{ "schemaVersion": 1, "revision": 101, "models": { "anthropic": [ { "id": "claude-opus-5-5", "name": "Claude Opus 5.5", "recommended": true, "api": "anthropic-messages", "reasoning": true, "input": [ "text", "image", "document" ], "contextWindow": 1000000, "maxTokens": 128000, "cost": { "input": 4, "output": 20, "cacheRead": 0.2, "cacheWrite": 8 }, "tokenizerFamily": "anthropic-4.7+", "capabilities": { "agentic": "native", "coding": "native", "reasoning": "native" }, "effort": { "levels": [ "low", "medium", "high", "xhigh", "max", "adaptive" ], "native": "adaptive", "default": "high", "adaptive": true, "note": "Always thinks; reasoning cannot be switched off. Anthropic's own default effort is medium." }, "compat": { "supportsDeferredToolUniverse": true }, "params": { "cache_control": { "type": "ephemeral", "ttl": "1h" } }, "systemRoleMessages": true, "forcedToolChoice": false, "releasedAt": "2026-09-22", "tokenUsage": "high", "tagline": "Flagship Claude for long-running agentic coding and knowledge work", "description": "Claude Opus 5.5 (22 Sep 2026): Anthropic's latest Opus, built for long-running agentic coding and knowledge work. Anthropic reports it performs at Fable 5.1's level on most work, ahead of it on agentic coding, knowledge work and computer use, and about 40% cheaper to run than Opus 5 on typical workloads. Sharper reading of charts, diagrams and screenshots. 1M-token context, 128K output, adaptive thinking always on. $4/$20 per 1M tokens with $0.20 cache reads.", "provider": "anthropic" }, { "id": "claude-fable-5-1", "name": "Claude Fable 5.1", "api": "anthropic-messages", "reasoning": true, "input": [ "text", "image", "document" ], "contextWindow": 1000000, "maxTokens": 128000, "cost": { "input": 10, "output": 50, "cacheRead": 0.25, "cacheWrite": 20 }, "tokenizerFamily": "anthropic-4.7+", "capabilities": { "agentic": "native", "coding": "native", "reasoning": "native" }, "effort": { "levels": [ "low", "medium", "high", "xhigh", "max", "adaptive" ], "native": "adaptive", "default": "high", "adaptive": true, "note": "Always thinks — reasoning cannot be switched off." }, "compat": { "supportsDeferredToolUniverse": true }, "params": { "cache_control": { "type": "ephemeral", "ttl": "1h" } }, "systemRoleMessages": true, "forcedToolChoice": false, "releasedAt": "2026-09-01", "tokenUsage": "high", "tagline": "Mythos-class Claude, at 2.5x Opus 5.5's price", "description": "Claude Fable 5.1 (1 Sep 2026): Anthropic's most capable model, the Mythos-class tier above Opus. It extends Fable 5 at the same $10/$50 per 1M token pricing but cuts cache reads by 75% to $0.25, roughly 25% cheaper for typical workloads and up to about 45% cheaper for highly agentic, long-context loops. Stronger long-horizon agentic coding, multistep research and document, spreadsheet and slide work. 1M-token context, 128K output, adaptive thinking always on. 2.5x Opus 5.5's input/output price, and Anthropic reports Opus 5.5 matches it on most work, so route it at work that earns it.", "provider": "anthropic" }, { "id": "claude-sonnet-5", "name": "Claude Sonnet 5", "api": "anthropic-messages", "reasoning": true, "input": [ "text", "image", "document" ], "contextWindow": 1000000, "maxTokens": 128000, "cost": { "input": 2, "output": 10, "cacheRead": 0.2, "cacheWrite": 4 }, "tokenizerFamily": "anthropic-4.7+", "capabilities": { "agentic": "native", "coding": "native", "reasoning": "native" }, "effort": { "levels": [ "off", "minimal", "low", "medium", "high", "xhigh", "max", "adaptive" ], "native": "adaptive", "default": "high", "adaptive": true, "note": "Thinking is on by default; off is an explicit opt-out." }, "compat": { "supportsDeferredToolUniverse": true }, "params": { "cache_control": { "type": "ephemeral", "ttl": "1h" } }, "releasedAt": "2026-06-30", "tokenUsage": "average", "tagline": "Best speed-to-intelligence Claude: near-Opus coding at Sonnet cost", "description": "Claude Sonnet 5 (30 Jun 2026): the next-generation Sonnet, reaching near-Opus quality on coding and agentic work at mid-tier price. Adaptive thinking on by default with low-to-max effort, high-resolution vision, a 1M-token context and 128K output. $2/$10 per 1M tokens. Uses a new tokenizer that produces ~30% more tokens than Sonnet 4.6.", "provider": "anthropic" }, { "id": "claude-haiku-4-5-20251001", "name": "Claude Haiku 4.5", "api": "anthropic-messages", "reasoning": true, "input": [ "text", "image", "document" ], "contextWindow": 200000, "maxTokens": 64000, "cost": { "input": 1, "output": 5, "cacheRead": 0.1, "cacheWrite": 2 }, "tokenizerFamily": "anthropic-pre-4.7", "capabilities": { "agentic": "limited", "tool-use": "limited", "reasoning": "limited", "chat": "native", "passthrough": "native" }, "effort": { "levels": [ "off", "minimal", "low", "medium", "high" ], "native": "off", "default": "high", "note": "Budget-based thinking only." }, "params": { "cache_control": { "type": "ephemeral", "ttl": "1h" } }, "releasedAt": "2025-10-15", "tokenUsage": "low", "tagline": "Fastest, cheapest Claude; near-frontier quality", "description": "Haiku 4.5 (15 Oct 2025): around 4-5x faster than Sonnet 4.5 at a fraction of the cost, while reaching about 90% of Sonnet 4.5's agentic-coding quality. 200k-token context, priced $1/$5. Ideal for high-volume, latency-sensitive work.", "provider": "anthropic" } ], "openai": [ { "id": "gpt-6-astra", "name": "GPT-6 Astra", "api": "openai-responses", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 1050000, "maxTokens": 128000, "cost": { "input": 10, "output": 50, "cacheRead": 1, "cacheWrite": 0 }, "tokenizerFamily": "openai-o200k", "capabilities": { "agentic": "native", "coding": "native" }, "effort": { "levels": [ "low", "medium", "high", "xhigh", "max" ], "native": "medium", "default": "high" }, "fast": { "kind": "service-tier", "tier": "priority" }, "compat": { "supportsToolSearch": true }, "releasedAt": "2026-09-03", "tokenUsage": "high", "tagline": "OpenAI's GPT-6 frontier tier, at 5x GPT-6 Sol's price", "description": "GPT-6 Astra (3 Sep 2026): OpenAI's first GPT-6 model and a generational step above the GPT-5.6 line for the hardest reasoning, coding and agentic work. 1.05M-token context (922k of it input), 128k-token output cap, and reasoning effort from low through max that cannot be switched off. At $10/$50 per 1M tokens it is five times GPT-6 Sol, so route it at work that earns it. Rollout is phased and began with OpenAI's cybersecurity access programme, so an account outside it may not resolve the id yet.", "provider": "openai" }, { "id": "gpt-6-sol", "name": "GPT-6 Sol", "recommended": true, "api": "openai-responses", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 1050000, "maxTokens": 128000, "cost": { "input": 2, "output": 10, "cacheRead": 0.2, "cacheWrite": 2.5 }, "tokenizerFamily": "openai-o200k", "capabilities": { "agentic": "native", "coding": "native" }, "effort": { "levels": [ "off", "low", "medium", "high", "xhigh", "max" ], "native": "medium", "default": "high", "toggles": { "verbosity": true } }, "fast": { "kind": "service-tier", "tier": "priority" }, "compat": { "supportsToolSearch": true }, "releasedAt": "2026-09-22", "tokenUsage": "average", "tagline": "Workhorse GPT-6 for coding and everyday work", "description": "GPT-6 Sol (22 Sep 2026): built for complex coding and agentic workflows, bringing much of GPT-6 Astra's strengths into a faster, cheaper model. OpenAI reports about half as many mistakes as GPT-5.6 Sol, with stronger factual reliability and clearer communication. 1.05M-token context (922k of it input), 128k-token output cap and reasoning effort from none through max. $2/$10 per 1M tokens.", "provider": "openai" }, { "id": "gpt-6-luna", "name": "GPT-6 Luna", "api": "openai-responses", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 1050000, "maxTokens": 128000, "cost": { "input": 0.1, "output": 0.5, "cacheRead": 0.01, "cacheWrite": 0.125 }, "tokenizerFamily": "openai-o200k", "capabilities": { "agentic": "native", "coding": "native" }, "effort": { "levels": [ "off", "low", "medium", "high", "xhigh", "max" ], "native": "medium", "default": "high", "toggles": { "verbosity": true } }, "fast": { "kind": "service-tier", "tier": "priority" }, "compat": { "supportsToolSearch": true }, "releasedAt": "2026-09-22", "tokenUsage": "low", "tagline": "OpenAI's most efficient model for high-volume work", "description": "GPT-6 Luna (22 Sep 2026): OpenAI's most efficient model for focused, high-volume tasks such as summarisation, extraction and focused coding, built on GPT-6 Astra's advances. 1.05M-token context (922k of it input), 128k-token output cap and reasoning effort from none through max. $0.10/$0.50 per 1M tokens.", "provider": "openai" }, { "id": "gpt-image-2", "name": "GPT Image 2", "reasoning": false, "input": [ "text" ], "tokenizerFamily": "openai", "capabilities": { "agentic": "limited", "vision": "unsupported", "image": "native", "chat": "unsupported", "tool-use": "unsupported" }, "effort": { "levels": [ "off" ], "native": "off", "default": "off" }, "releasedAt": "2026-04-21", "tokenUsage": "average", "tagline": "OpenAI image generation", "description": "gpt-image-2: OpenAI's current image generation model, billed per token on the API ($5 text in, $8 image in, $30 out per 1M). An image-generation specialist -- it does not chat or drive agent loops.", "provider": "openai" } ], "openai-codex": [ { "id": "gpt-6-astra", "name": "GPT-6 Astra", "api": "openai-codex-responses", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 872000, "maxTokens": 128000, "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, "tokenizerFamily": "openai-o200k", "capabilities": { "agentic": "native", "coding": "native" }, "effort": { "levels": [ "low", "medium", "high", "xhigh", "max" ], "native": "medium", "default": "high" }, "fast": { "kind": "service-tier", "tier": "priority" }, "compat": { "supportsToolSearch": true }, "releasedAt": "2026-09-03", "tokenUsage": "high", "tagline": "GPT-6 frontier tier on the ChatGPT subscription", "description": "GPT-6 Astra (3 Sep 2026) on the ChatGPT subscription surface: OpenAI's first GPT-6 model, a generational step above the GPT-5.6 tiers for the hardest reasoning, coding and agentic work. Plan-gated: confirmed for Pro, Business and Enterprise plans, and OpenAI's own pages disagree on Plus. The rollout is phased over days, so an eligible account may not see it immediately. Reasoning effort from low through max, plus a Fast tier. Plan-billed, so no per-token price.", "provider": "openai-codex" }, { "id": "gpt-6-sol", "name": "GPT-6 Sol", "recommended": true, "api": "openai-codex-responses", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 872000, "maxTokens": 128000, "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, "tokenizerFamily": "openai-o200k", "capabilities": { "agentic": "native", "coding": "native" }, "effort": { "levels": [ "low", "medium", "high", "xhigh", "max" ], "native": "medium", "default": "high", "toggles": { "verbosity": true } }, "fast": { "kind": "service-tier", "tier": "priority" }, "compat": { "supportsToolSearch": true }, "releasedAt": "2026-09-22", "tokenUsage": "average", "tagline": "Workhorse GPT-6 for coding and everyday work", "description": "GPT-6 Sol (22 Sep 2026) on the ChatGPT subscription surface: built for complex coding and agentic workflows, with stronger factual reliability and clearer communication than GPT-5.6 Sol. On Plus, Pro, Business, Enterprise and Edu plans in ChatGPT Work and Codex, rolling out, and Enterprise administrators must enable it. Reasoning effort from low through max, plus a Fast tier. Plan-billed, so no per-token price.", "provider": "openai-codex" }, { "id": "gpt-6-luna", "name": "GPT-6 Luna", "api": "openai-codex-responses", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 872000, "maxTokens": 128000, "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, "tokenizerFamily": "openai-o200k", "capabilities": { "agentic": "native", "coding": "native" }, "effort": { "levels": [ "low", "medium", "high", "xhigh", "max" ], "native": "medium", "default": "high", "toggles": { "verbosity": true } }, "fast": { "kind": "service-tier", "tier": "priority" }, "compat": { "supportsToolSearch": true }, "releasedAt": "2026-09-22", "tokenUsage": "low", "tagline": "Most efficient GPT-6 for focused, high-volume work", "description": "GPT-6 Luna (22 Sep 2026) on the ChatGPT subscription surface: OpenAI's most efficient model for focused, high-volume tasks, including summarisation, extraction and focused coding. On Plus, Pro, Business, Enterprise and Edu plans in ChatGPT Work and Codex, and in the desktop app for Free and Go. Reasoning effort from low through max, plus a Fast tier. Plan-billed, so no per-token price.", "provider": "openai-codex" }, { "id": "gpt-image-2", "name": "GPT Image 2", "reasoning": false, "input": [ "text", "image" ], "tokenizerFamily": "openai-codex", "capabilities": { "agentic": "limited", "image": "native", "chat": "unsupported", "tool-use": "unsupported" }, "effort": { "levels": [ "off" ], "native": "off", "default": "off" }, "releasedAt": "2026-04-21", "tokenUsage": "average", "tagline": "ChatGPT-subscription image generation", "description": "gpt-image-2: OpenAI's image generation model, billed to the connected ChatGPT plan's quota rather than API credits. Supports reference-image edits. An image-generation specialist -- it does not chat or drive agent loops. The ChatGPT surface serves a fixed size and quality: size, aspect-ratio and resolution overrides are not honoured.", "provider": "openai-codex" } ], "xai": [ { "id": "grok-4.7", "name": "Grok 4.7", "recommended": true, "api": "openai-completions", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 500000, "maxTokens": 64000, "cost": { "input": 2, "output": 6, "cacheRead": 0.5, "cacheWrite": 0 }, "tokenizerFamily": "xai", "capabilities": { "agentic": "native", "tool-use": "native", "coding": "native" }, "effort": { "levels": [ "low", "medium", "high", "xhigh" ], "native": "high", "default": "high" }, "releasedAt": "2026-09-21", "tokenUsage": "average", "tagline": "xAI's frontier Grok for coding, agentic and knowledge work", "description": "Grok 4.7 (21 Sep 2026): xAI's most capable model for coding, agentic tasks and knowledge work, on a new, larger base model with a longer reinforcement-learning run. It works longer on difficult tasks and checks its own work more carefully, at Grok 4.6's price and speed: $2/$6 per 1M in/out ($0.50 cached input), doubling above a 200k long-context band. 500k-token context, text-plus-image input.", "provider": "xai" }, { "id": "grok-4.20-0309-reasoning", "name": "Grok 4.20 Reasoning", "api": "openai-completions", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 1000000, "maxTokens": 64000, "cost": { "input": 1.25, "output": 2.5, "cacheRead": 0.2, "cacheWrite": 0 }, "tokenizerFamily": "xai", "capabilities": { "agentic": "native", "tool-use": "native" }, "effort": { "levels": [ "adaptive" ], "native": "adaptive", "default": "adaptive", "adaptive": true, "note": "Reasoning variant; use the separate non-reasoning model to disable thinking." }, "releasedAt": "2026-04-07", "tokenUsage": "average", "tagline": "Fast, low-hallucination agentic model with fixed reasoning", "description": "Grok 4.20 Reasoning (7 Apr 2026): the reasoning variant of xAI's fast, prompt-adherent 1M-context model, with text-plus-image input, structured outputs and agentic function calling. This exact dated id is served to the configured EU account.", "provider": "xai" }, { "id": "grok-4.20-0309-non-reasoning", "name": "Grok 4.20 Non-Reasoning", "api": "openai-completions", "reasoning": false, "input": [ "text", "image" ], "contextWindow": 1000000, "maxTokens": 64000, "cost": { "input": 1.25, "output": 2.5, "cacheRead": 0.2, "cacheWrite": 0 }, "tokenizerFamily": "xai", "capabilities": { "agentic": "limited", "tool-use": "native", "reasoning": "limited", "chat": "native" }, "effort": { "levels": [ "off" ], "native": "off", "default": "off" }, "releasedAt": "2026-04-07", "tokenUsage": "low", "tagline": "Fast 1M-context Grok 4.20 without reasoning latency", "description": "Grok 4.20 Non-Reasoning (7 Apr 2026): the low-latency variant of xAI's prompt-adherent 1M-context model, with text-plus-image input, structured outputs and function calling but no reasoning pass.", "provider": "xai" }, { "id": "grok-build-0.1", "name": "Grok Build 0.1", "api": "openai-completions", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 256000, "maxTokens": 64000, "cost": { "input": 1, "output": 2, "cacheRead": 0.2, "cacheWrite": 0 }, "tokenizerFamily": "xai", "capabilities": { "agentic": "limited", "tool-use": "limited", "coding": "native" }, "effort": { "levels": [ "adaptive" ], "native": "adaptive", "default": "adaptive", "adaptive": true, "note": "Always-on reasoning — not adjustable." }, "releasedAt": "2026-05-29", "tokenUsage": "average", "tagline": "xAI's coding model behind the Grok Build agent", "description": "grok-build-0.1 (29 May 2026): the public-beta API coding model powering xAI's Grok Build terminal agent. 256k-token context, text-plus-image input, 70.8% on SWE-Bench Verified. Tuned for software engineering and complex build tasks.", "provider": "xai" }, { "id": "grok-imagine-image-2.0", "name": "Grok Imagine Image 2.0", "reasoning": false, "input": [ "text" ], "tokenizerFamily": "xai", "capabilities": { "agentic": "limited", "vision": "unsupported", "image": "native", "chat": "unsupported", "tool-use": "unsupported" }, "effort": { "levels": [ "off" ], "native": "off", "default": "off" }, "tokenUsage": "average", "tagline": "xAI's recommended image generation model", "description": "grok-imagine-image-2.0: xAI's current Grok Imagine image model and the one xAI recommends, billed $0.04 per image. Takes aspect ratio and 1K/2K resolution. An image-generation specialist -- it does not chat or drive agent loops.", "provider": "xai" }, { "id": "grok-imagine-image-quality", "name": "Grok Imagine Image Quality", "reasoning": false, "input": [ "text" ], "tokenizerFamily": "xai", "capabilities": { "agentic": "limited", "vision": "unsupported", "image": "native", "chat": "unsupported", "tool-use": "unsupported" }, "effort": { "levels": [ "off" ], "native": "off", "default": "off" }, "tokenUsage": "high", "tagline": "Highest-quality Grok Imagine output", "description": "grok-imagine-image-quality: the highest-quality Grok Imagine tier, billed $0.05 per image. Takes aspect ratio and 1K/2K resolution. An image-generation specialist -- it does not chat or drive agent loops.", "provider": "xai" }, { "id": "grok-imagine-image", "name": "Grok Imagine Image", "reasoning": false, "input": [ "text" ], "tokenizerFamily": "xai", "capabilities": { "agentic": "limited", "vision": "unsupported", "image": "native", "chat": "unsupported", "tool-use": "unsupported" }, "effort": { "levels": [ "off" ], "native": "off", "default": "off" }, "tokenUsage": "low", "tagline": "Cheapest Grok Imagine tier", "description": "grok-imagine-image: the base Grok Imagine image model, billed $0.02 per image. Takes aspect ratio and 1K/2K resolution. An image-generation specialist -- it does not chat or drive agent loops.", "provider": "xai" } ], "google": [ { "id": "gemini-3.8-flash", "name": "Gemini 3.8 Flash", "recommended": true, "api": "google-generative-ai", "reasoning": true, "input": [ "text", "image", "document" ], "contextWindow": 1000000, "maxTokens": 65536, "cost": { "input": 1.5, "output": 7.5, "cacheRead": 0.15, "cacheWrite": 0 }, "tokenizerFamily": "google", "capabilities": { "agentic": "native", "coding": "native", "tool-use": "native" }, "effort": { "levels": [ "low", "medium", "high" ], "native": "medium", "default": "high" }, "releasedAt": "2026-09-02", "tokenUsage": "average", "tagline": "Google's newest Flash generation, tuned for long-horizon agents", "description": "Gemini 3.8 Flash (GA 2 Sep 2026): Google's newest Flash generation, tuned for long-horizon coding and autonomous agents. 1M-token context, 64K output, thinking, function calling and text, image, audio, video and PDF input. Standard $1.50/$7.50 per 1M in/out (introductory $0.75/$3.75 through 31 Dec 2026). Computer Use (preview). Thinking runs low, medium or high; there is no minimal tier.", "provider": "google" }, { "id": "gemini-3.1-pro-preview", "name": "Gemini 3.1 Pro", "api": "google-generative-ai", "reasoning": true, "input": [ "text", "image", "document" ], "contextWindow": 1000000, "maxTokens": 65536, "cost": { "input": 2.5, "output": 15, "cacheRead": 0.25, "cacheWrite": 0 }, "tokenizerFamily": "google", "capabilities": { "agentic": "native" }, "effort": { "levels": [ "low", "medium", "high", "adaptive" ], "native": "adaptive", "default": "high", "adaptive": true, "note": "Thinking cannot be disabled; high is dynamic." }, "releasedAt": "2026-02-19", "tokenUsage": "average", "tagline": "Google's most advanced reasoning model, natively multimodal", "description": "Gemini 3.1 Pro (preview, 19 Feb 2026): Google's most capable model for complex tasks, with a 1M-token context and native multimodal reasoning over text, audio, images, video, PDFs and whole code repositories. 77.1% on ARC-AGI-2 (more than double Gemini 3 Pro), stronger agentic and coding behaviour, and a new MEDIUM thinking level.", "provider": "google" }, { "id": "gemini-3.5-flash-lite", "name": "Gemini 3.5 Flash-Lite", "api": "google-generative-ai", "reasoning": true, "input": [ "text", "image", "document" ], "contextWindow": 1000000, "maxTokens": 65536, "cost": { "input": 0.3, "output": 2.5, "cacheRead": 0.03, "cacheWrite": 0 }, "tokenizerFamily": "google", "capabilities": { "agentic": "limited", "tool-use": "native", "reasoning": "limited", "chat": "native" }, "effort": { "levels": [ "minimal", "low", "medium", "high", "adaptive" ], "native": "minimal", "default": "high", "adaptive": true, "note": "Minimal is near-off, not guaranteed silent." }, "releasedAt": "2026-07-21", "tokenUsage": "low", "tagline": "Low-latency, low-cost Gemini for high-throughput subagent work", "description": "Gemini 3.5 Flash-Lite (GA 21 Jul 2026): Google's fast, cost-efficient model for high-volume automation, document extraction, structured output and autonomous subagent tasks. It has a 1M-token context, 64K output, thinking, function calling and Computer Use, at $0.30/$2.50 per 1M input/output tokens.", "provider": "google" }, { "id": "gemini-3.1-pro-preview-customtools", "name": "Gemini 3.1 Pro (Custom Tools)", "api": "google-generative-ai", "reasoning": true, "input": [ "text", "image", "document" ], "contextWindow": 1000000, "maxTokens": 65536, "cost": { "input": 2.5, "output": 15, "cacheRead": 0.25, "cacheWrite": 0 }, "tokenizerFamily": "google", "capabilities": { "agentic": "native", "tool-use": "native" }, "effort": { "levels": [ "low", "medium", "high", "adaptive" ], "native": "adaptive", "default": "high", "adaptive": true, "note": "Thinking cannot be disabled; high is dynamic." }, "releasedAt": "2026-02-19", "tokenUsage": "average", "tagline": "Gemini 3.1 Pro tuned for custom tool use", "description": "Gemini 3.1 Pro Preview (custom tools): the 19 Feb 2026 Gemini 3.1 Pro configured for custom tool / function calling, over the same 1M-token multimodal reasoning core.", "provider": "google" }, { "id": "gemini-3-flash-preview", "name": "Gemini 3 Flash", "api": "google-generative-ai", "reasoning": true, "input": [ "text", "image", "document" ], "contextWindow": 1000000, "maxTokens": 65536, "cost": { "input": 0.5, "output": 3, "cacheRead": 0.05, "cacheWrite": 0 }, "tokenizerFamily": "google", "capabilities": { "agentic": "limited", "chat": "native" }, "effort": { "levels": [ "minimal", "low", "medium", "high", "adaptive" ], "native": "adaptive", "default": "high", "adaptive": true }, "releasedAt": "2025-12-17", "tokenUsage": "low", "tagline": "Fast Gemini 3 with near-Pro performance", "description": "Gemini 3 Flash (17 Dec 2025): the fast tier of the Gemini 3 generation, delivering close to Pro-level quality at much lower latency and cost. Became the default model in the Gemini app, replacing 2.5 Flash.", "provider": "google" }, { "id": "gemini-3.1-flash-lite", "name": "Gemini 3.1 Flash-Lite", "api": "google-generative-ai", "reasoning": true, "input": [ "text", "image", "document" ], "contextWindow": 1000000, "maxTokens": 65536, "cost": { "input": 0.25, "output": 1.5, "cacheRead": 0.025, "cacheWrite": 0 }, "tokenizerFamily": "google", "capabilities": { "agentic": "limited", "reasoning": "limited", "chat": "native" }, "effort": { "levels": [ "minimal", "low", "medium", "high" ], "native": "minimal", "default": "high" }, "releasedAt": "2026-05-07", "tokenUsage": "low", "tagline": "Gemini's most cost-effective, low-latency tier", "description": "Gemini 3.1 Flash-Lite (GA 7 May 2026; replaces the shut-down preview id): the most cost-effective model in the Gemini 3.1 line, tuned for high-volume, latency-sensitive work while keeping Gemini 3 multimodal capabilities.", "provider": "google" }, { "id": "gemini-3.5-live-translate-preview", "name": "Gemini 3.5 Live Translate", "reasoning": true, "input": [ "text", "image", "document" ], "contextWindow": 1000000, "maxTokens": 65536, "tokenizerFamily": "google", "capabilities": { "agentic": "limited", "translate": "native", "chat": "unsupported", "tool-use": "unsupported" }, "effort": { "levels": [ "off", "minimal", "low", "medium", "high", "adaptive" ], "native": "off", "default": "medium" }, "releasedAt": "2026-05-20", "tokenUsage": "low", "tagline": "Real-time translation across 70+ languages", "description": "Gemini 3.5 Live Translate (preview): Google's dedicated real-time speech/text translation model covering 70+ languages. A translation specialist -- not a general chat or agent model.", "provider": "google" }, { "id": "gemini-3.1-flash-image", "name": "Gemini 3.1 Flash Image (Nano Banana 2)", "reasoning": true, "input": [ "text", "image", "document" ], "contextWindow": 131072, "maxTokens": 32768, "tokenizerFamily": "google", "capabilities": { "agentic": "limited", "image": "native", "chat": "unsupported", "tool-use": "unsupported" }, "effort": { "levels": [ "off", "minimal", "low", "medium", "high", "adaptive" ], "native": "off", "default": "medium" }, "releasedAt": "2026-05-28", "tokenUsage": "low", "tagline": "Fast, high-volume image generation and editing", "description": "Gemini 3.1 Flash Image, a.k.a. Nano Banana 2 (GA 28 May 2026): high-quality image generation and conversational editing optimized for speed and volume. An image-generation specialist -- it does not chat and cannot drive agent loops.", "provider": "google" }, { "id": "gemini-3.1-flash-lite-image", "name": "Gemini 3.1 Flash-Lite Image (Nano Banana 2 Lite)", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 65536, "maxTokens": 4096, "tokenizerFamily": "google", "capabilities": { "agentic": "limited", "image": "native", "chat": "unsupported", "tool-use": "unsupported" }, "effort": { "levels": [ "off", "minimal", "low", "medium", "high", "adaptive" ], "native": "off", "default": "medium" }, "releasedAt": "2026-06-30", "tokenUsage": "low", "tagline": "Ultra-low-latency image generation for interactive use", "description": "Gemini 3.1 Flash-Lite Image, a.k.a. Nano Banana 2 Lite (GA 30 Jun 2026): Google's ultra-low-latency, cost-efficient image generation and editing specialist for high-volume interactive work. It does not chat or drive agent loops.", "provider": "google" }, { "id": "gemini-3-pro-image", "name": "Gemini 3 Pro Image (Nano Banana Pro)", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 65536, "maxTokens": 32768, "tokenizerFamily": "google", "capabilities": { "agentic": "limited", "image": "native", "chat": "unsupported", "tool-use": "unsupported" }, "effort": { "levels": [ "off", "minimal", "low", "medium", "high", "adaptive" ], "native": "off", "default": "medium" }, "releasedAt": "2026-05-28", "tokenUsage": "average", "tagline": "Professional-grade image creation and precise editing", "description": "Gemini 3 Pro Image, a.k.a. Nano Banana Pro (GA 28 May 2026): professional-grade image creation and editing for complex design, high-fidelity mock-ups and accurate visual text. It does not chat or drive agent loops.", "provider": "google" } ], "cloud-ollama": [ { "id": "deepseek-v4-pro", "name": "DeepSeek V4 Pro", "recommended": true, "api": "ollama", "reasoning": true, "input": [ "text" ], "contextWindow": 1048576, "maxTokens": 32768, "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, "tokenizerFamily": "cloud-ollama", "capabilities": { "agentic": "native", "vision": "unsupported", "coding": "native", "reasoning": "native" }, "effort": { "levels": [ "off", "high", "max" ], "native": "high", "default": "high" }, "releasedAt": "2026-08-13", "tokenUsage": "high", "tagline": "Frontier open MoE: top coding, reasoning and agents", "description": "DeepSeek-V4-Pro (GA 13 Aug 2026, build 0813; in preview since 24 Apr): a 1.6T-parameter Mixture-of-Experts model (49B active) with a 1M-token context and three reasoning modes; the GA release sharpens agentic tool use, code execution and multi-step workflows. Frontier open-weight coding -- 93.5% LiveCodeBench, 3206 Codeforces, 80.6% SWE-bench Verified.", "provider": "cloud-ollama" }, { "id": "deepseek-v4.1-flash", "name": "DeepSeek V4.1 Flash", "api": "ollama", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 1048576, "maxTokens": 32768, "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, "tokenizerFamily": "cloud-ollama", "capabilities": { "agentic": "native" }, "effort": { "levels": [ "off", "high", "max" ], "native": "high", "default": "high" }, "releasedAt": "2026-09-10", "tokenUsage": "low", "tagline": "Multimodal DeepSeek Flash with 1M context", "description": "DeepSeek-V4.1-Flash (10 Sep 2026): the next Flash generation, a 552B-parameter sparse MoE on DeepSeek's Causal Encoder-Decoder architecture (8B active on input, 16B on output) with native visual understanding, the first DeepSeek Flash to take image input. 1M-token context, thinking and tool use. On DeepSeek's own API it is the `deepseek-flash` alias.", "provider": "cloud-ollama" }, { "id": "minimax-m3", "name": "MiniMax M3", "api": "ollama", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 524288, "maxTokens": 32768, "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, "tokenizerFamily": "cloud-ollama", "capabilities": { "agentic": "native", "coding": "native" }, "effort": { "levels": [ "off", "adaptive" ], "binary": true, "native": "adaptive", "default": "adaptive", "adaptive": true }, "releasedAt": "2026-06-01", "tokenUsage": "average", "tagline": "Open frontier coding with 1M-token context", "description": "MiniMax M3 (1 Jun 2026): MiniMax's agentic model using a sparse-attention architecture for a 1M-token context (512k guaranteed) with native text+image input. The first open model to combine frontier coding, million-token context and multimodality -- 83.5 on BrowseComp, ahead of Opus 4.7.", "provider": "cloud-ollama" }, { "id": "kimi-k2.7-code", "name": "Kimi K2.7 Code", "api": "ollama", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 262144, "maxTokens": 32768, "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, "tokenizerFamily": "cloud-ollama", "capabilities": { "agentic": "native", "coding": "native" }, "effort": { "levels": [ "adaptive" ], "native": "adaptive", "default": "adaptive", "adaptive": true, "note": "Always thinks — cannot be disabled." }, "releasedAt": "2026-06-12", "tokenUsage": "average", "tagline": "Moonshot's token-efficient 1T coding agent", "description": "Kimi K2.7 Code (12 Jun 2026): Moonshot AI's coding-focused 1.04T-parameter open MoE with a 256k-token context and vision, tuned for end-to-end software engineering across 10+ languages and production tech stacks. Uses ~30% fewer thinking tokens than K2.6 while scoring markedly higher on coding -- +21.8% on Kimi Code Bench v2 and +11.0% on Program Bench.", "provider": "cloud-ollama" }, { "id": "qwen3.5:397b", "name": "Qwen 3.5 397B", "api": "ollama", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 262144, "maxTokens": 32768, "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, "tokenizerFamily": "cloud-ollama", "capabilities": { "agentic": "native" }, "effort": { "levels": [ "off", "high" ], "binary": true, "native": "high", "default": "high", "note": "Boolean thinking only — level strings are rejected." }, "releasedAt": "2026-02-16", "tokenUsage": "high", "tagline": "Alibaba's flagship reasoning + vision model", "description": "Qwen3.5 397B (16 Feb 2026): Alibaba's flagship open multimodal model -- a 256k-token context with unified vision-language training and strong reasoning, coding and agentic use. Competitive with GPT-5.2 and Opus-class models; 88.6 on MathVision.", "provider": "cloud-ollama" }, { "id": "mistral-large-3:675b", "name": "Mistral Large 3 675B", "api": "ollama", "reasoning": false, "input": [ "text", "image" ], "contextWindow": 262144, "maxTokens": 32768, "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, "tokenizerFamily": "cloud-ollama", "capabilities": { "agentic": "limited", "reasoning": "limited", "chat": "native" }, "effort": { "levels": [ "off" ], "native": "off", "default": "off" }, "releasedAt": "2025-12-02", "tokenUsage": "high", "tagline": "Mistral's multimodal enterprise flagship", "description": "Mistral Large 3 675B (2 Dec 2025): Mistral AI's general-purpose multimodal MoE for production workloads -- a 256k-token context, vision, strong multilingual support and agentic function calling, released under Apache 2.0.", "provider": "cloud-ollama" }, { "id": "nemotron-3-ultra", "name": "Nemotron 3 Ultra", "api": "ollama", "reasoning": true, "input": [ "text" ], "contextWindow": 262144, "maxTokens": 32768, "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, "tokenizerFamily": "cloud-ollama", "capabilities": { "agentic": "native", "vision": "unsupported" }, "effort": { "levels": [ "off", "high" ], "binary": true, "native": "high", "default": "high" }, "releasedAt": "2026-06-04", "tokenUsage": "high", "tagline": "NVIDIA's 550B open flagship for long-horizon agents", "description": "NVIDIA Nemotron 3 Ultra (4 Jun 2026): a fully open 550B-parameter Mixture-of-Experts reasoning model (55B active) built for long-running agentic workflows -- agent orchestration, coding agents and deep research across hundreds of tool calls. NVFP4 4-bit optimisation delivers leading throughput at up to 30% lower cost than other open models, while leading on agent-productivity, instruction-following and long-context accuracy. 1M-token capable; served on Ollama Cloud at a 256k context.", "provider": "cloud-ollama" }, { "id": "nemotron-3-super", "name": "Nemotron 3 Super", "api": "ollama", "reasoning": true, "input": [ "text" ], "contextWindow": 262144, "maxTokens": 32768, "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, "tokenizerFamily": "cloud-ollama", "capabilities": { "agentic": "native", "vision": "unsupported" }, "effort": { "levels": [ "off", "high" ], "binary": true, "native": "high", "default": "high" }, "releasedAt": "2026-03-11", "tokenUsage": "low", "tagline": "NVIDIA's efficient agentic MoE", "description": "NVIDIA Nemotron 3 Super (11 Mar 2026): a 120B open Mixture-of-Experts model activating just 12B parameters, with a 256k-token context. Tuned for reasoning, tool use and multi-agent workflows at high computational efficiency.", "provider": "cloud-ollama" }, { "id": "gemma4:31b", "name": "Gemma 4 31B", "api": "ollama", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 262144, "maxTokens": 16384, "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, "tokenizerFamily": "cloud-ollama", "capabilities": { "agentic": "limited", "vision": "native" }, "effort": { "levels": [ "off", "high" ], "binary": true, "native": "high", "default": "high" }, "releasedAt": "2026-06-03", "tokenUsage": "low", "tagline": "Google's open multimodal model, 31B dense", "description": "Gemma 4 31B (3 Jun 2026): the largest dense tier of Google's Apache-2.0 Gemma 4 family -- a 256k-token context with configurable thinking and native text+image input (the smaller E2B/E4B tiers add audio). Strong for its size: 85.2% MMLU-Pro, 89.2% AIME 2026, 80.0% LiveCodeBench v6, 2150 Codeforces ELO. A capable generalist and vision model, not a frontier agentic driver.", "provider": "cloud-ollama" }, { "id": "glm-5.3", "name": "GLM 5.3", "api": "ollama", "reasoning": true, "input": [ "text" ], "contextWindow": 1048576, "maxTokens": 32768, "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, "tokenizerFamily": "cloud-ollama", "capabilities": { "agentic": "native", "vision": "unsupported", "coding": "native", "reasoning": "native" }, "effort": { "levels": [ "off", "high", "max" ], "native": "max", "default": "max" }, "releasedAt": "2026-08-26", "tokenUsage": "average", "tagline": "Z.ai's GLM-5.3 reasoning flagship, open weights", "description": "GLM-5.3 (26 Aug 2026): Z.ai's GLM-5.3 reasoning flagship for complex software engineering and long-horizon agents; open weights served on Ollama Cloud at a ~1M context with thinking and tools.", "provider": "cloud-ollama" }, { "id": "glm-5.3-flash", "name": "GLM 5.3 Flash", "api": "ollama", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 1048576, "maxTokens": 32768, "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, "tokenizerFamily": "cloud-ollama", "capabilities": { "agentic": "native", "coding": "native" }, "effort": { "levels": [ "off", "high", "max" ], "native": "max", "default": "max" }, "releasedAt": "2026-08-26", "tokenUsage": "low", "tagline": "Z.ai's first natively multimodal GLM-5, open weights", "description": "GLM-5.3-Flash (26 Aug 2026): Z.ai's first natively multimodal GLM-5 -- a 320B-A18B MoE with text+image input, thinking and tools; open weights served on Ollama Cloud at a ~1M context.", "provider": "cloud-ollama" } ], "zai": [ { "id": "glm-5.3", "name": "GLM-5.3", "recommended": true, "api": "openai-completions", "reasoning": true, "input": [ "text" ], "contextWindow": 1310720, "maxTokens": 131072, "cost": { "input": 1.4, "output": 4.4, "cacheRead": 0.26, "cacheWrite": 0 }, "tokenizerFamily": "zai", "capabilities": { "agentic": "native", "vision": "unsupported", "coding": "native", "reasoning": "native" }, "effort": { "levels": [ "low", "high", "max" ], "native": "max", "default": "max", "note": "Always thinks; effort maps to reasoning_effort low/high/max (default max)." }, "releasedAt": "2026-08-26", "tokenUsage": "average", "tagline": "Z.ai's GLM-5.3 flagship: reasoning-heavy engineering and agents", "description": "GLM-5.3 (26 Aug 2026): Z.ai's GLM-5.3 flagship, a large-scale reasoning model for complex software engineering and long-horizon agent tasks. 1.31M-token context, 131k output, reasoning always on with low/high/max effort. $1.40/$4.40 per 1M in/out.", "provider": "zai" }, { "id": "glm-5.3-flash", "name": "GLM-5.3 Flash", "api": "openai-completions", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 1310720, "maxTokens": 131072, "cost": { "input": 0.15, "output": 0.5, "cacheRead": 0.03, "cacheWrite": 0 }, "tokenizerFamily": "zai", "capabilities": { "agentic": "native", "coding": "native" }, "effort": { "levels": [ "low", "high", "max" ], "native": "max", "default": "max", "note": "Always thinks; effort maps to reasoning_effort low/high/max (default max)." }, "releasedAt": "2026-08-26", "tokenUsage": "low", "tagline": "Z.ai's first natively multimodal GLM-5, at flash price", "description": "GLM-5.3-Flash (26 Aug 2026): Z.ai's first natively multimodal GLM-5 -- a 320B-A18B MoE (18B active) with a 1.31M-token context, 131k output, text+image input and open weights. Beats GLM-5.2 on Z.ai's agentic and coding benchmarks (63.4 vs 46.2 DeepSWE v1.1, 48.8 vs 26.2 AutomationBench) at $0.15/$0.50 per 1M in/out ($0.03 cached). Always thinks.", "provider": "zai" } ], "moonshot": [ { "id": "kimi-k3", "name": "Kimi K3", "recommended": true, "api": "openai-completions", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 1048576, "maxTokens": 131072, "cost": { "input": 3, "output": 15, "cacheRead": 0.3, "cacheWrite": 0 }, "tokenizerFamily": "moonshot", "capabilities": { "agentic": "native", "coding": "native" }, "effort": { "levels": [ "low", "high", "max" ], "native": "max", "default": "max", "note": "Always thinks; effort is reasoning_effort low/high/max." }, "releasedAt": "2026-07-16", "tokenUsage": "high", "tagline": "Moonshot's 2.8T flagship: 1M context, always thinking", "description": "Kimi K3 (16 Jul 2026): Moonshot AI's 2.8T-parameter MoE flagship (Kimi Delta Attention, 16-of-896 active experts) with a 1M-token context, native vision and always-on thinking, built for long-horizon coding and end-to-end knowledge work. Vision input takes base64 or uploaded files only (no public URLs). Output defaults to 131k tokens and can be raised to 1M.", "provider": "moonshot" }, { "id": "kimi-k2.7-code", "name": "Kimi K2.7 Code", "api": "openai-completions", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 262144, "maxTokens": 32768, "cost": { "input": 0.95, "output": 4, "cacheRead": 0.19, "cacheWrite": 0 }, "tokenizerFamily": "moonshot", "capabilities": { "agentic": "native", "coding": "native" }, "effort": { "levels": [ "adaptive" ], "native": "adaptive", "default": "adaptive", "adaptive": true, "note": "Always thinks — cannot be disabled." }, "releasedAt": "2026-06-12", "tokenUsage": "average", "tagline": "Moonshot's token-efficient 1T coding agent", "description": "Kimi K2.7 Code (12 Jun 2026): Moonshot AI's coding-dedicated 1T-class open MoE with a 256k-token context and vision, tuned for end-to-end software engineering across 10+ languages and production tech stacks. Uses ~30% fewer thinking tokens than K2.6 while scoring markedly higher on coding.", "provider": "moonshot" }, { "id": "kimi-k2.7-code-highspeed", "name": "Kimi K2.7 Code HighSpeed", "api": "openai-completions", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 262144, "maxTokens": 32768, "cost": { "input": 1.9, "output": 8, "cacheRead": 0.38, "cacheWrite": 0 }, "tokenizerFamily": "moonshot", "capabilities": { "agentic": "native" }, "effort": { "levels": [ "adaptive" ], "native": "adaptive", "default": "adaptive", "adaptive": true, "note": "Always thinks — cannot be disabled." }, "releasedAt": "2026-07-09", "tokenUsage": "high", "tagline": "K2.7 Code served ~2x faster at 2x the price", "description": "Kimi K2.7 Code HighSpeed (9 Jul 2026): the same K2.7 Code model on a faster serving tier (~180 tok/s, bursts up to 260 tok/s) at double the per-token price. Pick it when wall-clock latency matters more than cost.", "provider": "moonshot" } ], "kimi": [ { "id": "kimi-for-coding", "name": "Kimi for Coding", "recommended": true, "api": "anthropic-messages", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 262144, "maxTokens": 32768, "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, "tokenizerFamily": "kimi", "capabilities": { "agentic": "native", "coding": "native" }, "effort": { "levels": [ "adaptive" ], "native": "adaptive", "default": "adaptive", "adaptive": true, "note": "Always thinks — effort is not configurable on this model." }, "releasedAt": "2026-06-12", "tokenUsage": "low", "tagline": "Kimi K2.7 Code on the coding subscription — the default", "description": "Kimi for Coding: Kimi K2.7 Code served on the Kimi Code subscription — plan-billed coding with a 256k-token context, vision and always-on thinking. The default Kimi Code model, available on every paid membership tier.", "provider": "kimi" }, { "id": "k3", "name": "Kimi K3 (subscription)", "api": "anthropic-messages", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 1048576, "maxTokens": 131072, "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, "tokenizerFamily": "kimi", "capabilities": { "agentic": "native", "coding": "native" }, "effort": { "levels": [ "low", "high", "max" ], "native": "high", "default": "high", "note": "Always thinks; effort is reasoning_effort low/high/max (default high)." }, "releasedAt": "2026-07-16", "tokenUsage": "high", "tagline": "Kimi K3 flagship with the full 1M context", "description": "Kimi K3 on the Kimi Code subscription: the 2.8T flagship with the full 1M-token context, vision and configurable thinking (low/high/max, default high). Requires the Allegretto tier or higher; the 1M context draws roughly twice the quota of k3-256k.", "provider": "kimi" }, { "id": "k3-256k", "name": "Kimi K3 256k (subscription)", "api": "anthropic-messages", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 262144, "maxTokens": 131072, "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, "tokenizerFamily": "kimi", "capabilities": { "agentic": "native", "coding": "native" }, "effort": { "levels": [ "low", "high", "max" ], "native": "high", "default": "high", "note": "Always thinks; effort is reasoning_effort low/high/max (default high)." }, "releasedAt": "2026-07-16", "tokenUsage": "average", "tagline": "Kimi K3 capped at 256k for lighter quota use", "description": "Kimi K3 on the Kimi Code subscription capped at a 256k-token context: the same K3 quality within 256k at roughly half the quota of the full 1M row. Requires the Moderato tier or higher.", "provider": "kimi" }, { "id": "kimi-for-coding-highspeed", "name": "Kimi for Coding HighSpeed", "api": "anthropic-messages", "reasoning": true, "input": [ "text", "image" ], "contextWindow": 262144, "maxTokens": 32768, "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, "tokenizerFamily": "kimi", "capabilities": { "agentic": "native" }, "effort": { "levels": [ "adaptive" ], "native": "adaptive", "default": "adaptive", "adaptive": true, "note": "Always thinks — effort is not configurable on this model." }, "releasedAt": "2026-07-09", "tokenUsage": "high", "tagline": "K2.7 Code on Kimi Code's fast serving tier", "description": "Kimi for Coding HighSpeed: the same K2.7 Code model on Kimi Code's fast serving tier at higher quota consumption. Requires the Allegretto tier or higher.", "provider": "kimi" } ], "fal": [ { "id": "fal-ai/flux/dev", "name": "FLUX.1 [dev]", "reasoning": false, "input": [ "text" ], "tokenizerFamily": "fal", "capabilities": { "agentic": "limited", "vision": "unsupported", "image": "native", "chat": "unsupported", "tool-use": "unsupported" }, "effort": { "levels": [ "off" ], "native": "off", "default": "off" }, "releasedAt": "2024-08-01", "tokenUsage": "average", "tagline": "Open-weight FLUX image generation on fal", "description": "FLUX.1 [dev] on fal: a 12B open-weight image model for high-quality text-to-image and single-reference editing. An image-generation specialist -- it does not chat or drive agent loops.", "provider": "fal" } ], "typesafe": [ { "id": "jev-1.13.0", "name": "Jev 1.13", "reasoning": false, "input": [ "text" ], "contextWindow": 64000, "cost": { "input": 0.042, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, "tokenizerFamily": "typesafe", "capabilities": { "agentic": "limited", "vision": "unsupported", "judgement": "native", "chat": "unsupported", "tool-use": "unsupported" }, "effort": { "levels": [ "off" ], "native": "off", "default": "off" }, "releasedAt": "2026-09-10", "tokenUsage": "low", "tagline": "Typed judgments with probabilities, in private preview", "description": "Jev 1.13 from TypeSafe answers narrow typed questions about a piece of text -- pick one option, rate against a rubric, or yes/no -- and returns probabilities in a fraction of a second. A judgment specialist: it writes no text, calls no tools, and cannot chat or drive agent loops. TypeSafe is in private preview, so a key needs an invitation.", "provider": "typesafe" } ] }, "superseded": { "anthropic/claude-opus-4": "claude-opus-5-5", "anthropic/claude-opus-4-1": "claude-opus-5-5", "anthropic/claude-opus-4-5": "claude-opus-5-5", "anthropic/claude-opus-4-6": "claude-opus-5-5", "anthropic/claude-opus-4-7": "claude-opus-5-5", "anthropic/claude-opus-4-8": "claude-opus-5-5", "anthropic/claude-opus-5": "claude-opus-5-5", "anthropic/claude-sonnet-4": "claude-sonnet-5", "anthropic/claude-sonnet-4-5": "claude-sonnet-5", "anthropic/claude-sonnet-4-6": "claude-sonnet-5", "anthropic/claude-fable-5": "claude-fable-5-1", "openai/gpt-5.4": "gpt-6-sol", "openai/gpt-5.4-mini": "gpt-6-luna", "openai/gpt-5.4-nano": "gpt-6-luna", "openai/gpt-5.4-pro": "gpt-6-astra", "openai/gpt-5.5": "gpt-6-sol", "openai/gpt-5.5-pro": "gpt-6-astra", "openai/gpt-image-1": "gpt-image-2", "openai/gpt-5.6-sol": "gpt-6-sol", "openai/gpt-5.6-terra": "gpt-6-sol", "openai/gpt-5.6-luna": "gpt-6-luna", "openai-codex/gpt-5.4": "gpt-6-sol", "openai-codex/gpt-5.4-mini": "gpt-6-luna", "openai-codex/gpt-5.5": "gpt-6-sol", "openai-codex/gpt-5.6-sol": "gpt-6-sol", "openai-codex/gpt-5.6-terra": "gpt-6-sol", "openai-codex/gpt-5.6-luna": "gpt-6-luna", "xai/grok-4.6": "grok-4.7", "xai/grok-4.5": "grok-4.7", "google/gemini-3.7-flash": "gemini-3.8-flash", "google/gemini-3.6-flash": "gemini-3.8-flash", "google/gemini-3.5-flash": "gemini-3.8-flash", "cloud-ollama/glm-5.2": "glm-5.3", "cloud-ollama/glm-5.1": "glm-5.3", "cloud-ollama/deepseek-v4-flash": "deepseek-v4.1-flash", "zai/glm-5.2": "glm-5.3", "zai/glm-5.1": "glm-5.3", "zai/glm-5": "glm-5.3-flash", "zai/glm-5-turbo": "glm-5.3-flash", "zai/glm-4.7": "glm-5.3-flash" }, "removed": [ { "pattern": "(?:^|\\/)(?:gpt-5[.-]3-codex)$", "flags": "i" }, { "pattern": "(?:^|\\/)(?:gpt-5[.-][45]-codex)$", "flags": "i" }, { "pattern": "^google\\/(?:gemini-3\\.1-flash-image-preview)$", "flags": "i" }, { "pattern": "^google\\/(?:gemini-3-pro-image-preview)$", "flags": "i" }, { "pattern": "^cloud-ollama\\/(?:qwen3-next:80b|qwen3-vl:235b|qwen3-coder:480b|devstral-2:123b|gemini-3-flash-preview)$", "flags": "i" }, { "pattern": "^moonshot\\/(?:kimi-k2-0711-preview|kimi-k2-0905-preview|kimi-k2-turbo-preview|kimi-k2-turbo|kimi-k2-thinking|kimi-k2-thinking-turbo|kimi-latest|kimi-thinking-preview)$", "flags": "i" } ], "disabled": [ "anthropic/claude-sonnet-4-6", "openai/gpt-5.3" ], "news": null }