FrontierMath-2025-02-28-Private
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 101 Ergebnissen.
| # | Modell | Organisation | Ergebnis | Nachweis |
|---|---|---|---|---|
| 1 | GPT-5.5 Pro (high) API access | OpenAI | 52,4 % ± 2,9 % | – |
| 2 | GPT-5.5 (xhigh) API access | OpenAI | 51,7 % ± 2,9 % | – |
| 3 | GPT-5.5 Pro (xhigh) API access | OpenAI | 51 % ± 2,9 % | – |
| 4 | GPT-5.4 Pro (xhigh) API access | OpenAI | 50 % ± 2,9 % | – |
| 5 | GPT-5.4 (xhigh) API access | OpenAI | 47,6 % ± 2,9 % | – |
| 6 | Claude Opus 4.8 API access | Anthropic | 47,2 % ± 2,9 % | – |
| 7 | Claude Opus 4.7 (xhigh) API access | Anthropic | 43,8 % ± 2,9 % | – |
| 8 | Claude Opus 4.6 (max) API access | Anthropic | 40,7 % ± 2,9 % | – |
| 9 | GPT-5.2 (xhigh) API access | OpenAI | 40,7 % ± 2,9 % | – |
| 10 | GPT-5.2 (high) API access | OpenAI | 40,3 % ± 2,9 % | – |
| 11 | Claude Opus 4.6 (32k thinking) API access | Anthropic | 40 % ± 2,9 % | – |
| 12 | Claude Opus 4.6 (64k thinking) API access | Anthropic | 39,7 % ± 2,9 % | – |
| 13 | Muse Spark API access | Meta AI | 39 % ± 2,9 % | – |
| 14 | Gemini 3.5 Flash API access | Google DeepMind | 39 % ± 2,9 % | – |
| 15 | Kimi K2.6 Open weights (unrestricted) | Moonshot | 39 % ± 2,9 % | – |
| 16 | Claude Opus 4.6 (no thinking) API access | Anthropic | 38,3 % ± 2,9 % | – |
| 17 | Gemini 3 Pro Preview API access | Google DeepMind | 37,6 % ± 2,8 % | – |
| 18 | Gemini 3.1 Pro Preview API access | Google DeepMind | 36,9 % ± 2,8 % | – |
| 19 | GPT-5.2 (medium) API access | OpenAI | 36,9 % ± 2,8 % | – |
| 20 | Gemini 3 Flash Preview API access | Google DeepMind | 35,6 % ± 2,8 % | – |
| 21 | GLM-5.1 Open weights (unrestricted) | Z.ai (Zhipu AI) | 33,4 % ± 2,8 % | – |
| 22 | GPT-5 (high) API access | OpenAI | 32,4 % ± 2,8 % | – |
| 23 | Claude Sonnet 4.6 (16k thinking) API access | Anthropic | 32,4 % ± 2,8 % | – |
| 24 | GPT-5.1 (high) API access | OpenAI | 31 % ± 2,7 % | – |
| 25 | gemini-2.5-deep-think-2025-08-01-webapp Hosted access (no API) | Google,Google DeepMind | 29 % ± 2,7 % | – |
| 26 | GPT-5.4 mini (high) API access | OpenAI | 28,3 % ± 2,1 % | – |
| 27 | Kimi K2.5 (Fireworks) Open weights (unrestricted) | Moonshot | 27,9 % ± 2,6 % | – |
| 28 | GPT-5 (medium) API access | OpenAI | 27,2 % ± 2,6 % | – |
| 29 | GPT-5 mini (high) API access | OpenAI | 27,2 % ± 2,6 % | – |
| 30 | GPT-5.1 (medium) API access | OpenAI | 26,9 % ± 2,6 % | – |
| 31 | GPT-5.2 (low) API access | OpenAI | 26,6 % ± 2,6 % | – |
| 32 | Qwen 3.6 Plus (2026-04-02) API access | Alibaba | 26,2 % ± 2,6 % | – |
| 33 | GPT-5.4 nano (high) API access | OpenAI | 25,9 % ± 2,6 % | – |
| 34 | o4-mini (high) API access | OpenAI | 24,8 % ± 2,5 % | – |
| 35 | Qwen 3.6 Max (Preview) | Alibaba | 23,1 % ± 2,5 % | – |
| 36 | DeepSeek-V3.2 (Thinking; Fireworks) Open weights (unrestricted) | DeepSeek | 22,1 % ± 4,1 % | – |
| 37 | Kimi K2 Thinking (Together) Open weights (restricted use) | Moonshot | 21,4 % ± 2,4 % | – |
| 38 | Qwen3.5 Plus | Alibaba | 21 % ± 2,4 % | – |
| 39 | Claude Opus 4.5 (no thinking) API access | Anthropic | 20,7 % ± 2,4 % | – |
| 40 | Claude Opus 4.5 (32k thinking) API access | Anthropic | 20,7 % ± 2,4 % | – |
| 41 | Claude Opus 4.5 (16k thinking) API access | Anthropic | 20,3 % ± 2,4 % | – |
| 42 | GPT-5 mini (medium) API access | OpenAI | 20,3 % ± 2,4 % | – |
| 43 | grok-4-0709 API access | xAI | 19,7 % ± 2,3 % | – |
| 44 | o4-mini (medium) API access | OpenAI | 19 % ± 2,3 % | – |
| 45 | o3 (high) API access | OpenAI | 18,7 % ± 2,3 % | – |
| 46 | GPT-5.1 (low) API access | OpenAI | 17,3 % ± 2,2 % | – |
| 47 | o3 (medium) API access | OpenAI | 16,9 % ± 2,2 % | – |
| 48 | GLM-5 Open weights (unrestricted) | Z.ai (Zhipu AI) | 16,4 % ± 2,2 % | – |
| 49 | Claude Sonnet 4.5 (32k thinking) API access | Anthropic | 15,2 % ± 2,1 % | – |
| 50 | gemini-2.5-pro API access | Google DeepMind | 14,1 % ± 2 % | – |
| 51 | Claude Sonnet 4.5 (59k thinking) API access | Anthropic | 13,5 % ± 2 % | – |
| 52 | o3-mini (high) API access | OpenAI | 12,4 % ± 1,9 % | – |
| 53 | o4-mini (low) API access | OpenAI | 10,7 % ± 1,8 % | – |
| 54 | Gemini 2.5 Pro Preview (Jun 2025) API access | Google DeepMind | 10,3 % ± 1,8 % | – |
| 55 | Qwen 3.6 Flash (2026-04-16) | Alibaba | 10,3 % ± 1,8 % | – |
| 56 | o3 (low) API access | OpenAI | 9,7 % ± 1,7 % | – |
| 57 | Claude Sonnet 4.5 (no thinking) API access | Anthropic | 9,3 % ± 1,7 % | – |
| 58 | o1 (high) API access | OpenAI | 9,3 % ± 1,7 % | – |
| 59 | Qwen3-235B-A22B-Thinking-2507 Open weights (unrestricted) | Alibaba | 8,5 % ± 1,7 % | – |
| 60 | GPT-5 nano (high) API access | OpenAI | 8,3 % ± 1,6 % | – |
| 61 | o3-mini (medium) API access | OpenAI | 8,1 % ± 1,3 % | – |
| 62 | claude-opus-4-1-20250805_27K API access | Anthropic | 7,2 % ± 1,5 % | – |
| 63 | GPT-5 nano (medium) API access | OpenAI | 7,2 % ± 1,5 % | – |
| 64 | Qwen3.5 Flash | Alibaba | 6,2 % ± 1,4 % | – |
| 65 | claude-haiku-4-5-20251001_32K API access | Anthropic | 5,9 % ± 1,4 % | – |
| 66 | claude-opus-4-1-20250805 API access | Anthropic | 5,9 % ± 1,4 % | – |
| 67 | grok-3-mini-beta_high API access | xAI | 5,9 % ± 1,4 % | – |
| 68 | GPT-4.1 API access | OpenAI | 5,5 % ± 1,3 % | – |
| 69 | gemini-2.5-flash API access | Google DeepMind | 4,8 % ± 1,3 % | – |
| 70 | Claude Opus 4 API access | Anthropic | 4,5 % ± 1,2 % | – |
| 71 | GPT-4.1 mini API access | OpenAI | 4,5 % ± 1,2 % | – |
| 72 | Claude 3.7 Sonnet (16k thinking) API access | Anthropic | 4,1 % ± 1,2 % | – |
| 73 | claude-haiku-4-5-20251001 API access | Anthropic | 4,1 % ± 1,2 % | – |
| 74 | Claude Opus 4 API access | Anthropic | 4,1 % ± 1,2 % | – |
| 75 | claude-sonnet-4-20250514 API access | Anthropic | 4,1 % ± 1,2 % | – |
| 76 | GLM-4.6 (Together) Open weights (unrestricted) | Z.ai (Zhipu AI),Tsinghua University | 3,8 % ± 1,1 % | – |
| 77 | grok-3-beta API access | xAI | 3,8 % ± 1,1 % | – |
| 78 | Claude 3.7 Sonnet (32k thinking) API access | Anthropic | 3,4 % ± 1,1 % | – |
| 79 | claude-3-7-sonnet-20250219 API access | Anthropic | 3,1 % ± 1 % | – |
| 80 | Claude 3.7 Sonnet (64k thinking) API access | Anthropic | 3,1 % ± 1 % | – |
| 81 | grok-3-mini-beta_low API access | xAI | 2,8 % ± 1 % | – |
| 82 | GLM-4.7 (Together) Open weights (unrestricted) | Z.ai (Zhipu AI) | 2,4 % ± 0,9 % | – |
| 83 | GPT-5.1 (no thinking) API access | OpenAI | 2,1 % ± 0,8 % | – |
| 84 | Claude 3.5 Sonnet (Oct 2024) API access | Anthropic | 2,1 % ± 0,8 % | – |
| 85 | DeepSeek-V3 Open weights (restricted use) | DeepSeek | 1,7 % ± 0,8 % | – |
| 86 | Gemini 2.0 Flash (Feb 2025) API access | Google DeepMind,Google | 1,7 % ± 0,8 % | – |
| 87 | o1-mini (medium) API access | OpenAI | 1,7 % ± 0,8 % | – |
| 88 | qwen-plus-2025-04-28 API access | Alibaba | 1,7 % ± 0,8 % | – |
| 89 | o1-mini (high) API access | OpenAI | 1,4 % ± 0,7 % | – |
| 90 | Claude 3.5 Sonnet (Jun 2024) API access | Anthropic | 1 % ± 0,6 % | – |
| 91 | GPT-4.1 nano API access | OpenAI | 1 % ± 0,6 % | – |
| 92 | qwen-max-2025-01-25 API access | Alibaba | 1 % ± 0,6 % | – |
| 93 | Llama 4 Maverick (FP8) Open weights (restricted use) | Meta AI | 0,7 % ± 0,5 % | – |
| 94 | grok-2-1212 API access | xAI | 0,7 % ± 0,5 % | – |
| 95 | mistral-medium-2505 API access | Mistral AI | 0,3 % ± 0,3 % | – |
| 96 | Claude 3.5 Haiku (Oct 2024) API access | Anthropic | 0,3 % ± 0,3 % | – |
| 97 | GPT-4o (Aug 2024) API access | OpenAI | 0,3 % ± 0,3 % | – |
| 98 | GPT-4o (Nov 2024) API access | OpenAI | 0,3 % ± 0,3 % | – |
| 99 | mistral-large-2411 Open weights (non-commercial) | Mistral AI | 0,3 % ± 0,3 % | – |
| 100 | Llama-4-Scout-17B-16E-Instruct Open weights (restricted use) | Meta AI | 0 % ± 0 % | – |
