AI models
Compare documented token limits, supported inputs and capabilities across AI models.
| Attribute | Claude Fable 5.2Anthropic | Claude Haiku 5.5Anthropic |
|---|---|---|
| Model identifier | Unknown | Unknown |
| OpenRouter model ID | anthropic/claude-fable-5.2 OpenRouter Models API · | anthropic/claude-haiku-5.5 OpenRouter Models API · |
| OpenRouter listed context | 1,000,000 tokens OpenRouter Models API · | 1,000,000 tokens OpenRouter Models API · |
| OpenRouter top provider output limit | 128,000 tokens OpenRouter Models API · | 128,000 tokens OpenRouter Models API · |
| OpenRouter listed input types | file, image, text OpenRouter Models API · | file, image, text OpenRouter Models API · |
| Family | Unknown | Unknown |
| Lifecycle | Unknown | Unknown |
| Weights | Unknown | Unknown |
| Published license | Unknown | Unknown |
| Documented context window | Unknown | Unknown |
| Separate input limit | Unknown | Unknown |
| Output limit | Unknown | Unknown |
| Input modalities | Unknown | Unknown |
| Function calling | Unknown | Unknown |
| Structured outputs | Unknown | Unknown |
| Reasoning controls | Unknown | Unknown |
Snapshot · Catalog and evidence ↗
Model benchmark ranking
Models ranked by independent benchmark score. Coding uses each model's best Terminal-Bench result in any agent; Overall uses the Artificial Analysis Intelligence Index at each model's highest-effort configuration. Models documented as lacking function calling are excluded; models whose function-calling support is not yet recorded are ranked and marked unverified. Models without a current score are listed as not benchmarked.
Terminal-Bench 4.1 · measured 2026-11-19 by Terminal-Bench 4.1 leaderboard. Best public leaderboard resolution rate for each model in any agent harness. The agent and effort behind the score are named with it.
- Claude Fable 5.2 · 65.8% Claude Code · max tools unverified
- Claude Opus 5.5 · 65.8% Claude Code · max tools unverified
- Gemini 4 Carbon · 64.5% Antigravity CLI · high
- Claude Sonnet 5.5 · 63.9% Claude Code · max tools unverified
- GPT-6.2 Astra · 62.1% Codex · max
- GPT-6.2 Sol · 61.5% Codex · max
- Gemini 4 Hydrogen · 56.4% Antigravity CLI · high
- Gemini 4 Argon · 55.5% Antigravity CLI · high
- GPT-6.1 Terra · 53.3% Codex · max
- Claude Haiku 5.5 · 40.0% Claude Code · max tools unverified
- Grok 4.7 · 38.8% Grok Build · xhigh tools unverified
- DeepSeek V4.1 Flash · 38.5% Claude Code · max tools unverified
- Mistral Large 4 · 27.6% Claude Code · max tools unverified
- Qwen3.8 Max (0902) · 27.0% Claude Code · max tools unverified
- GPT-5.6 Terra · 21.5% Codex · max tools unverified
- Gemini 3.8 Flash · 18.2% mini-SWE-agent · high
- GPT-6 Luna · 15.2% Codex · max
- DeepSeek V4 Pro · 6.7% Claude Code · max tools unverified
Not benchmarked: Claude Sonnet 4.6, Llama 4 Maverick, Llama 4 Scout, Qwen3-14B, Qwen3-32B, Qwen3.8 27B.
Artificial Analysis Intelligence Index v5 · measured 2026-11-19 by Artificial Analysis. Independent evaluation by Artificial Analysis on its standard harness, each model at the configuration named with its score.
- Claude Fable 5.2 · 64.6 Claude Fable 5.2 (Max, Default Fallback) tools unverified
- GPT-6.2 Astra · 62.1 GPT-6.2 Astra (Max)
- Gemini 4 Carbon · 60.0 Gemini 4 Carbon (High)
- Claude Opus 5.5 · 59.1 Claude Opus 5.5 (Max, Default Fallback) tools unverified
- GPT-6.2 Sol · 57.3 GPT-6.2 Sol (Max)
- Claude Sonnet 5.5 · 54.5 Claude Sonnet 5.5 (Max, Default Fallback) tools unverified
- Gemini 4 Hydrogen · 53.9 Gemini 4 Hydrogen (Xhigh)
- GPT-6.1 Terra · 51.3 GPT-6.1 Terra (Max)
- Gemini 4 Argon · 50.4 Gemini 4 Argon (High)
- Claude Haiku 5.5 · 36.5 Claude Haiku 5.5 (Max) tools unverified
- Grok 4.7 · 35.5 Grok 4.7 (Xhigh) tools unverified
- Qwen3.8 Max (0902) · 33.7 Qwen3.8 Max (0902) tools unverified
- DeepSeek V4.1 Flash · 32.6 DeepSeek V4.1 Flash 1017 (Max) tools unverified
- GPT-5.6 Terra · 32.3 GPT-5.6 Terra (Max) tools unverified
- Gemini 3.8 Flash · 30.8 Gemini 3.8 Flash (High)
- GPT-6 Luna · 27.9 GPT-6 Luna (Max)
- Mistral Large 4 · 26.1 Mistral Large 4 (High) tools unverified
- DeepSeek V4 Pro · 23.1 DeepSeek V4 Pro 0813 (Max) tools unverified
- Qwen3.8 27B · 20.1 Qwen3.8 27B (Xhigh) tools unverified
Not benchmarked: Claude Sonnet 4.6, Llama 4 Maverick, Llama 4 Scout, Qwen3-14B, Qwen3-32B.
Benchmark history
Benchmark trends need repeated, dated measurements under the same conditions.