APEX-Agents
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 von 65 Ergebnissen; die Tabelle führt alle auf.
| # | Modell | Organisation | Ergebnis | Nachweis |
|---|---|---|---|---|
| 1 | Gemini 4 Carbon (unknown) API access | Google DeepMind | 84,2 % | – |
| 2 | Gemini 4 Argon (unknown) API access | Google DeepMind | 82,4 % | – |
| 3 | Claude Fable 5.2 (unknown) API access | Anthropic | 80,1 % | – |
| 4 | Claude Sonnet 5.5 (max) API access | Anthropic | 75,5 % | – |
| 5 | Claude Opus 5.5 (max) API access | Anthropic | 73,5 % | – |
| 6 | DeepSeek V4.1 Flash 1017 (max) Open weights (unrestricted) | DeepSeek | 69,8 % | – |
| 7 | Claude Fable 5.1 (unknown) API access | Anthropic | 68,6 % | – |
| 8 | GPT-6.1 Terra (max) API access | OpenAI | 68,3 % | – |
| 9 | Gemini 3.7 Flash (unknown) API access | Google DeepMind | 67,8 % | – |
| 10 | Muse Wonder (max) API access | Meta AI | 66,1 % | – |
| 11 | Claude Opus 5 (max) API access | Anthropic | 65,8 % | – |
| 12 | Kimi K3.1 Pro (unknown) Open weights (restricted use) | Moonshot | 65,5 % | – |
| 13 | Grok 4.6 (unknown) API access | xAI | 65,3 % | – |
| 14 | GPT-6 Astra (unknown thinking) API access | OpenAI | 64,7 % | – |
| 15 | Gemini 3.8 Flash (unknown) API access | Google DeepMind | 64,3 % | – |
| 16 | Muse Ember (unknown) API access | Meta AI | 64,3 % | – |
| 17 | Claude Fable 5 API access | Anthropic | 63,6 % | – |
| 18 | Qwen3.8 Max (unknown) API access | Alibaba | 63,3 % | – |
| 19 | GPT-6.1 Sol (max) API access | OpenAI | 60 % | – |
| 20 | Claude Fable 5.1 (high) API access | Anthropic | 59,7 % | – |
| 21 | Muse Spark 1.3 (unknown) API access | Meta AI | 58,6 % | – |
| 22 | GPT-5.6 Terra (max) API access | OpenAI | 58,2 % | – |
| 23 | GLM-5.3 (unknown thinking) Open weights (restricted use) | Z.ai (Zhipu AI) | 56,6 % | – |
| 24 | Grok 4.5 (unknown thinking) API access | xAI | 56,2 % | – |
| 25 | Claude Haiku 5.5 (max) API access | Anthropic | 56,1 % | – |
| 26 | GPT-5.5 (unknown thinking) API access | OpenAI | 55,1 % | – |
| 27 | Grok 4.7 (unknown) API access | xAI | 54,6 % | – |
| 28 | Claude Sonnet 5 (unknown thinking) API access | Anthropic | 54,5 % | – |
| 29 | MiMo-V2.6 Pro | Xiaomi Corp | 54,4 % | – |
| 30 | GPT-6 Sol (max) API access | OpenAI | 54,3 % | – |
| 31 | GLM-5.3-Flash (unknown) Open weights (unrestricted) | Z.ai (Zhipu AI) | 52,8 % | – |
| 32 | Claude Opus 5.5 (medium) API access | Anthropic | 52,5 % | – |
| 33 | GPT-5.4 (unknown thinking) API access | OpenAI | 52,4 % | – |
| 34 | Mistral Large 4 Preview (high) Open weights (restricted use) | Mistral AI | 51,6 % | – |
| 35 | GPT-5.6 Sol (pro, max) API access | OpenAI | 51,4 % | – |
| 36 | Claude Opus 4.7 (max) API access | Anthropic | 49,2 % | – |
| 37 | Claude Opus 4.8 API access | Anthropic | 48,9 % | – |
| 38 | Beam 1.0 (max) Open weights (unrestricted) | Reflection | 48,8 % | – |
| 39 | Kimi K3 (unknown) Open weights (non-commercial) | Moonshot | 47,8 % | – |
| 40 | Muse Spark 1.3 (max) API access | Meta AI | 47,6 % | – |
| 41 | Qwen3.8 27B Open weights (unrestricted) | Alibaba | 47,5 % | – |
| 42 | DeepSeek V4 Pro 0813 (unknown) Open weights (unrestricted) | DeepSeek | 47,3 % | – |
| 43 | Gemini 3.6 Flash API access | Google DeepMind | 46,9 % | – |
| 44 | Inkling 2 Small (unknown) | Thinking Machines | 46,8 % | – |
| 45 | Claude Opus 4.6 (max) API access | Anthropic | 46,3 % | – |
| 46 | Muse Spark 1.2 API access | Meta AI | 45,4 % | – |
| 47 | GLM-5.2 Open weights (unrestricted) | Z.ai (Zhipu AI) | 45,2 % | – |
| 48 | Claude Sonnet 5.5 (medium) API access | Anthropic | 44,6 % | – |
| 49 | GPT-6 Luna (max) API access | OpenAI | 44,3 % | – |
| 50 | Claude Sonnet 4.6 (high) API access | Anthropic | 43 % | – |
| 51 | GPT-5.6 Luna (unknown thinking) API access | OpenAI | 43 % | – |
| 52 | GLM-5.1 Open weights (unrestricted) | Z.ai (Zhipu AI) | 40,9 % | – |
| 53 | DeepSeek V4.1 Flash (unknown) Open weights (unrestricted) | DeepSeek | 39,5 % | – |
| 54 | MiniMax-M3 Open weights (restricted use) | MiniMax | 37,7 % | – |
| 55 | Kimi K2.7 Code Open weights (unrestricted) | Moonshot | 37,6 % | – |
| 56 | Gemini 3.1 Pro Preview API access | Google DeepMind | 35,1 % | – |
| 57 | Inkling Open weights (unrestricted) | Thinking Machines | 33,8 % | – |
| 58 | Muse Spark Open (unknown) Open weights (restricted use) | Meta AI | 33,4 % | – |
| 59 | Muse Spark 1.1 API access | Meta AI | 31,8 % | – |
| 60 | Gemini 3.5 Flash-Lite API access | Google DeepMind | 29,3 % | – |
| 61 | Gemini 3.5 Flash API access | Google DeepMind | 27,5 % | – |
| 62 | Qwen3.5 397B-A17B Open weights (unrestricted) | Alibaba | 24,9 % | – |
| 63 | nemotron-3-ultra Open weights (unrestricted) | Nvidia | 22,7 % | – |
| 64 | DeepSeek-V3.2 (Thinking; Novita) Open weights (unrestricted) | DeepSeek | 21,3 % | – |
| 65 | gpt-oss-120b Open weights (unrestricted) | OpenAI | 4,4 % | – |
