestamp(timestamp: number) {
+  return new Date(timestamp * 1000).toISOString().slice(0, 10);
+}
+
+function reasoningOptions(model: HyperModel) {
+  const effortLevels = model.reasoning?.effort_levels?.map((level) => level.value) ?? [];
+  if (effortLevels.length === 0) return [];
+  const values = effortLevels.filter(isReasoningEffort);
+  if (values.length === 0) return [{ type: "toggle" as const }];
+  return [{ type: "effort" as const, values }];
+}
+
+function isReasoningEffort(value: string): value is z.infer<typeof ReasoningEffort> {
+  return ReasoningEffort.safeParse(value).success;
+}
+
+function price(value: number) {
+  return Math.round(value * 1_000_000) / 1_000_000;
+}
+
+function positivePrice(value: number | undefined) {
+  return value !== undefined && value > 0 ? price(value) : undefined;
+}
+
+function buildCost(model: HyperModel, existing: ExistingModel["cost"] | undefined) {
+  const pricing = model.pricing;
+  if (pricing?.input === undefined || pricing.output === undefined) return existing;
+
+  return {
+    input: price(pricing.input),
+    output: price(pricing.output),
+    cache_read: positivePrice(pricing.cache_hit)
+      ?? (pricing.cache_hit === undefined ? existing?.cache_read : undefined),
+    cache_write: positivePrice(pricing.cache_create)
+      ?? (pricing.cache_create === undefined ? existing?.cache_write : undefined),
+    reasoning: existing?.reasoning,
+  };
+}
+
+function hyperModalities(vision: boolean) {
+  const input = vision ? ["text" as const, "image" as const] : ["text" as const];
+  return {
+    input,
+    output: ["text" as const],
+  };
+}
+
+export function buildHyperModel(
+  model: HyperModel,
+  existing: ExistingModel | undefined,
+  baseModel: string,
+  today = new Date().toISOString().slice(0, 10),
+): SyncedModel {
+  const limit = {
+    context: model.context_window,
+    input: existing?.limit?.input,
+    output: model.max_output_tokens,
+  };
+  const modalities = hyperModalities(model.capabilities?.vision ?? false);
+  const values: Partial<SyncedFullModel> = {
+    attachment: modalities.input.some((value) => value !== "text"),
+    modalities,
+    reasoning: model.reasoning != null,
+    reasoning_options: model.reasoning != null ? reasoningOptions(model) : undefined,
+    release_date: existing?.release_date ?? dateFromTimestamp(model.created),
+    last_updated: existing?.last_updated ?? today,
+    interleaved: existing?.interleaved,
+    cost: buildCost(model, existing?.cost),
+    limit,
+  };
+
+  return factorBaseModel(baseModel, values, limit, existing?.base_model_omit);
+}
@@ -19,6 +19,7 @@ import {
  resolveDigitalOceanBaseModel,
  type DigitalOceanSourceModel,
 } from "../src/sync/providers/digitalocean.js";
+import { buildHyperModel, type HyperModel } from "../src/sync/providers/hyper.js";
 import {
  buildEmpiriolabsModel,
  empiriolabs,
@@ -791,6 +792,66 @@ function deepInfraModel(model_name: string, tags: string[]): DeepInfraModel {
  };
 }

+test("syncs Hyper pricing from catalog input/output fields", () => {
+  const model = hyperModel({
+    id: "minimax-m2.7",
+    reasoning: undefined,
+    pricing: {
+      input: 0.3,
+      output: 1.2,
+      cache_hit: 0.06,
+      cache_create: 0.03,
+    },
+  });
+
+  expect(buildHyperModel(model, undefined, "minimax/MiniMax-M2.7")).toMatchObject({
+    cost: { input: 0.3, output: 1.2, cache_read: 0.06, cache_write: 0.03 },
+    reasoning: false,
+  });
+  expect(buildHyperModel(model, undefined, "minimax/MiniMax-M2.7")).not.toHaveProperty("reasoning_options");
+});
+
+test("rounds Hyper pricing to six decimal places", () => {
+  const model = hyperModel({
+    id: "deepseek-v4-flash",
+    pricing: {
+      input: 0.20000010875000002,
+      output: 0.40000021750000003,
+      cache_hit: 0.039999586250000004,
+    },
+  });
+
+  expect(buildHyperModel(model, undefined, "deepseek/deepseek-v4-flash")).toMatchObject({
+    cost: { input: 0.2, output: 0.4, cache_read: 0.04 },
+  });
+});
+
+test("sets Hyper reasoning false when API omits reasoning metadata", () => {
+  const model = hyperModel({ id: "llama-3.3-70b-instruct", reasoning: undefined });
+
+  expect(buildHyperModel(model, undefined, "meta/llama-3.3-70b-instruct")).toMatchObject({
+    attachment: false,
+  });
+  expect(buildHyperModel(model, undefined, "meta/llama-3.3-70b-instruct")).not.toHaveProperty("reasoning");
+  expect(buildHyperModel(model, undefined, "meta/llama-3.3-70b-instruct")).not.toHaveProperty("reasoning_options");
+
+  expect(buildHyperModel(hyperModel({ id: "minimax-m2.7", reasoning: undefined }), undefined, "minimax/MiniMax-M2.7")).toMatchObject({
+    reasoning: false,
+  });
+});
+
+test("preserves existing Hyper cost when API pricing is missing", () => {
+  const existing = {
+    cost: { input: 1, output: 2 },
+    release_date: "2026-01-01",
+    last_updated: "2026-01-01",
+  };
+
+  expect(buildHyperModel(hyperModel({ id: "minimax-m2.7" }), existing, "minimax/MiniMax-M2.7")).toMatchObject({
+    cost: { input: 1, output: 2 },
+  });
+});
+
 test("formats interleaved as a root field before reasoning option tables", () => {
  const content = formatToml({
    id: "example/model",
@@ -1509,6 +1570,23 @@ function llmGatewayModel(overrides: Partial<LLMGatewayModel> = {}): LLMGatewayMo
  };
 }

+function hyperModel(overrides: Partial<HyperModel> = {}): HyperModel {
+  return {
+    id: "deepseek-v4-flash",
+    created: 1_780_592_628,
+    display_name: "DeepSeek V4 Flash",
+    reasoning: {
+      effort_levels: [
+        { value: "high" },
+        { value: "xhigh" },
+      ],
+    },
+    context_window: 1_000_000,
+    max_output_tokens: 384_000,
+    ...overrides,
+  };
+}
+
 function openRouterModel(overrides: Partial<OpenRouterModel> = {}): OpenRouterModel {
  return {
    id: "anthropic/claude-sonnet-5",
@@ -0,0 +1,42 @@
+# Sources (accessed 2026-07-28):
+# - Model card, limits (1M context, 999,990 max output), capabilities, GLM-5.2 base:
+#   https://docs.abliteration.ai/models
+# - Reasoning effort ladder and per-endpoint request fields:
+#   https://docs.abliteration.ai/capabilities/thinking
+# - Pricing ($5 per 1M tokens, flat input + output):
+#   https://docs.abliteration.ai/pricing
+#   https://abliteration.ai/pricing#api-pricing
+# - Launch announcement (base model, fine-tuning, benchmarks):
+#   https://abliteration.ai/blog/introducing-abliterated-model-large
+name = "Abliterated Model Large"
+description = "GLM-5.2 model abliterated and finetuned for cyber, ML red teaming, and agent testing"
+release_date = "2026-07-25"
+last_updated = "2026-07-28"
+attachment = false
+reasoning = true
+tool_call = true
+structured_output = true
+temperature = true
+open_weights = false
+
+# Two distinct reasoning depths: high and max. The API also accepts the other
+# ladder values as aliases, mapping minimal-high -> high and xhigh-max -> max.
+[[reasoning_options]]
+type = "effort" # API: {"reasoning_effort": "<value>"} on /v1/chat/completions
+values = ["none", "high", "max"]
+
+[[reasoning_options]]
+type = "toggle" # API: {"thinking": false} on /v1/messages (legacy alias on /v1/chat/completions)
+
+[cost]
+input = 5.00
+output = 5.00
+
+[limit]
+context = 1_000_000
+input = 1_000_000
+output = 999_990
+
+[modalities]
+input = ["text"]
+output = ["text"]
@@ -1,20 +1,27 @@
+# Sources (accessed 2026-07-28):
+# - Model card and capabilities: https://docs.abliteration.ai/models
+# - Reasoning effort ladder and per-endpoint request fields (reasoning on by
+#   default; "none" disables; top-level thinking:false disables on
+#   /v1/messages and as a legacy alias on /v1/chat/completions):
+#   https://docs.abliteration.ai/capabilities/thinking
 name = "Abliterated Model"
 description = "Multimodal model for analyzing text, images, documents, and rich media"
-# Reasoning HTTP format (accessed 2026-06-25):
-# This model thinks by default. On POST /v1/chat/completions or /v1/messages,
-# top-level `thinking: false` skips thinking; omission keeps it enabled.
-# Sources:
-# https://docs.abliteration.ai/models
-# https://docs.abliteration.ai/capabilities/thinking
 release_date = "2026-01-06"
-last_updated = "2026-01-06"
+last_updated = "2026-07-28"
 attachment = true
-reasoning = false
+reasoning = true
 tool_call = true
 structured_output = false
 temperature = true
 open_weights = true

+[[reasoning_options]]
+type = "effort" # API: {"reasoning_effort": "<value>"} on /v1/chat/completions
+values = ["none", "minimal", "low", "medium", "high", "xhigh", "max"]
+
+[[reasoning_options]]
+type = "toggle" # API: {"thinking": false} on /v1/messages (legacy alias on /v1/chat/completions)
+
 [cost]
 input = 3.00
 output = 3.00
@@ -1,12 +1,16 @@
+# Reasoning controls for the whole API surface (verified 2026-07-28):
+# https://docs.abliteration.ai/capabilities/thinking
+# Both models reason by default. Effort ladder, least to most:
+# none (disables reasoning), minimal, low, medium, high, xhigh, max.
+# Request field per endpoint:
+# - POST /v1/chat/completions: top-level "reasoning_effort": "<level>"
+#   (legacy top-level "thinking": false also disables reasoning)
+# - POST /v1/responses: "reasoning": { "effort": "<level>" }
+# - POST /v1/messages: "output_config": { "effort": "<level>" } or
+#   "thinking": { "type": "enabled", "budget_tokens": <n> };
+#   "thinking": false disables reasoning
 name = "abliteration.ai"
 env = ["ABLIT_KEY"]
 npm = "@ai-sdk/openai-compatible"
-# Reasoning HTTP format (accessed 2026-06-25):
-# POST /v1/chat/completions and POST /v1/messages: top-level `thinking` is true
-# by default; false skips thinking. POST /v1/responses has no thinking toggle.
-# No effort or numeric reasoning-budget request field is documented.
-# Sources:
-# https://docs.abliteration.ai/capabilities/thinking
-# https://docs.abliteration.ai/compatibility-matrix
 api = "https://api.abliteration.ai/v1"
 doc = "https://docs.abliteration.ai/models"
@@ -0,0 +1,24 @@
+# Source: GET https://api.aiand.com/v1/models (USD list prices; accessed
+# 2026-07-28). Pricing is aiand-specific: $3.00 input, $0.50 cache_read,
+# $12.50 output (differs from Moonshot official rates).
+# Modalities: text + image + pdf accepted per GET /v1/models; video rejected
+# ("does not support video input") — overridden from shared base
+# text+image+video. PDF kept: sibling aiand Moonshot entries (kimi-k2.6,
+# kimi-k2.7-code) include pdf after catalog/probe evidence; aiand treats
+# PDF as a provider-level Files API modality.
+# reasoning_effort verified live: gateway schema accepts
+# none/minimal/low/medium/high/xhigh/max, but the K3 backend only accepts
+# none/low/high/max (minimal/medium/xhigh rejected). Invalid values rejected
+# with 400 (negative control).
+base_model = "moonshotai/kimi-k3"
+
+reasoning_options = [{ type = "effort", values = ["none", "low", "high", "max"] }]
+
+[cost]
+input = 3.00
+cache_read = 0.50
+output = 12.50
+
+[modalities]
+input = ["text", "image", "pdf"]
+output = ["text"]
@@ -0,0 +1,26 @@
+# AIHubMix model page documents Kimi K3 pricing, supported input modalities
+# (text, vision, video), and a 1M-token context window.
+# Source accessed 2026-07-27:
+# https://aihubmix.com/model/kimi-k3
+
+base_model = "moonshotai/kimi-k3"
+last_updated = "2026-07-27"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "toggle"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "high", "max"]
+
+[cost]
+input = 3.00
+output = 15.00
+cache_read = 0.30
+
+[modalities]
+input = ["text", "image", "video"]
+output = ["text"]
@@ -5,7 +5,7 @@ temperature = true
 reasoning_options = []

 [cost]
-input = 0.82
-output = 3.75
+input = 0.73
+output = 3.5
 cache_read = 0.15
 cache_write = 0
@@ -1 +0,0 @@
-../../azure/models/cohere-command-r-08-2024.toml
@@ -1 +0,0 @@
-../../azure/models/cohere-command-r-plus-08-2024.toml
@@ -1 +0,0 @@
-../../azure/models/deepseek-r1-0528.toml
@@ -1 +0,0 @@
-../../azure/models/deepseek-v3-0324.toml
@@ -1 +0,0 @@
-../../azure/models/deepseek-v3.1.toml
@@ -1 +0,0 @@
-../../azure/models/gpt-3.5-turbo-0301.toml
@@ -1 +0,0 @@
-../../azure/models/gpt-3.5-turbo-0613.toml
@@ -1 +0,0 @@
-../../azure/models/gpt-4-32k.toml
@@ -1 +0,0 @@
-../../azure/models/gpt-4.toml
@@ -1 +0,0 @@
-../../azure/models/gpt-5-chat.toml
@@ -1 +0,0 @@
-../../azure/models/gpt-5.1-chat.toml
@@ -1 +0,0 @@
-../../azure/models/gpt-5.2-chat.toml
@@ -1 +0,0 @@
-../../azure/models/grok-4-fast-reasoning.toml
@@ -1 +0,0 @@
-../../azure/models/kimi-k2-thinking.toml
@@ -1 +0,0 @@
-../../azure/models/llama-3.2-11b-vision-instruct.toml
@@ -1 +0,0 @@
-../../azure/models/llama-3.2-90b-vision-instruct.toml
@@ -1 +0,0 @@
-../../azure/models/meta-llama-3-70b-instruct.toml
@@ -1 +0,0 @@
-../../azure/models/meta-llama-3-8b-instruct.toml
@@ -1 +0,0 @@
-../../azure/models/meta-llama-3.1-405b-instruct.toml
@@ -1 +0,0 @@
-../../azure/models/meta-llama-3.1-70b-instruct.toml
@@ -1 +0,0 @@
-../../azure/models/meta-llama-3.1-8b-instruct.toml
@@ -1 +0,0 @@
-../../azure/models/mistral-large-2411.toml
@@ -1 +0,0 @@
-../../azure/models/mistral-nemo.toml
@@ -1 +0,0 @@
-../../azure/models/o1-mini.toml
@@ -1 +0,0 @@
-../../azure/models/phi-3-medium-128k-instruct.toml
@@ -1 +0,0 @@
-../../azure/models/phi-3-medium-4k-instruct.toml
@@ -1 +0,0 @@
-../../azure/models/phi-3-mini-128k-instruct.toml
@@ -1 +0,0 @@
-../../azure/models/phi-3-mini-4k-instruct.toml
@@ -1 +0,0 @@
-../../azure/models/phi-3-small-128k-instruct.toml
@@ -1 +0,0 @@
-../../azure/models/phi-3-small-8k-instruct.toml
@@ -1 +0,0 @@
-../../azure/models/phi-3.5-mini-instruct.toml
@@ -1 +0,0 @@
-../../azure/models/phi-3.5-moe-instruct.toml
@@ -10,6 +10,7 @@ temperature = false
 knowledge = "2024-04"
 tool_call = true
 open_weights = false
+status = "deprecated"

 [cost]
 input = 1.50
@@ -1,5 +0,0 @@
-base_model = "cohere/command-r-08-2024"
-
-[cost]
-input = 0.15
-output = 0.6
@@ -1,5 +0,0 @@
-base_model = "cohere/command-r-plus-08-2024"
-
-[cost]
-input = 2.5
-output = 10
@@ -1,24 +0,0 @@
-name = "DeepSeek-R1-0528"
-description = "DeepSeek reasoning model for multi-step analysis, math, coding, and tools"
-family = "deepseek-thinking"
-release_date = "2025-05-28"
-last_updated = "2025-05-28"
-attachment = false
-reasoning = true
-reasoning_options = []
-temperature = true
-knowledge = "2024-07"
-tool_call = true
-open_weights = true
-
-[cost]
-input = 1.35
-output = 5.4
-
-[limit]
-context = 163_840
-output = 163_840
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -10,6 +10,7 @@ temperature = true
 knowledge = "2024-07"
 tool_call = false
 open_weights = true
+status = "deprecated"

 [cost]
 input = 1.35
@@ -1,23 +0,0 @@
-name = "DeepSeek-V3-0324"
-description = "DeepSeek chat model for instruction following, coding, and analysis"
-family = "deepseek"
-release_date = "2025-03-24"
-last_updated = "2025-03-24"
-attachment = false
-reasoning = false
-temperature = true
-knowledge = "2024-07"
-tool_call = true
-open_weights = true
-
-[cost]
-input = 1.14
-output = 4.56
-
-[limit]
-context = 131_072
-output = 131_072
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,24 +0,0 @@
-name = "DeepSeek-V3.1"
-description = "DeepSeek chat model for instruction following, coding, and analysis"
-family = "deepseek"
-release_date = "2025-08-21"
-last_updated = "2025-08-21"
-attachment = false
-reasoning = true
-reasoning_options = []
-temperature = true
-knowledge = "2024-07"
-tool_call = true
-open_weights = true
-
-[cost]
-input = 0.56
-output = 1.68
-
-[limit]
-context = 131_072
-output = 131_072
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -9,6 +9,7 @@ temperature = true
 knowledge = "2021-08"
 tool_call = false
 open_weights = false
+status = "deprecated"

 [cost]
 input = 0.50
@@ -1,23 +0,0 @@
-name = "GPT-3.5 Turbo 0301"
-description = "Compact GPT model for low-latency assistance and high-volume workloads"
-family = "gpt"
-release_date = "2023-03-01"
-last_updated = "2023-03-01"
-attachment = false
-reasoning = false
-temperature = true
-knowledge = "2021-08"
-tool_call = false
-open_weights = false
-
-[cost]
-input = 1.50
-output = 2.00
-
-[limit]
-context = 4096
-output = 4096
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,23 +0,0 @@
-name = "GPT-3.5 Turbo 0613"
-description = "Compact GPT model for low-latency assistance and high-volume workloads"
-family = "gpt"
-release_date = "2023-06-13"
-last_updated = "2023-06-13"
-attachment = false
-reasoning = false
-temperature = true
-knowledge = "2021-08"
-tool_call = false
-open_weights = false
-
-[cost]
-input = 3.00
-output = 4.00
-
-[limit]
-context = 16384
-output = 16384
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -9,6 +9,7 @@ temperature = true
 knowledge = "2021-08"
 tool_call = false
 open_weights = false
+status = "deprecated"

 [cost]
 input = 1.00
@@ -9,6 +9,7 @@ temperature = true
 knowledge = "2021-08"
 tool_call = false
 open_weights = false
+status = "deprecated"

 [cost]
 input = 1.50
@@ -1,23 +0,0 @@
-name = "GPT-4 32K"
-description = "GPT model for general reasoning, writing, coding, and tool-assisted tasks"
-family = "gpt"
-release_date = "2023-03-14"
-last_updated = "2023-03-14"
-attachment = false
-reasoning = false
-temperature = true
-knowledge = "2023-11"
-tool_call = true
-open_weights = false
-
-[cost]
-input = 60.00
-output = 120.00
-
-[limit]
-context = 32768
-output = 32768
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -9,6 +9,7 @@ temperature = true
 knowledge = "2023-11"
 tool_call = true
 open_weights = false
+status = "deprecated"

 [cost]
 input = 10.00
@@ -1,4 +1,5 @@
 base_model = "openai/gpt-4-turbo"
+status = "deprecated"
 base_model_omit = ["structured_output"]

 [cost]
@@ -1,4 +1,5 @@
 base_model = "openai/gpt-4.1-mini"
+status = "deprecated"
 base_model_omit = ["structured_output"]

 [cost]
@@ -1,4 +1,5 @@
 base_model = "openai/gpt-4.1-nano"
+status = "deprecated"
 base_model_omit = ["structured_output"]

 [cost]
@@ -1,4 +1,5 @@
 base_model = "openai/gpt-4.1"
+status = "deprecated"
 base_model_omit = ["structured_output"]

 [cost]
@@ -1,23 +0,0 @@
-name = "GPT-4"
-description = "GPT model for general reasoning, writing, coding, and tool-assisted tasks"
-family = "gpt"
-release_date = "2023-03-14"
-last_updated = "2023-03-14"
-attachment = false
-reasoning = false
-temperature = true
-knowledge = "2023-11"
-tool_call = true
-open_weights = false
-
-[cost]
-input = 60.00
-output = 120.00
-
-[limit]
-context = 8192
-output = 8192
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,4 +1,5 @@
 base_model = "openai/gpt-4o-mini"
+status = "deprecated"
 base_model_omit = ["structured_output"]

 [cost]
@@ -1,4 +1,5 @@
 base_model = "openai/gpt-4o"
+status = "deprecated"
 base_model_omit = ["structured_output"]

 [cost]
@@ -1,25 +0,0 @@
-name = "GPT-5 Chat"
-description = "Chat-tuned GPT model for conversational assistance, writing, and tool workflows"
-family = "gpt-codex"
-release_date = "2025-08-07"
-last_updated = "2025-08-07"
-attachment = true
-reasoning = true
-reasoning_options = []
-temperature = false
-knowledge = "2024-10-24"
-tool_call = false
-open_weights = false
-
-[cost]
-input = 1.25
-output = 10.00
-cache_read = 0.13
-
-[limit]
-context = 128_000
-output = 16_384
-
-[modalities]
-input = ["text", "image"]
-output = ["text"]
@@ -1,26 +0,0 @@
-name = "GPT-5.1 Chat"
-description = "Speech generation model for controllable voice, narration, and audio delivery"
-family = "gpt-codex"
-release_date = "2025-11-14"
-last_updated = "2025-11-14"
-attachment = true
-reasoning = true
-reasoning_options = [{ type = "effort", values = ["medium"] }]
-temperature = false
-knowledge = "2024-09-30"
-tool_call = true
-structured_output = true
-open_weights = false
-
-[cost]
-input = 1.25
-output = 10.00
-cache_read = 0.125
-
-[limit]
-context = 128_000
-output = 16_384
-
-[modalities]
-input = ["text", "image", "audio"]
-output = ["text", "image", "audio"]
@@ -1,26 +0,0 @@
-name = "GPT-5.2 Chat"
-description = "Chat-tuned GPT model for conversational assistance, writing, and tool workflows"
-family = "gpt-codex"
-release_date = "2025-12-11"
-last_updated = "2025-12-11"
-attachment = true
-reasoning = true
-reasoning_options = [{ type = "effort", values = ["medium"] }]
-temperature = false
-knowledge = "2025-08-31"
-tool_call = true
-structured_output = true
-open_weights = false
-
-[cost]
-input = 1.75
-output = 14.00
-cache_read = 0.175
-
-[limit]
-context = 128_000
-output = 16_384
-
-[modalities]
-input = ["text", "image"]
-output = ["text"]
@@ -1,26 +0,0 @@
-name = "GPT-5.3 Chat"
-description = "Chat-tuned GPT model for conversational assistance, writing, and tool workflows"
-family = "gpt-codex"
-release_date = "2026-03-03"
-last_updated = "2026-03-03"
-attachment = true
-reasoning = true
-reasoning_options = [{ type = "effort", values = ["medium"] }]
-temperature = false
-knowledge = "2025-08-31"
-tool_call = true
-structured_output = true
-open_weights = false
-
-[cost]
-input = 1.75
-output = 14.00
-cache_read = 0.175
-
-[limit]
-context = 128_000
-output = 16_384
-
-[modalities]
-input = ["text", "image"]
-output = ["text"]
@@ -1,25 +0,0 @@
-name = "Grok 4 Fast (Reasoning)"
-description = "Fast Grok model for responsive chat, reasoning, and tool-assisted work"
-family = "grok"
-release_date = "2025-09-19"
-last_updated = "2025-09-19"
-attachment = true
-reasoning = true
-reasoning_options = []
-temperature = true
-knowledge = "2025-07"
-tool_call = 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"]
@@ -1,26 +0,0 @@
-name = "Kimi K2 Thinking"
-description = "Kimi reasoning model for long-horizon research, planning, and tool use"
-family = "kimi-thinking"
-release_date = "2025-11-06"
-last_updated = "2025-12-02"
-attachment = false
-reasoning = true
-reasoning_options = []
-temperature = true
-knowledge = "2024-08"
-tool_call = true
-interleaved = true
-open_weights = true
-
-[cost]
-input = 0.60
-output = 2.50
-cache_read = 0.15
-
-[limit]
-context = 262_144
-output = 262_144
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,23 +0,0 @@
-name = "Llama-3.2-11B-Vision-Instruct"
-description = "Open Llama multimodal model for image understanding and text reasoning"
-family = "llama"
-release_date = "2024-09-25"
-last_updated = "2024-09-25"
-attachment = true
-reasoning = false
-temperature = true
-knowledge = "2023-12"
-tool_call = true
-open_weights = true
-
-[cost]
-input = 0.37
-output = 0.37
-
-[limit]
-context = 128_000
-output = 8_192
-
-[modalities]
-input = ["text", "image"]
-output = ["text"]
@@ -1,23 +0,0 @@
-name = "Llama-3.2-90B-Vision-Instruct"
-description = "Open Llama multimodal model for image understanding and text reasoning"
-family = "llama"
-release_date = "2024-09-25"
-last_updated = "2024-09-25"
-attachment = true
-reasoning = false
-temperature = true
-knowledge = "2023-12"
-tool_call = true
-open_weights = true
-
-[cost]
-input = 2.04
-output = 2.04
-
-[limit]
-context = 128_000
-output = 8_192
-
-[modalities]
-input = ["text", "image"]
-output = ["text"]
@@ -1,23 +0,0 @@
-name = "Meta-Llama-3-70B-Instruct"
-description = "Open Llama instruction model for multilingual chat, reasoning, and coding"
-family = "llama"
-release_date = "2024-04-18"
-last_updated = "2024-04-18"
-attachment = false
-reasoning = false
-temperature = true
-knowledge = "2023-12"
-tool_call = false
-open_weights = true
-
-[cost]
-input = 2.68
-output = 3.54
-
-[limit]
-context = 8_192
-output = 2_048
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,23 +0,0 @@
-name = "Meta-Llama-3-8B-Instruct"
-description = "Open Llama instruction model for multilingual chat, reasoning, and coding"
-family = "llama"
-release_date = "2024-04-18"
-last_updated = "2024-04-18"
-attachment = false
-reasoning = false
-temperature = true
-knowledge = "2023-12"
-tool_call = false
-open_weights = true
-
-[cost]
-input = 0.30
-output = 0.61
-
-[limit]
-context = 8_192
-output = 2_048
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,23 +0,0 @@
-name = "Meta-Llama-3.1-405B-Instruct"
-description = "Open Llama instruction model for multilingual chat, reasoning, and coding"
-family = "llama"
-release_date = "2024-07-23"
-last_updated = "2024-07-23"
-attachment = false
-reasoning = false
-temperature = true
-knowledge = "2023-12"
-tool_call = true
-open_weights = true
-
-[cost]
-input = 5.33
-output = 16.00
-
-[limit]
-context = 128_000
-output = 32_768
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,23 +0,0 @@
-name = "Meta-Llama-3.1-70B-Instruct"
-description = "Open Llama instruction model for multilingual chat, reasoning, and coding"
-family = "llama"
-release_date = "2024-07-23"
-last_updated = "2024-07-23"
-attachment = false
-reasoning = false
-temperature = true
-knowledge = "2023-12"
-tool_call = true
-open_weights = true
-
-[cost]
-input = 2.68
-output = 3.54
-
-[limit]
-context = 128_000
-output = 32_768
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,23 +0,0 @@
-name = "Meta-Llama-3.1-8B-Instruct"
-description = "Open Llama instruction model for multilingual chat, reasoning, and coding"
-family = "llama"
-release_date = "2024-07-23"
-last_updated = "2024-07-23"
-attachment = false
-reasoning = false
-temperature = true
-knowledge = "2023-12"
-tool_call = true
-open_weights = true
-
-[cost]
-input = 0.30
-output = 0.61
-
-[limit]
-context = 128_000
-output = 32_768
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,23 +0,0 @@
-name = "Mistral Large 24.11"
-description = "Flagship Mistral model for advanced reasoning, coding, and multilingual work"
-family = "mistral-large"
-release_date = "2024-11-01"
-last_updated = "2024-11-01"
-attachment = false
-reasoning = false
-temperature = true
-knowledge = "2024-09"
-tool_call = true
-open_weights = false
-
-[cost]
-input = 2.00
-output = 6.00
-
-[limit]
-context = 128_000
-output = 32_768
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,23 +0,0 @@
-name = "Mistral Nemo"
-description = "Mistral model for multilingual chat, reasoning, and tool-assisted workflows"
-family = "mistral-nemo"
-release_date = "2024-07-18"
-last_updated = "2024-07-18"
-attachment = false
-reasoning = false
-temperature = true
-knowledge = "2024-07"
-tool_call = true
-open_weights = true
-
-[cost]
-input = 0.15
-output = 0.15
-
-[limit]
-context = 128_000
-output = 128_000
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,25 +0,0 @@
-name = "o1-mini"
-description = "O-series reasoning model for hard analysis, math, coding, and planning"
-family = "o-mini"
-release_date = "2024-09-12"
-last_updated = "2024-09-12"
-attachment = false
-reasoning = true
-reasoning_options = []
-temperature = false
-knowledge = "2023-09"
-tool_call = true
-open_weights = false
-
-[cost]
-input = 1.10
-output = 4.40
-cache_read = 0.55
-
-[limit]
-context = 128_000
-output = 65_536
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -10,6 +10,7 @@ temperature = false
 knowledge = "2023-09"
 tool_call = true
 open_weights = false
+status = "deprecated"

 [cost]
 input = 15.00
@@ -1,4 +1,5 @@
 base_model = "openai/o3-mini"
+status = "deprecated"
 reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }]
 base_model_omit = ["structured_output"]

@@ -1,4 +1,5 @@
 base_model = "openai/o4-mini"
+status = "deprecated"
 reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }]
 base_model_omit = ["structured_output"]

@@ -1,23 +0,0 @@
-name = "Phi-3-medium-instruct (128k)"
-description = "Open-weight instruction model for adaptable chat and self-hosted production workloads"
-family = "phi"
-release_date = "2024-04-23"
-last_updated = "2024-04-23"
-attachment = false
-reasoning = false
-temperature = true
-knowledge = "2023-10"
-tool_call = false
-open_weights = true
-
-[cost]
-input = 0.17
-output = 0.68
-
-[limit]
-context = 128_000
-output = 4_096
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,23 +0,0 @@
-name = "Phi-3-medium-instruct (4k)"
-description = "Open-weight instruction model for adaptable chat and self-hosted production workloads"
-family = "phi"
-release_date = "2024-04-23"
-last_updated = "2024-04-23"
-attachment = false
-reasoning = false
-temperature = true
-knowledge = "2023-10"
-tool_call = false
-open_weights = true
-
-[cost]
-input = 0.17
-output = 0.68
-
-[limit]
-context = 4_096
-output = 1_024
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,23 +0,0 @@
-name = "Phi-3-mini-instruct (128k)"
-description = "Efficient model for low-latency assistance, extraction, and routine automation"
-family = "phi"
-release_date = "2024-04-23"
-last_updated = "2024-04-23"
-attachment = false
-reasoning = false
-temperature = true
-knowledge = "2023-10"
-tool_call = false
-open_weights = true
-
-[cost]
-input = 0.13
-output = 0.52
-
-[limit]
-context = 128_000
-output = 4_096
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,23 +0,0 @@
-name = "Phi-3-mini-instruct (4k)"
-description = "Efficient model for low-latency assistance, extraction, and routine automation"
-family = "phi"
-release_date = "2024-04-23"
-last_updated = "2024-04-23"
-attachment = false
-reasoning = false
-temperature = true
-knowledge = "2023-10"
-tool_call = false
-open_weights = true
-
-[cost]
-input = 0.13
-output = 0.52
-
-[limit]
-context = 4_096
-output = 1_024
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,23 +0,0 @@
-name = "Phi-3-small-instruct (128k)"
-description = "Efficient model for low-latency assistance, extraction, and routine automation"
-family = "phi"
-release_date = "2024-04-23"
-last_updated = "2024-04-23"
-attachment = false
-reasoning = false
-temperature = true
-knowledge = "2023-10"
-tool_call = false
-open_weights = true
-
-[cost]
-input = 0.15
-output = 0.60
-
-[limit]
-context = 128_000
-output = 4_096
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,23 +0,0 @@
-name = "Phi-3-small-instruct (8k)"
-description = "Efficient model for low-latency assistance, extraction, and routine automation"
-family = "phi"
-release_date = "2024-04-23"
-last_updated = "2024-04-23"
-attachment = false
-reasoning = false
-temperature = true
-knowledge = "2023-10"
-tool_call = false
-open_weights = true
-
-[cost]
-input = 0.15
-output = 0.60
-
-[limit]
-context = 8_192
-output = 2_048
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,23 +0,0 @@
-name = "Phi-3.5-mini-instruct"
-description = "Efficient model for low-latency assistance, extraction, and routine automation"
-family = "phi"
-release_date = "2024-08-20"
-last_updated = "2024-08-20"
-attachment = false
-reasoning = false
-temperature = true
-knowledge = "2023-10"
-tool_call = false
-open_weights = true
-
-[cost]
-input = 0.13
-output = 0.52
-
-[limit]
-context = 128_000
-output = 4_096
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -1,23 +0,0 @@
-name = "Phi-3.5-MoE-instruct"
-description = "Open-weight instruction model for adaptable chat and self-hosted production workloads"
-family = "phi"
-release_date = "2024-08-20"
-last_updated = "2024-08-20"
-attachment = false
-reasoning = false
-temperature = true
-knowledge = "2023-10"
-tool_call = false
-open_weights = true
-
-[cost]
-input = 0.16
-output = 0.64
-
-[limit]
-context = 128_000
-output = 4_096
-
-[modalities]
-input = ["text"]
-output = ["text"]
@@ -0,0 +1,14 @@
+base_model = "moonshotai/kimi-k3"
+description = "Kimi multimodal agent model for visual understanding, coding, and planning"
+temperature = true
+reasoning_options = []
+
+[cost]
+input = 3
+output = 15
+
+[limit]
+output = 262_144
+
+[modalities]
+input = ["text", "image"]
@@ -19,4 +19,4 @@ cache_read = 0.21

 [limit]
 context = 524_288
-output = 524_288
+output = 262_144
@@ -17,4 +17,4 @@ cache_read = 0.3

 [limit]
 context = 524_288
-output = 524_288
+output = 262_144
@@ -0,0 +1,8 @@
+base_model = "anthropic/claude-opus-5"
+reasoning_options = [{ type = "effort", values = ["low", "medium", "high", "xhigh", "max"] }]
+
+[cost]
+input = 5
+output = 25
+cache_read = 0.5
+cache_write = 6.25
@@ -0,0 +1,17 @@
+base_model = "moonshotai/kimi-k3"
+reasoning_options = [{ type = "effort", values = ["none", "low", "high", "max"] }]
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 2
+output = 8
+cache_read = 0.25
+
+[limit]
+context = 1_000_000
+output = 262_144
+
+[provider]
+npm = "@ai-sdk/openai-compatible"
@@ -0,0 +1,14 @@
+base_model = "anthropic/claude-opus-5"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high", "xhigh", "max"]
+
+[cost]
+input = 5
+output = 25
+cache_read = 0.5
+cache_write = 6.25
+
+[modalities]
+input = ["text", "image"]
@@ -0,0 +1,20 @@
+# DeepInfra OpenAI chat (POST https://api.deepinfra.com/v1/openai/chat/completions)
+# model = "moonshotai/Kimi-K3".
+# Always-on reasoning; effort via top-level reasoning_effort = "low"|"high"|"max" (default max).
+# No thinking toggle — K3 always returns reasoning_content.
+# DeepInfra OpenAPI accepts max among effort values; model-native values are low/high/max.
+# https://docs.deepinfra.com/api-reference/chat-completions/openai-chat-completions
+# https://platform.kimi.ai/docs/guide/use-reasoning-effort
+base_model = "moonshotai/kimi-k3"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "high", "max"]
+
+[cost]
+input = 2.7
+output = 13.5
+cache_read = 0.27
+
+[modalities]
+input = ["text", "image"]
@@ -1,23 +1,26 @@
-name = "Qwen3-32B"
+name = "Qwen3 32B"
 description = "Qwen instruction model for multilingual chat, reasoning, and tool use"
 family = "qwen"
 release_date = "2025-04-30"
 last_updated = "2026-04-16"
 attachment = false
 reasoning = true
-reasoning_options = [{ type = "effort", values = ["none", "low", "medium", "high", "max"] }]
 temperature = true
 tool_call = true
 structured_output = true
 open_weights = true

+[[reasoning_options]]
+type = "effort"
+values = ["none", "low", "medium", "high", "max"]
+
 [cost]
 input = 0.25
 output = 0.55

 [limit]
-context = 131_000
-output = 40_960
+context = 32_768
+output = 6_554

 [modalities]
 input = ["text"]
@@ -1,26 +1,29 @@
-name = "Claude Opus 4.1"
+name = "Anthropic Claude 4.1 Opus"
 description = "Flagship Claude model for deep reasoning, coding, and long-horizon agents"
 family = "claude-opus"
 release_date = "2025-08-05"
 last_updated = "2025-08-05"
 attachment = true
 reasoning = true
-reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }]
 temperature = true
 tool_call = true
 knowledge = "2025-03-31"
 open_weights = false

+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high"]
+
 [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
-output = 32_000
+output = 8_192

 [modalities]
-input = ["text", "image", "pdf"]
+input = ["text", "image"]
 output = ["text"]
@@ -1,33 +1,36 @@
-name = "Claude Sonnet 4.5"
+name = "Anthropic Claude 4.5 Sonnet"
 description = "Balanced Claude model for coding, analysis, agent workflows, and cost control"
 family = "claude-sonnet"
 release_date = "2025-09-29"
 last_updated = "2025-09-29"
 attachment = true
 reasoning = true
-reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }]
 temperature = true
 tool_call = true
 knowledge = "2025-07-31"
 open_weights = false

+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high"]
+
 [cost]
-input = 3.00
-output = 15.00
-cache_read = 0.30
+input = 3
+output = 15
+cache_read = 0.3
 cache_write = 3.75

 [[cost.tiers]]
-tier = { size = 200_000 }
-input = 6.00
-output = 22.50
-cache_read = 0.30
-cache_write = 3.75
+tier = { type = "context", size = 200_000 }
+input = 6
+output = 22.5
+cache_read = 0.6
+cache_write = 7.5

 [limit]
-context = 1_000_000
+context = 200_000
 output = 64_000

 [modalities]
-input = ["text", "image", "pdf"]
+input = ["text", "image"]
 output = ["text"]
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
github-actions[bot]	18cd726008	chore(sync): update DigitalOcean model catalog	2026-07-28 10:09:45 +00:00
Billy Cao	52d5045ee7	feat(synthetic): Add Kimi K3 model (#3794 ) Deploy / deploy (push) Has been cancelled * Add Synthetic's Kimi K3 offering * Update cache read price * fix(synthetic): declare effort-only reasoning for Kimi K3 per Synthetic API docs Synthetic's OpenAI-compatible chat completions API documents reasoning_effort with values low \| medium \| high and no reasoning on/off toggle, so drop the toggle option and align effort values with the documented surface (matching the existing Synthetic Kimi K2.6 / K2.7-Code entries). https://dev.synthetic.new/docs/openai/chat-completions Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * Remove unnecessary comment * Retrigger transient actions failure --------- Co-authored-by: Claude Fable 5 <noreply@anthropic.com>	2026-07-27 23:36:15 -05:00
github-actions[bot]	4faf76317a	chore(sync): update Venice model catalog (#3791 ) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>	2026-07-27 23:36:01 -05:00
github-actions[bot]	efb5d8ea0d	chore(sync): update Baseten model catalog (#3798 ) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>	2026-07-27 23:35:53 -05:00
Fenil Modi	0273194a65	feat(aiand): add Kimi K3 (#3799 ) * feat(aiand): add Kimi K3 * fix(aiand): remove pdf from kimi-k3 modalities, text+image only * fix(aiand): restore pdf modality for kimi-k3 based on /v1/models evidence PDF was removed to match fireworks/ollama-cloud, but that is not valid evidence for aiand. Sibling aiand entries (kimi-k2.6, kimi-k2.7-code) keep pdf after catalog/probe evidence. Restoring pdf per original GET /v1/models data which showed document support.	2026-07-27 23:35:42 -05:00
Abliteration AI	a034112075	Add abliterated-model-large (#3793 ) * Add abliterated-model-large * Fix reasoning abliterated-model-large.toml * fix provider * Update abliterated-model-large.toml * removed interleaved reasoning * fixed docs and effort * Address review: verified reasoning controls, citations, provider docs - abliterated-model: reasoning = true with effort ladder and toggle, per docs.abliteration.ai/capabilities/thinking - abliterated-model-large: replace unverified effort values with the documented ladder (none..max via reasoning_effort) plus thinking toggle; add API-syntax comments; move all source citations into the leading header block; align max output with docs (999,990) - provider.toml: restore reasoning notes with the current verified per-endpoint request fields * Narrow abliterated-model-large effort values to distinct modes The API maps minimal-high -> high and xhigh-max -> max, so only none, high, and max are distinct outcomes. Alias mapping kept as a comment.	2026-07-27 23:27:01 -05:00
Oskar	b91080aa0e	feat(hyper): add Charm Hyper provider and sync module (#3352 ) * feat(hyper): add Charm Hyper provider and sync module * feat: resync models * fix: remove references to /provider endpoint * feat: simplify model resolution * fix logo * feat: add base model resolution * update models * feat: add reasoning_options with base model fallback * fix: undo env relaxation * feat: round prices * fix(hyper): sync modalities from vision * . * fix(hyper): remove base model reasoning fallback	2026-07-27 23:23:43 -05:00
Aiden Cline	6fda2e07c2	fix(nvidia): add missing NIM chat models and correct API ids (#3744 ) * fix(nvidia): add missing NIM chat models and correct API ids Add high-demand NVIDIA NIM models used by OpenCode (Nemotron Super/Ultra/Nano, Inkling, Laguna XS, Mistral Medium 3.5, Ministral 14B, Gemma 3, Cosmos Reason2) and rename catalog ids that used underscores so they match integrate.api.nvidia.com. Fixes anomalyco/opencode#38865 * fix(nvidia): audit NIM reasoning_options against infer docs Keep only verified controls (mistral-medium-3.5-128b reasoning_effort none\|high). Set reasoning_options=[] and drop interleaved where NIM OpenAPI does not document a control. Narrow inkling modalities to text+image and cite max_tokens bounds for Super/Laguna output limits. * fix(nvidia): restore verified Nemotron prompt toggles and Inkling audio First-party NIM model cards document reasoning ON/OFF via system prompts for Super v1/v1.5, Ultra 253B, and Nano 8B. NVIDIA's Inkling card lists text/image/audio inputs. Keep empty reasoning_options only where no control is documented (Laguna, VL models). Align max_tokens with infer OpenAPI bounds.	2026-07-27 20:45:13 -05:00
github-actions[bot]	24b7a2aa4c	chore(sync): update OpenRouter model catalog (#3758 ) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>	2026-07-27 20:44:47 -05:00
Vladimir Glafirov	d1d08289ac	feat: add gitlab duo-chat-opus-5 model (#3765 )	2026-07-27 20:44:27 -05:00
rakshith1928	9875219078	feat(kimi-k3): add Kimi K3 model configuration with pricing and modalities (#3789 )	2026-07-27 20:43:33 -05:00
amrrs	ec23529c0c	fix(nebius): fix Kimi K3 reasoning_options for Nebius Token Factory (#3792 ) * fix(nebius): curate Kimi K3 reasoning_options from verified API behavior PR #3780 merged Kimi K3 for Nebius with reasoning_options = [] (no verified control surface). Live testing against api.tokenfactory.nebius.com/v1/chat/completions shows reasoning_effort is a real, validated parameter: invalid values 422, and valid values visibly change reasoning_content length. Curate the accepted literal list instead of leaving it empty. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * fix(nebius): narrow Kimi K3 reasoning_options to backend-verified values Live testing invoking each literal (not just triggering the generic gateway validator) shows the sglang model backend itself rejects "minimal" and "xhigh" with a 400: "Input should be 'none', 'low', 'medium', 'high' or 'max'". Those two were only accepted by the gateway's shared schema, not by this model. Narrow the list to the 5 values that actually work end-to-end. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> --------- Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>	2026-07-27 20:43:18 -05:00
KiKaraage	62ef55a446	feat(crof): add Kimi K3 (#3795 ) * feat(crof): add Kimi K3 * fix(crof): change reasoning levels to low-high-max * fix(crof): add "none" reasoning back for Kimi K3	2026-07-27 20:42:40 -05:00
github-actions[bot]	6eaf975918	chore(sync): update Venice model catalog (#3786 ) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>	2026-07-27 16:06:11 -05:00
rakshith1928	f8ac4b4fb1	feat(ollama-cloud): add kimi k3 model (#3787 ) * feat(ollama-cloud): add kimi k3 model * update ollama reasoning * Revise Kimi K3 model documentation and sources	2026-07-27 16:06:00 -05:00
github-actions[bot]	03e2178662	chore(sync): update Baseten model catalog (#3770 ) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>	2026-07-27 14:50:15 -05:00
github-actions[bot]	03e495d946	chore(sync): update Ambient model catalog (#3771 ) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>	2026-07-27 14:46:40 -05:00
github-actions[bot]	1f5a03df40	chore(sync): update Vercel AI Gateway model catalog (#3772 ) * chore(sync): update Vercel AI Gateway model catalog * fix(vercel): set Kimi K3 Fast reasoning effort options K3 always reasons; expose verified low/high/max via reasoning_effort. --------- Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Aiden Cline <aidenpcline@gmail.com>	2026-07-27 14:46:30 -05:00
github-actions[bot]	0968fea09f	chore(sync): update LLM Gateway model catalog (#3779 ) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>	2026-07-27 14:43:34 -05:00
github-actions[bot]	1fb770040a	chore(sync): update Deep Infra model catalog (#3782 ) * chore(sync): update Deep Infra model catalog * fix(deepinfra): set Kimi-K3 reasoning effort options K3 always reasons; expose verified low/high/max via reasoning_effort. --------- Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Aiden Cline <aidenpcline@gmail.com>	2026-07-27 14:43:20 -05:00
Kevin	1efc768ae5	Add Kimi K3 to Nebius Token Factory (#3780 ) Register moonshotai/Kimi-K3 with Nebius pricing and limits from https://tokenfactory.nebius.com/api/public/models_info.	2026-07-27 14:42:37 -05:00
github-actions[bot]	c3aab14477	chore(sync): update Hugging Face model catalog (#3784 ) * chore(sync): update Hugging Face model catalog * fix(huggingface): set Kimi-K3 reasoning effort options K3 always reasons; expose verified low/high/max via reasoning_effort. --------- Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Aiden Cline <aidenpcline@gmail.com>	2026-07-27 14:42:20 -05:00
Zain Hasan	236d2dd99a	add kimi k3 (#3783 )	2026-07-27 14:37:56 -05:00
github-actions[bot]	ad211c8f8f	chore(sync): update Venice model catalog (#3781 ) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>	2026-07-27 14:37:42 -05:00
Jack	38ccccc20d	add kimi k3 to Zen	2026-07-28 01:08:00 +08:00
github-actions[bot]	cce20188e5	chore(sync): update Venice model catalog (#3775 ) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>	2026-07-27 11:33:49 -05:00
Ahmad Shahzad	2acddd4818	feat(fireworks-ai): add Kimi K3 and Kimi K3 Fast (#3777 )	2026-07-27 11:33:33 -05:00
Ahmad Shahzad	c67dbc2e02	fix(fireworks-ai): remove deprecated GLM 5.1 and GLM 5.1 Fast (#3730 ) Deploy / deploy (push) Has been cancelled Fireworks AI will decommission GLM 5.1 and GLM 5.1 Fast serverless endpoints on 2026-07-26, with GLM 5.2 and GLM 5.2 Fast serving as their recommended replacements: GLM 5.1 -> GLM 5.2 (accounts/fireworks/models/glm-5p2) GLM 5.1 Fast -> GLM 5.2 Fast (accounts/fireworks/routers/glm-5p2-fast) Remove the two model files ahead of the decommission date. Dedicated deployments are unaffected.	2026-07-26 22:48:41 -05:00
github-actions[bot]	790e5cb842	chore(sync): update Ambient model catalog (#3747 ) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>	2026-07-26 22:48:00 -05:00
github-actions[bot]	73160c42bd	chore(sync): update OpenRouter model catalog (#3748 ) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>	2026-07-26 22:47:50 -05:00
Carlo Francisco	ff9bc91921	fix(thinkingmachines): name 256K variant "Inkling (256K)" (#3755 ) Both Tinker Inkling tiers rendered with the same display name "Inkling" because the :peft:262144 variant inherits it via base_model. Downstream consumers (e.g. opencode) show two indistinguishable entries despite different context windows and pricing. Override the name to match the "Inkling (256K)" label used on Tinker's pricing page.	2026-07-26 22:47:35 -05:00
Nathan Nguyen	9c249c78cb	feat(cloudflare-ai-gateway): add Claude Opus 5 (#3736 )	2026-07-26 15:33:34 -05:00
github-actions[bot]	c40d2ae925	chore(sync): update OpenRouter model catalog (#3733 ) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>	2026-07-25 22:57:04 -05:00
github-actions[bot]	71b3ca345d	chore(sync): update Vercel AI Gateway model catalog (#3732 ) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>	2026-07-25 22:56:55 -05:00
github-actions[bot]	0b0414d78e	chore(sync): update Weights & Biases model catalog (#3731 ) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>	2026-07-25 22:56:49 -05:00
github-actions[bot]	f5edd52931	chore(sync): update Ambient model catalog (#3745 ) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>	2026-07-25 22:34:31 -05:00
opencode-agent[bot]	d53651e3d9	fix(reviewer): compare reasoning options by API surface (#3746 ) Co-authored-by: Aiden Cline <63023139+rekram1-node@users.noreply.github.com>	2026-07-25 22:34:20 -05:00
Aiden Cline	2e25bad01c	chore(azure): remove retired models, mark deprecated still-serving (#3729 ) * chore(azure): remove retired models, mark deprecated still-serving Delete Foundry models past retirement (chat snapshots, Phi-3, old GPT-4, retired Meta/Cohere/DeepSeek/Mistral/xAI/Moonshot entries). Clean broken azure-cognitive-services symlinks that pointed at deleted azure models. Mark still-serving Deprecated/Legacy models with status = "deprecated" (gpt-4.1, gpt-4o, o1/o3-mini/o4-mini, codex-mini, gpt-image-1, deepseek-r1, claude-opus-4-1). Sources: - https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule - https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/retired-models * fix(azure): address review — Preview status + Nov-2025 cohort - Remove status=deprecated from gpt-image-1 and claude-opus-4-1 (official lifecycle is Preview, not Deprecated) - Delete remaining Nov-2025 OpenAI cohort for consistency with o1-mini: gpt-3.5-turbo-0125/1106/instruct, gpt-4-turbo, gpt-4-turbo-vision - Drop broken azure-cognitive-services symlinks * fix(azure): restore Nov-2025 OpenAI cohort as deprecated Azure schedule/retired-models pages do not list gpt-4-turbo or gpt-3.5-turbo-0125/1106/instruct as Retired. OpenAI still serves the turbo family (catalog marks deprecated). Restore these IDs with status=deprecated instead of deleting, matching OpenAI catalog policy. Keep o1-mini deleted (long shut down on OpenAI API).	2026-07-25 15:04:01 -05:00
github-actions[bot]	fcf16dcf64	chore(sync): update Ambient model catalog (#3727 ) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>	2026-07-25 13:52:39 -05:00
github-actions[bot]	8a61715de2	chore(sync): update CrossModel model catalog (#3743 ) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>	2026-07-25 13:14:58 -05:00
github-actions[bot]	5d913d45eb	chore(sync): update OpenRouter model catalog (#3723 ) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>	2026-07-24 20:26:19 -05:00
				`@@ -1 +0,0 @@`
				`../../azure/models/cohere-command-r-08-2024.toml`
				`@@ -1 +0,0 @@`
				`../../azure/models/cohere-command-r-plus-08-2024.toml`
				`@@ -1 +0,0 @@`
				`../../azure/models/grok-4-fast-reasoning.toml`
				`@@ -1 +0,0 @@`
				`../../azure/models/llama-3.2-11b-vision-instruct.toml`
				`@@ -1 +0,0 @@`
				`../../azure/models/llama-3.2-90b-vision-instruct.toml`
				`@@ -1 +0,0 @@`
				`../../azure/models/meta-llama-3-70b-instruct.toml`
				`@@ -1 +0,0 @@`
				`../../azure/models/meta-llama-3-8b-instruct.toml`
				`@@ -1 +0,0 @@`
				`../../azure/models/meta-llama-3.1-405b-instruct.toml`
				`@@ -1 +0,0 @@`
				`../../azure/models/meta-llama-3.1-70b-instruct.toml`