ARC-AGI
Das Ergebnis jedes Modells, so, wie der Benchmarking Hub von Epoch AI es veröffentlicht, von Epoch gemessen oder aus der genannten Quelle gesammelt. Facthory hat sie nicht selbst gemessen.
Daten: Epoch AI, „Capabilities & benchmarking“, epoch.ai/benchmarks, lizenziert unter CC BY 4.0. Gelesen am 16.11.2026.
Das Diagramm zeigt die 25 besten, die Tabelle die 100 besten von 287 Ergebnissen.
| # | Modell | Organisation | Ergebnis | Nachweis |
|---|---|---|---|---|
| 1 | Claude Fable 5.2 (xhigh) API access | Anthropic | 98,5 % | – |
| 2 | Claude Fable 5 (max) API access | Anthropic | 98,5 % | – |
| 3 | Claude Fable 5 (xhigh) API access | Anthropic | 98,5 % | – |
| 4 | Claude Opus 5.5 (high) API access | Anthropic | 98,5 % | – |
| 5 | Gemini 3.8 Flash (high) API access | Google DeepMind | 98,5 % | – |
| 6 | Gemini 4 Argon (high) API access | Google DeepMind | 98,5 % | – |
| 7 | Gemini 4 Argon (medium) API access | Google DeepMind | 98,5 % | – |
| 8 | Gemini 4 Carbon (medium) API access | Google DeepMind | 98,5 % | – |
| 9 | GPT-6 Astra (high) API access | OpenAI | 98,5 % | – |
| 10 | GPT-6 Astra (xhigh) API access | OpenAI | 98,5 % | – |
| 11 | GPT-6.1 Sol (high) API access | OpenAI | 98,5 % | – |
| 12 | GPT-6.1 Sol (xhigh) API access | OpenAI | 98,5 % | – |
| 13 | Gemini 3.1 Pro Preview API access | Google DeepMind | 98 % | Protokoll oder Quelle |
| 14 | Claude Fable 5.1 (max) API access | Anthropic | 97,5 % | – |
| 15 | Claude Fable 5.2 (high) API access | Anthropic | 97,5 % | – |
| 16 | Claude Fable 5.2 (max) API access | Anthropic | 97,5 % | – |
| 17 | Claude Fable 5.2 (medium) API access | Anthropic | 97,5 % | – |
| 18 | Claude Opus 5.5 (max) API access | Anthropic | 97,5 % | – |
| 19 | Claude Opus 5.5 (medium) API access | Anthropic | 97,5 % | – |
| 20 | Claude Opus 5.5 (xhigh) API access | Anthropic | 97,5 % | – |
| 21 | Claude Opus 5 (high) API access | Anthropic | 97,5 % | – |
| 22 | Claude Opus 5 (max) API access | Anthropic | 97,5 % | – |
| 23 | Gemini 3.8 Flash (medium) API access | Google DeepMind | 97,5 % | – |
| 24 | Gemini 4 Carbon (high) API access | Google DeepMind | 97,5 % | – |
| 25 | GPT-5.6 Sol (xhigh) API access | OpenAI | 97,5 % | – |
| 26 | GPT-6 Astra (max) API access | OpenAI | 97,5 % | – |
| 27 | GPT-6 Astra (medium) API access | OpenAI | 97,5 % | – |
| 28 | GPT-5.6 Sol (high) API access | OpenAI | 97 % | – |
| 29 | Claude Fable 5.1 (xhigh) API access | Anthropic | 96,5 % | – |
| 30 | Claude Sonnet 5.5 (max) API access | Anthropic | 96,5 % | – |
| 31 | Claude Sonnet 5.5 (xhigh) API access | Anthropic | 96,5 % | – |
| 32 | GPT-5.5 Pro (high) API access | OpenAI | 96,5 % | – |
| 33 | GPT-5.6 Sol (max) API access | OpenAI | 96,5 % | – |
| 34 | GPT-5.6 Terra (max) API access | OpenAI | 96,5 % | – |
| 35 | GPT-6 Astra (low) API access | OpenAI | 96,5 % | – |
| 36 | GPT-6.1 Sol (max) API access | OpenAI | 96,5 % | – |
| 37 | Inkling 2 Small (xhigh) | Thinking Machines | 96,5 % | – |
| 38 | Claude Fable 5.1 (high) API access | Anthropic | 96 % | – |
| 39 | Gemini 3 Deep Think Hosted access (no API) | Google DeepMind | 96 % | – |
| 40 | Claude Fable 5 (high) API access | Anthropic | 95,5 % | – |
| 41 | Gemini 3.7 Flash (high) API access | Google DeepMind | 95,5 % | – |
| 42 | GPT-6 Sol (max) API access | OpenAI | 95,5 % | – |
| 43 | GPT-6.1 Sol (medium) API access | OpenAI | 95,5 % | – |
| 44 | GPT-5.5 Pro (xhigh) API access | OpenAI | 95 % | – |
| 45 | GPT-5.5 (xhigh) API access | OpenAI | 95 % | – |
| 46 | Claude Fable 5.1 (medium) API access | Anthropic | 94,5 % | – |
| 47 | Claude Fable 5.2 (low) API access | Anthropic | 94,5 % | – |
| 48 | DeepSeek V4.1 Flash 1017 (high) Open weights (unrestricted) | DeepSeek | 94,5 % | – |
| 49 | DeepSeek V4.1 Flash (max) Open weights (unrestricted) | DeepSeek | 94,5 % | – |
| 50 | GPT-5.4 Pro (xhigh) API access | OpenAI | 94,5 % | – |
| 51 | GPT-5.5 (high) API access | OpenAI | 94,5 % | – |
| 52 | Kimi K3 (max) Open weights (non-commercial) | Moonshot | 94,5 % | – |
| 53 | Claude Opus 4.6 (120k thinking) API access | Anthropic | 94 % | – |
| 54 | GPT-5.6 Terra (xhigh) API access | OpenAI | 94 % | – |
| 55 | GPT-5.4 (xhigh) API access | OpenAI | 93,7 % | – |
| 56 | Claude Opus 4.7 (high) API access | Anthropic | 93,5 % | – |
| 57 | DeepSeek V4.1 Flash 1017 (max) Open weights (unrestricted) | DeepSeek | 93,5 % | – |
| 58 | GPT-6.1 Sol (low) API access | OpenAI | 93,5 % | – |
| 59 | GPT-5.4 (high) API access | OpenAI | 92,7 % | – |
| 60 | GPT-6 Sol (xhigh) API access | OpenAI | 92,7 % | – |
| 61 | Claude Fable 5 (medium) API access | Anthropic | 92,5 % | – |
| 62 | Claude Opus 4.8 API access | Anthropic | 92,5 % | – |
| 63 | Gemini 3.5 Flash API access | Google DeepMind | 92,5 % | – |
| 64 | GPT-5.6 Sol (medium) API access | OpenAI | 92,5 % | – |
| 65 | GPT-5.5 (medium) API access | OpenAI | 92,2 % | – |
| 66 | Claude Opus 4.7 (max) API access | Anthropic | 92 % | – |
| 67 | Claude Opus 4.8 API access | Anthropic | 92 % | – |
| 68 | Claude Sonnet 5.5 (high) API access | Anthropic | 92 % | – |
| 69 | GPT-5.6 Terra (high) API access | OpenAI | 92 % | – |
| 70 | Claude Opus 4.8 API access | Anthropic | 91,5 % | – |
| 71 | DeepSeek V4.1 Flash 1017 (low) Open weights (unrestricted) | DeepSeek | 91,5 % | – |
| 72 | Gemini 3.6 Flash API access | Google DeepMind | 91,2 % | – |
| 73 | Gemini 3.7 Flash (medium) API access | Google DeepMind | 91,2 % | – |
| 74 | Claude Opus 4.7 (low) API access | Anthropic | 91 % | – |
| 75 | Claude Opus 4.7 (medium) API access | Anthropic | 91 % | – |
| 76 | GLM-5.3-Flash (max) Open weights (unrestricted) | Z.ai (Zhipu AI) | 91 % | – |
| 77 | GPT-6 Sol (high) API access | OpenAI | 91 % | – |
| 78 | Claude Fable 5 (low) API access | Anthropic | 90,5 % | – |
| 79 | DeepSeek V4 Pro 0813 (low) Open weights (unrestricted) | DeepSeek | 90,5 % | – |
| 80 | DeepSeek V4.1 Flash (low) Open weights (unrestricted) | DeepSeek | 90,5 % | – |
| 81 | Gemini 3.8 Flash (low) API access | Google DeepMind | 90,5 % | – |
| 82 | GPT-5.2 Pro API access | OpenAI | 90,5 % | – |
| 83 | Grok 4.7 (high) API access | xAI | 90,2 % | – |
| 84 | Claude Fable 5.1 (low) API access | Anthropic | 90 % | – |
| 85 | DeepSeek V4 Pro 0813 (max) Open weights (unrestricted) | DeepSeek | 90 % | – |
| 86 | Grok 4.7 (medium) API access | xAI | 90 % | – |
| 87 | grok-4-20 API access | xAI | 89,5 % | – |
| 88 | Grok 4.7 (xhigh) API access | xAI | 89,5 % | – |
| 89 | Muse Spark Open (xhigh) Open weights (restricted use) | Meta AI | 89 % | – |
| 90 | Claude Haiku 5.5 (max) API access | Anthropic | 89 % | – |
| 91 | DeepSeek V4 Flash 0731 (max) Open weights (unrestricted) | DeepSeek | 89 % | – |
| 92 | Muse Spark Open (high) Open weights (restricted use) | Meta AI | 88,5 % | – |
| 93 | Claude Opus 5.5 (low) API access | Anthropic | 88,5 % | – |
| 94 | DeepSeek V4.1 Flash (high) Open weights (unrestricted) | DeepSeek | 88,5 % | – |
| 95 | Claude Opus 4.8 API access | Anthropic | 88 % | – |
| 96 | GPT-5.6 Luna (max) API access | OpenAI | 88 % | – |
| 97 | GPT-5.6 Luna (xhigh) API access | OpenAI | 87,7 % | – |
| 98 | Claude Haiku 5.5 (xhigh) API access | Anthropic | 87,5 % | – |
| 99 | Grok 4.6 (medium) API access | xAI | 87,5 % | – |
| 100 | Qwen3.8 27B (xhigh) Open weights (unrestricted) | Alibaba | 87,5 % | – |
