WeirdML
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 172 Ergebnissen.
| # | Modell | Organisation | Ergebnis | Nachweis |
|---|---|---|---|---|
| 1 | GPT-6 Astra (pro, max) API access | OpenAI | 93,6 % | – |
| 2 | GPT-6 Astra (max) API access | OpenAI | 93,3 % | – |
| 3 | Claude Fable 5.1 (max) API access | Anthropic | 92,9 % | – |
| 4 | GPT-6 Astra (high) API access | OpenAI | 92,9 % | – |
| 5 | Claude Fable 5.1 (high) API access | Anthropic | 92,3 % | – |
| 6 | Claude Fable 5 (max) API access | Anthropic | 91,9 % | – |
| 7 | Claude Opus 5 (max) API access | Anthropic | 91,8 % | – |
| 8 | Claude Opus 5 (high) API access | Anthropic | 91,6 % | – |
| 9 | GPT-5.6 Sol (pro, max) API access | OpenAI | 89,4 % | – |
| 10 | GPT-5.6 Sol (high) API access | OpenAI | 88,8 % | – |
| 11 | Claude Fable 5 (high) API access | Anthropic | 87,9 % | – |
| 12 | GPT-5.6 Sol (max) API access | OpenAI | 87 % | – |
| 13 | Claude Opus 5 API access | Anthropic | 86,3 % | – |
| 14 | GPT-5.5 (xhigh) API access | OpenAI | 84,9 % | – |
| 15 | Gemini 3.8 Flash (high) API access | Google DeepMind | 84,8 % | – |
| 16 | GPT-5.5 (high) API access | OpenAI | 83,9 % | – |
| 17 | Claude Opus 4.8 API access | Anthropic | 82,9 % | – |
| 18 | Kimi K3 (max) Open weights (non-commercial) | Moonshot | 82,6 % | – |
| 19 | GPT-5.3 Codex API access | OpenAI | 79,3 % | – |
| 20 | GPT-5.6 Terra (high) API access | OpenAI | 78,3 % | – |
| 21 | Claude Opus 4.6 (high) API access | Anthropic | 78 % | – |
| 22 | Claude Opus 4.6 (unknown thinking) API access | Anthropic | 77,9 % | – |
| 23 | GPT-5.3 Codex (xhigh) API access | OpenAI | 77,9 % | – |
| 24 | GPT-5.4 (xhigh) API access | OpenAI | 77,7 % | Protokoll oder Quelle |
| 25 | Claude Opus 4.7 (high) API access | Anthropic | 76,4 % | – |
| 26 | Claude Opus 4.7 (no thinking) API access | Anthropic | 76,4 % | – |
| 27 | Claude Opus 4.7 (unknown) API access | Anthropic | 76,4 % | – |
| 28 | Claude Opus 4.8 API access | Anthropic | 76 % | – |
| 29 | Claude Opus 4.7 (max) API access | Anthropic | 75,5 % | – |
| 30 | GLM-5.3 (max) Open weights (restricted use) | Z.ai (Zhipu AI) | 75,4 % | – |
| 31 | GPT-5.2 (xhigh) API access | OpenAI | 72,2 % | Protokoll oder Quelle |
| 32 | Gemini 3.1 Pro Preview API access | Google DeepMind | 72,1 % | – |
| 33 | Claude Opus 4.8 API access | Anthropic | 70,5 % | – |
| 34 | GLM-5.2 Open weights (unrestricted) | Z.ai (Zhipu AI) | 70,1 % | – |
| 35 | Gemini 3 Pro Preview API access | Google DeepMind | 69,9 % | Protokoll oder Quelle |
| 36 | Claude Sonnet 5 (high) API access | Anthropic | 68,8 % | – |
| 37 | GLM-5.2 Open weights (unrestricted) | Z.ai (Zhipu AI) | 67,3 % | – |
| 38 | Grok 4.6 (high) API access | xAI | 67,3 % | – |
| 39 | GPT-5.5 (no thinking) API access | OpenAI | 67,2 % | – |
| 40 | DeepSeek V4 Pro 0813 (max) Open weights (unrestricted) | DeepSeek | 66,2 % | – |
| 41 | Claude Sonnet 4.6 (medium) API access | Anthropic | 66,1 % | – |
| 42 | Claude Opus 4.6 (no thinking) API access | Anthropic | 65,9 % | Protokoll oder Quelle |
| 43 | Claude Opus 4.5 (16k thinking) API access | Anthropic | 63,7 % | Protokoll oder Quelle |
| 44 | GPT-5.2 (medium) API access | OpenAI | 63,4 % | Protokoll oder Quelle |
| 45 | DeepSeek V4 Flash 0731 (max) Open weights (unrestricted) | DeepSeek | 63 % | – |
| 46 | Gemini 3.5 Flash API access | Google DeepMind | 62,6 % | – |
| 47 | Gemini 3 Flash Preview API access | Google DeepMind | 61,6 % | Protokoll oder Quelle |
| 48 | GPT-5.6 Luna (high) API access | OpenAI | 60,9 % | – |
| 49 | GPT-5.1 (high) API access | OpenAI | 60,8 % | Protokoll oder Quelle |
| 50 | GPT-5 (high) API access | OpenAI | 60,7 % | Protokoll oder Quelle |
| 51 | GPT-5 Pro API access | OpenAI | 60,4 % | Protokoll oder Quelle |
| 52 | GPT-5.4 mini (high) API access | OpenAI | 60,3 % | – |
| 53 | Muse Spark 1.2 (xhigh) API access | Meta AI | 60,3 % | – |
| 54 | o3-pro-2025-06-10_high API access | OpenAI | 58,2 % | Protokoll oder Quelle |
| 55 | GPT-5.4 (none) API access | OpenAI | 57,4 % | – |
| 56 | GPT-5.4 Pro (no thinking) API access | OpenAI | 57,4 % | – |
| 57 | GLM-5.1 Open weights (unrestricted) | Z.ai (Zhipu AI) | 57,1 % | – |
| 58 | DeepSeek V4 Flash 0731 (high) Open weights (unrestricted) | DeepSeek | 57,1 % | – |
| 59 | Gemini 3.6 Flash API access | Google DeepMind | 56,1 % | – |
| 60 | Kimi K2.6 Open weights (unrestricted) | Moonshot | 55,9 % | – |
| 61 | GPT-5-codex API access | OpenAI | 54,5 % | Protokoll oder Quelle |
| 62 | GPT-5-codex (high) API access | OpenAI | 54,5 % | – |
| 63 | Kimi K2.7 Code Open weights (unrestricted) | Moonshot | 54,1 % | – |
| 64 | gemini-2.5-pro_16K API access | Google DeepMind | 54 % | Protokoll oder Quelle |
| 65 | GPT-5 mini (high) API access | OpenAI | 52,7 % | Protokoll oder Quelle |
| 66 | o4-mini (high) API access | OpenAI | 52,6 % | Protokoll oder Quelle |
| 67 | o3 (high) API access | OpenAI | 52,4 % | Protokoll oder Quelle |
| 68 | Gemma 4 31B IT Open weights (restricted use) | Google DeepMind | 52,3 % | – |
| 69 | grok-4-20 API access | xAI | 52,3 % | – |
| 70 | Gemini 3.1 Flash-Lite API access | 52,2 % | – | |
| 71 | grok-4.3_unknown Hosted access (no API) | xAI | 49,9 % | – |
| 72 | GPT-5.2 (none) API access | OpenAI | 49,6 % | – |
| 73 | GPT-5.2 (low) API access | OpenAI | 49,6 % | Protokoll oder Quelle |
| 74 | GPT-5.4 nano (high) API access | OpenAI | 49,2 % | – |
| 75 | DeepSeek v4 Pro (max) Open weights (unrestricted) | DeepSeek | 48,9 % | – |
| 76 | GLM-5 Open weights (unrestricted) | Z.ai (Zhipu AI) | 48,2 % | – |
| 77 | gpt-oss-120b_high Open weights (unrestricted) | OpenAI | 48,2 % | Protokoll oder Quelle |
| 78 | openai/gpt-oss-120b_high Open weights (unrestricted) | OpenAI | 48,2 % | – |
| 79 | Claude Sonnet 4.5 (16k thinking) API access | Anthropic | 47,7 % | Protokoll oder Quelle |
| 80 | o1-preview API access | OpenAI | 47,6 % | Protokoll oder Quelle |
| 81 | DeepSeek-V3.2-Speciale Open weights (unrestricted) | DeepSeek | 46,7 % | – |
| 82 | Claude Sonnet 4.5 (no thinking) API access | Anthropic | 46,7 % | Protokoll oder Quelle |
| 83 | DeepSeek v4 Pro (high) Open weights (unrestricted) | DeepSeek | 46,5 % | – |
| 84 | Grok 4.5 (unknown thinking) API access | xAI | 46,4 % | – |
| 85 | claude-sonnet-4-20250514_16K API access | Anthropic | 46,1 % | Protokoll oder Quelle |
| 86 | o1 (high) API access | OpenAI | 46,1 % | Protokoll oder Quelle |
| 87 | claude-opus-4-1-20250805_16K API access | Anthropic | 45,9 % | Protokoll oder Quelle |
| 88 | grok-4-0709 API access | xAI | 45,7 % | Protokoll oder Quelle |
| 89 | DeepSeek v4 Flash (max) Open weights (unrestricted) | DeepSeek | 45,6 % | – |
| 90 | Kimi K2.5 Open weights (unrestricted) | Moonshot | 45,6 % | Protokoll oder Quelle |
| 91 | claude-haiku-4-5-20251001 API access | Anthropic | 45,4 % | Protokoll oder Quelle |
| 92 | claude-haiku-4-5-20251001_16K API access | Anthropic | 44,2 % | Protokoll oder Quelle |
| 93 | claude-sonnet-4-20250514 API access | Anthropic | 43,9 % | Protokoll oder Quelle |
| 94 | DeepSeek v4 Flash (high) Open weights (unrestricted) | DeepSeek | 43,8 % | – |
| 95 | Claude Opus 4 API access | Anthropic | 43,7 % | Protokoll oder Quelle |
| 96 | Mistral Medium 3.5 Open weights (restricted use) | Mistral AI | 43,7 % | – |
| 97 | o3-mini (high) API access | OpenAI | 43,7 % | Protokoll oder Quelle |
| 98 | nemotron-3-ultra Open weights (unrestricted) | Nvidia | 43,5 % | – |
| 99 | Mercury 2 API access | Inception Labs | 43,2 % | – |
| 100 | grok-4-fast API access | xAI | 42,9 % | Protokoll oder Quelle |
