FrontierCode
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 46 Ergebnissen; die Tabelle führt alle auf.
| # | Modell | Organisation | Ergebnis | Nachweis |
|---|---|---|---|---|
| 1 | Claude Opus 5.5 (medium) API access | Anthropic | 54,6 % | – |
| 2 | Claude Fable 5 (unknown) API access | Anthropic | 53,5 % | – |
| 3 | Claude Opus 5 (max) API access | Anthropic | 53,4 % | – |
| 4 | GPT-6 Astra (max) API access | OpenAI | 53,3 % | – |
| 5 | Claude Sonnet 5.5 (xhigh) API access | Anthropic | 52,1 % | – |
| 6 | Claude Fable 5.1 (medium) API access | Anthropic | 50,9 % | – |
| 7 | GPT-6.1 Sol (medium) API access | OpenAI | 50,2 % | – |
| 8 | SWE-2 (max) Hosted access (no API) | Cognition | 50 % | – |
| 9 | GPT-6 Sol (max) API access | OpenAI | 49,3 % | – |
| 10 | Grok 4.6 (unknown) API access | xAI | 48 % | – |
| 11 | Grok 4.7 (unknown) API access | xAI | 47,6 % | – |
| 12 | Gemini 4 Argon (medium) API access | Google DeepMind | 47,6 % | – |
| 13 | GPT-5.6 Sol (unknown thinking) API access | OpenAI | 47,5 % | – |
| 14 | Claude Opus 4.8 API access | Anthropic | 46,5 % | – |
| 15 | Claude Haiku 5.5 (max) API access | Anthropic | 46,4 % | – |
| 16 | Kimi K3 (unknown) Open weights (non-commercial) | Moonshot | 44,2 % | – |
| 17 | DeepSeek V4.1 Flash (high) Open weights (unrestricted) | DeepSeek | 43,9 % | – |
| 18 | Gemini 3.7 Flash (unknown) API access | Google DeepMind | 43,6 % | – |
| 19 | GPT-5.5 (unknown thinking) API access | OpenAI | 43 % | – |
| 20 | Claude Sonnet 5 (unknown thinking) API access | Anthropic | 42,7 % | – |
| 21 | Grok 4.5 (unknown thinking) API access | xAI | 42,4 % | – |
| 22 | GPT-6 Luna (max) API access | OpenAI | 42,4 % | – |
| 23 | SWE-1.7 | Cognition | 42 % | – |
| 24 | GPT-5.6 Terra (unknown thinking) API access | OpenAI | 41,3 % | – |
| 25 | Gemini 3.8 Flash (medium) API access | Google DeepMind | 41,2 % | – |
| 26 | GLM-5.3 (max) Open weights (restricted use) | Z.ai (Zhipu AI) | 40,1 % | – |
| 27 | GPT-5.6 Luna (unknown thinking) API access | OpenAI | 39,8 % | – |
| 28 | Mistral Large 4 Preview (high) Open weights (restricted use) | Mistral AI | 39,4 % | – |
| 29 | Claude Opus 4.7 (unknown) API access | Anthropic | 38,5 % | – |
| 30 | Gemini 3.6 Flash API access | Google DeepMind | 34,4 % | – |
| 31 | GLM-5.3-Flash (max) Open weights (unrestricted) | Z.ai (Zhipu AI) | 31,8 % | – |
| 32 | Kimi K2.7 Code Open weights (unrestricted) | Moonshot | 30,1 % | – |
| 33 | DeepSeek V4 Pro 0813 (high) Open weights (unrestricted) | DeepSeek | 28,6 % | – |
| 34 | GPT-5.4 mini (unknown thinking) API access | OpenAI | 27 % | – |
| 35 | Claude Opus 4.6 (unknown thinking) API access | Anthropic | 26,6 % | – |
| 36 | Composer 2.5 API access | Cursor | 25,6 % | – |
| 37 | GLM-5.2 Open weights (unrestricted) | Z.ai (Zhipu AI) | 24,5 % | – |
| 38 | Claude Sonnet 4.6 (unknown thinking) API access | Anthropic | 24,3 % | – |
| 39 | DeepSeek V4 Flash 0731 (unknown) Open weights (unrestricted) | DeepSeek | 18,8 % | – |
| 40 | DeepSeek v4 Pro (unknown thinking) Open weights (unrestricted) | DeepSeek | 17,6 % | – |
| 41 | MiniMax-M3 Open weights (restricted use) | MiniMax | 14,7 % | – |
| 42 | Inkling Open weights (unrestricted) | Thinking Machines | 14 % | – |
| 43 | nemotron-3-ultra Open weights (unrestricted) | Nvidia | 13,6 % | – |
| 44 | Qwen3.7 Plus API access | Alibaba | 10,2 % | – |
| 45 | SWE-1.6 Hosted access (no API) | Cognition | 9,4 % | – |
| 46 | Mistral Medium 3.5 Open weights (restricted use) | Mistral AI | 8 % | – |
