FrontierMath-Tier-4-2025-07-01-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 von 72 Ergebnissen; die Tabelle führt alle auf.
| # | Modell | Organisation | Ergebnis | Nachweis |
|---|---|---|---|---|
| 1 | AI co-mathematician Unreleased | Google DeepMind | 47,9 % ± 7,2 % | – |
| 2 | GPT-5.5 Pro (high) API access | OpenAI | 39,6 % ± 7,1 % | – |
| 3 | GPT-5.5 Pro (xhigh) API access | OpenAI | 39,6 % ± 7,1 % | – |
| 4 | GPT-5.4 Pro (web) API access | OpenAI | 37,5 % ± 7 % | – |
| 5 | GPT-5.5 (xhigh) API access | OpenAI | 35,4 % ± 6,9 % | – |
| 6 | GPT-5.2 Pro (web) API access | OpenAI | 31,3 % ± 6,7 % | – |
| 7 | Claude Opus 4.8 API access | Anthropic | 31,3 % ± 6,8 % | – |
| 8 | GPT-5.4 (xhigh) API access | OpenAI | 27,1 % ± 6,4 % | – |
| 9 | Claude Opus 4.7 (xhigh) API access | Anthropic | 22,9 % ± 6,1 % | – |
| 10 | Claude Opus 4.6 (max) API access | Anthropic | 22,9 % ± 6,1 % | – |
| 11 | Claude Opus 4.6 (32k thinking) API access | Anthropic | 20,8 % ± 5,9 % | – |
| 12 | Claude Opus 4.6 (64k thinking) API access | Anthropic | 20,8 % ± 5,9 % | – |
| 13 | GPT-5.2 (high) API access | OpenAI | 18,8 % ± 5,6 % | – |
| 14 | GPT-5.2 (xhigh) API access | OpenAI | 18,8 % ± 5,6 % | – |
| 15 | Gemini 3 Pro Preview API access | Google DeepMind | 18,8 % ± 5,7 % | – |
| 16 | Gemini 3.1 Pro Preview API access | Google DeepMind | 16,7 % ± 5,4 % | – |
| 17 | GPT-5.2 (medium) API access | OpenAI | 16,7 % ± 5,4 % | – |
| 18 | GPT-5 Pro API access | OpenAI | 14,6 % ± 5,1 % | – |
| 19 | Muse Spark API access | Meta AI | 14,6 % ± 5,1 % | – |
| 20 | Claude Opus 4.6 (no thinking) API access | Anthropic | 14,6 % ± 5,1 % | – |
| 21 | Gemini 3.5 Flash API access | Google DeepMind | 14,6 % ± 5,1 % | – |
| 22 | Kimi K2.6 Open weights (unrestricted) | Moonshot | 14,6 % ± 5,2 % | – |
| 23 | GLM-5.1 Open weights (unrestricted) | Z.ai (Zhipu AI) | 12,5 % ± 4,8 % | – |
| 24 | GPT-5 (high) API access | OpenAI | 12,5 % ± 4,8 % | – |
| 25 | GPT-5.1 (high) API access | OpenAI | 12,5 % ± 4,8 % | – |
| 26 | gemini-2.5-deep-think-2025-08-01-webapp Hosted access (no API) | Google,Google DeepMind | 10,4 % ± 4,4 % | – |
| 27 | Qwen 3.6 Plus (2026-04-02) API access | Alibaba | 8,3 % ± 4 % | – |
| 28 | Claude Sonnet 4.6 (16k thinking) API access | Anthropic | 8,3 % ± 4 % | – |
| 29 | GPT-5 (medium) API access | OpenAI | 6,3 % ± 3,5 % | – |
| 30 | GPT-5 mini (high) API access | OpenAI | 6,3 % ± 3,5 % | – |
| 31 | GPT-5.2 (low) API access | OpenAI | 6,3 % ± 3,5 % | – |
| 32 | GPT-5.4 nano (high) API access | OpenAI | 6,3 % ± 3,5 % | – |
| 33 | o4-mini (high) API access | OpenAI | 6,3 % ± 3,5 % | – |
| 34 | Kimi K2.5 (Fireworks) Open weights (unrestricted) | Moonshot | 4,2 % ± 2,9 % | – |
| 35 | claude-opus-4-1-20250805_27K API access | Anthropic | 4,2 % ± 2,9 % | – |
| 36 | Claude Opus 4 API access | Anthropic | 4,2 % ± 2,9 % | – |
| 37 | Claude Opus 4.5 (no thinking) API access | Anthropic | 4,2 % ± 2,9 % | – |
| 38 | Claude Opus 4.5 (32k thinking) API access | Anthropic | 4,2 % ± 2,9 % | – |
| 39 | Claude Sonnet 4.5 (32k thinking) API access | Anthropic | 4,2 % ± 2,9 % | – |
| 40 | gemini-2.5-flash API access | Google DeepMind | 4,2 % ± 2,9 % | – |
| 41 | gemini-2.5-pro API access | Google DeepMind | 4,2 % ± 2,9 % | – |
| 42 | Gemini 3 Flash Preview API access | Google DeepMind | 4,2 % ± 2,9 % | – |
| 43 | GPT-5 mini (medium) API access | OpenAI | 4,2 % ± 2,9 % | – |
| 44 | GPT-5.1 (medium) API access | OpenAI | 4,2 % ± 2,9 % | – |
| 45 | o3-mini (high) API access | OpenAI | 4,2 % ± 2,9 % | – |
| 46 | Qwen 3.6 Max (Preview) | Alibaba | 4,2 % ± 2,9 % | – |
| 47 | GLM-4.6 (Together) Open weights (unrestricted) | Z.ai (Zhipu AI),Tsinghua University | 2,1 % ± 2,1 % | – |
| 48 | DeepSeek-V3.2 (Thinking; Fireworks) Open weights (unrestricted) | DeepSeek | 2,1 % ± 2,1 % | – |
| 49 | GLM-5 Open weights (unrestricted) | Z.ai (Zhipu AI) | 2,1 % ± 2,1 % | – |
| 50 | claude-haiku-4-5-20251001_32K API access | Anthropic | 2,1 % ± 2,1 % | – |
| 51 | Claude Opus 4.5 (16k thinking) API access | Anthropic | 2,1 % ± 2,1 % | – |
| 52 | Claude Sonnet 4.5 (no thinking) API access | Anthropic | 2,1 % ± 2,1 % | – |
| 53 | Gemini 2.5 Pro Preview (Jun 2025) API access | Google DeepMind | 2,1 % ± 2,1 % | – |
| 54 | GPT-5 nano (medium) API access | OpenAI | 2,1 % ± 2,1 % | – |
| 55 | grok-4-0709 API access | xAI | 2,1 % ± 2,1 % | – |
| 56 | o3 (high) API access | OpenAI | 2,1 % ± 2,1 % | – |
| 57 | o4-mini (medium) API access | OpenAI | 2,1 % ± 2,1 % | – |
| 58 | Qwen3.5 Plus | Alibaba | 2,1 % ± 2,1 % | – |
| 59 | GPT-5.4 mini (high) API access | OpenAI | 2,1 % ± 2,1 % | – |
| 60 | Grok 4 Heavy (web app) Unreleased | xAI | 2,1 % ± 2,1 % | – |
| 61 | Qwen3-235B-A22B-Thinking-2507 Open weights (unrestricted) | Alibaba | 0 % ± 0 % | – |
| 62 | Claude 3.5 Sonnet (Jun 2024) API access | Anthropic | 0 % ± 0 % | – |
| 63 | Claude 3.5 Sonnet (Oct 2024) API access | Anthropic | 0 % ± 0 % | – |
| 64 | Claude Opus 4 API access | Anthropic | 0 % ± 0 % | – |
| 65 | claude-sonnet-4-20250514 API access | Anthropic | 0 % ± 0 % | – |
| 66 | GPT-4.1 API access | OpenAI | 0 % ± 0 % | – |
| 67 | GPT-5 nano (high) API access | OpenAI | 0 % ± 0 % | – |
| 68 | grok-3-beta API access | xAI | 0 % ± 0 % | – |
| 69 | Kimi K2 Thinking (Together) Open weights (restricted use) | Moonshot | 0 % ± 0 % | – |
| 70 | Qwen3.5 Flash | Alibaba | 0 % ± 0 % | – |
| 71 | Qwen 3.6 Flash (2026-04-16) | Alibaba | 0 % ± 0 % | – |
| 72 | GLM-4.7 (Together) Open weights (unrestricted) | Z.ai (Zhipu AI) | 0 % ± 0 % | – |
