FrontierMath-Tier-4-v2-Private
Re-run by Epoch AIThe hardest tier of FrontierMath (second version): unpublished research-level mathematics problems written for Epoch AI.
- Kind
- Run independently by Epoch AI, with public logs. Our own re-runs are not built yet.
- Score shown
- Epoch's column "Best score (across scorers)", as a percentage, higher is better; random guessing scores about 0.0%. Error bars as the source gives them.
- Results
- 83 in all: 16 for open-weight models, 67 for closed ones. 15 linked to a model page here.
Best open-weight result: moonshotai/Kimi-K3.1-Pro, 65.9%, number 27 of 83 results.
| # | Model | Weights | Setting | Score | Date | Source |
|---|---|---|---|---|---|---|
| 1 | claude-fable-5-2_maxAnthropic | closed | effort max | 100.0% | 6 Nov 2026 | Epoch AI |
| 2 | gpt-6.1-sol_maxOpenAI | closed | effort max | 100.0% | 29 Sep 2026 | Epoch AI |
| 3 | gpt-6.2-astra_maxOpenAI | closed | effort max | 100.0% | 27 Oct 2026 | Epoch AI |
| 4 | gpt-6-astra_highOpenAI | closed | effort high | 97.6%±2.4 | 30 Aug 2026 | Epoch AI |
| 5 | gpt-6-astra_maxOpenAI | closed | effort max | 97.6%±2.4 | 30 Aug 2026 | Epoch AI |
| 6 | gpt-6-astra_xhighOpenAI | closed | effort xhigh | 97.6%±2.4 | 30 Aug 2026 | Epoch AI |
| 7 | gpt-6-astra_mediumOpenAI | closed | effort medium | 97.6%±2.4 | 30 Aug 2026 | Epoch AI |
| 8 | gemini-4-carbon_highGoogle DeepMind | closed | effort high | 95.1%±3.4 | 2 Nov 2026 | Epoch AI |
| 9 | claude-opus-5-5_maxAnthropic | closed | effort max | 95.0%±4.1 | 22 Sep 2026 | Epoch AI |
| 10 | gpt-6.1-terra_maxOpenAI | closed | effort max | 92.7%±4.1 | 26 Oct 2026 | Epoch AI |
| 11 | claude-fable-5_maxAnthropic | closed | effort max | 90.2%±4.6 | 9 Jun 2026 | Epoch AI log |
| 12 | gpt-6-sol_maxOpenAI | closed | effort max | 90.0%±5.2 | 22 Sep 2026 | Epoch AI |
| 13 | claude-fable-5-1_maxAnthropic | closed | effort max | 87.8%±5.2 | 1 Sep 2026 | Epoch AI |
| 14 | gemini-4-hydrogen_xhighGoogle DeepMind | closed | effort xhigh | 87.8%±5.2 | 13 Nov 2026 | Epoch AI |
| 15 | gpt-6-astra_lowOpenAI | closed | effort low | 87.8%±5.2 | 30 Aug 2026 | Epoch AI |
| 16 | gpt-5.6-sol_maxOpenAI | closed | effort max | 82.9%±5.9 | 9 Jul 2026 | Epoch AI log |
| 17 | gpt-6-astra_noneOpenAI | closed | effort none | 82.9%±5.9 | 30 Aug 2026 | Epoch AI |
| 18 | claude-sonnet-5-5_maxAnthropic | closed | effort max | 80.5%±6.3 | 29 Sep 2026 | Epoch AI |
| 19 | gpt-5.6-sol_promaxOpenAI | closed | default | 80.5%±6.3 | 9 Jul 2026 | Epoch AI log |
| 20 | gemini-4-argon_highGoogle DeepMind | closed | effort high | 78.0%±6.5 | 13 Oct 2026 | Epoch AI |
| 21 | gpt-5.5-pro_xhighOpenAI | closed | effort xhigh | 78.0%±6.5 | 12 Jun 2026 | Epoch AI |
| 22 | gdm-ai-co-mathematicianGoogle DeepMind | closed | default | 75.6%±6.7 | 12 Jun 2026 | Epoch AI |
| 23 | claude-opus-5_maxAnthropic | closed | effort max | 73.2%±7.0 | 24 Jul 2026 | Epoch AI log |
| 24 | gpt-5.5_xhighOpenAI | closed | effort xhigh | 72.5%±7.1 | 11 Jun 2026 | Epoch AI |
| 25 | gpt-5.6-terra_maxOpenAI | closed | effort max | 70.7%±7.2 | 9 Jul 2026 | Epoch AI log |
| 26 | muse-wonder-1.0_maxMeta AI | closed | effort max | 70.7%±7.2 | 12 Oct 2026 | Epoch AI |
| 27 | moonshotai/Kimi-K3.1-ProEpoch: kimi-k3.1-pro-together_max, Moonshot | open | effort max | 65.9%±7.5 | 6 Nov 2026 | Epoch AI |
| 28 | gpt-5.6-luna_maxOpenAI | closed | effort max | 61.0%±7.7 | 9 Jul 2026 | Epoch AI log |
| 29 | gpt-5.4-pro-2026-03-05_xhighOpenAI | closed | effort xhigh | 58.5%±7.8 | 13 Jun 2026 | Epoch AI log |
| 30 | claude-opus-4-8_maxAnthropic | closed | effort max | 56.1%±7.8 | 10 Jun 2026 | Epoch AI log |
| 31 | gpt-6-luna_maxOpenAI | closed | effort max | 56.1%±7.8 | 22 Sep 2026 | Epoch AI |
| 32 | gpt-5.4-2026-03-05_xhighOpenAI | closed | effort xhigh | 49.0%±8.0 | 11 Jun 2026 | Epoch AI log |
| 33 | claude-haiku-5-5_maxAnthropic | closed | effort max | 46.3%±7.9 | 7 Oct 2026 | Epoch AI |
| 34 | deepseek-ai/DeepSeek-V4.1-Flash-1017Epoch: deepseek-v4.1-flash-1017_max, DeepSeek | open | effort max | 46.3%±7.9 | 19 Oct 2026 | Epoch AI |
| 35 | muse-spark-1.3_maxMeta AI | closed | effort max | 46.3%±7.9 | 18 Sep 2026 | Epoch AI |
| 36 | qwen3.8-max_xhighAlibaba | closed | effort xhigh | 46.3%±7.9 | 4 Aug 2026 | Epoch AI log |
| 37 | gpt-5.2-pro-2025-12-11_xhighOpenAI | closed | effort xhigh | 46.0%±7.8 | 13 Jun 2026 | Epoch AI |
| 38 | muse-spark-1.3_xhighMeta AI | closed | effort xhigh | 41.5%±7.8 | 16 Sep 2026 | Epoch AI |
| 39 | moonshotai/Kimi-K3Epoch: kimi-k3_max, Moonshot | open | effort max | 39.0%±7.7 | 17 Jul 2026 | Epoch AI log |
| 40 | gemini-3.7-flash_highGoogle DeepMind | closed | effort high | 36.6%±7.6 | 14 Aug 2026 | Epoch AI log |
| 41 | qwen3.7-maxAlibaba | closed | default | 34.1%±7.5 | 13 Jun 2026 | Epoch AI log |
| 42 | qwen3.8-max-0902_xhighAlibaba | closed | effort xhigh | 34.1%±7.5 | 2 Sep 2026 | Epoch AI |
| 43 | claude-opus-4-7_maxAnthropic | closed | effort max | 31.7%±7.4 | 10 Jun 2026 | Epoch AI log |
| 44 | grok-4.6_xhighxAI | closed | effort xhigh | 31.7%±7.4 | 14 Aug 2026 | Epoch AI log |
| 45 | meta-models/Muse-Spark-Open-816B-A42BEpoch: Muse-Spark-Open-816B-A42B_xhigh, Meta AI | open | effort xhigh | 31.7%±7.4 | 15 Oct 2026 | Epoch AI |
| 46 | gpt-5.2-2025-12-11_xhighOpenAI | closed | effort xhigh | 31.7%±7.3 | 11 Jun 2026 | Epoch AI |
| 47 | claude-sonnet-5_maxAnthropic | closed | effort max | 29.3%±7.2 | 30 Jun 2026 | Epoch AI log |
| 48 | zai-org/GLM-5.2Epoch: glm-5.2_max, Z.ai (Zhipu AI) | open | effort max | 29.3%±7.2 | 19 Jun 2026 | Epoch AI log |
| 49 | zai-org/GLM-5.3Epoch: glm-5.3_max, Z.ai (Zhipu AI) | open | effort max | 29.3%±7.2 | 25 Aug 2026 | Epoch AI |
| 50 | claude-opus-4-6_maxAnthropic | closed | effort max | 26.8%±7.0 | 11 Jun 2026 | Epoch AI log |
| 51 | deepseek-ai/DeepSeek-V4-Pro-0813Epoch: deepseek-v4-pro-0813_max, DeepSeek | open | effort max | 26.8%±7.0 | 19 Aug 2026 | Epoch AI log |
| 52 | deepseek-ai/DeepSeek-V4.1-FlashEpoch: deepseek-v4.1-flash_max, DeepSeek | open | effort max | 26.8%±7.0 | 8 Oct 2026 | Epoch AI |
| 53 | gemini-3.1-pro-previewGoogle DeepMind | closed | default | 26.8%±7.0 | 11 Jun 2026 | Epoch AI log |
| 54 | gemini-3.5-flash_highGoogle DeepMind | closed | effort high | 26.8%±7.0 | 10 Jun 2026 | Epoch AI log |
| 55 | mistral-large-4-0_highMistral AIno page match | open | effort high | 26.8%±7.0 | 12 Oct 2026 | Epoch AI |
| 56 | moonshotai/Kimi-K2.6Epoch: kimi-k2.6, Moonshot | open | default | 25.6%±7.1 | 10 Jun 2026 | Epoch AI log |
| 57 | deepseek-ai/DeepSeek-V4-Flash-0731Epoch: deepseek-v4-flash-0731_max, DeepSeek | open | effort max | 24.4%±6.8 | 2 Aug 2026 | Epoch AI log |
| 58 | grok-4.5_highxAI | closed | effort high | 24.4%±6.8 | 9 Jul 2026 | Epoch AI |
| 59 | gemini-3.6-flash_highGoogle DeepMind | closed | effort high | 22.0%±6.5 | 2 Aug 2026 | Epoch AI log |
| 60 | gemini-3.8-flash_highGoogle DeepMind | closed | effort high | 22.0%±6.5 | 2 Sep 2026 | Epoch AI |
| 61 | gpt-5-2025-08-07_highOpenAI | closed | effort high | 22.0%±6.5 | 11 Jun 2026 | Epoch AI |
| 62 | gpt-5-pro-2025-10-06_highOpenAI | closed | effort high | 19.5%±6.3 | 12 Jun 2026 | Epoch AI log |
| 63 | gemini-3-flash-previewGoogle DeepMind | closed | default | 17.1%±5.9 | 11 Jun 2026 | Epoch AI log |
| 64 | zai-org/GLM-5.3-FlashEpoch: glm-5.3-flash_max, Z.ai (Zhipu AI) | open | effort max | 17.1%±5.9 | 27 Aug 2026 | Epoch AI |
| 65 | grok-4.20-0309-reasoningxAI | closed | default | 17.1%±5.9 | 13 Jul 2026 | Epoch AI log |
| 66 | grok-4.7_xhighxAI | closed | effort xhigh | 17.1%±5.9 | 22 Sep 2026 | Epoch AI |
| 67 | thinkingmachines/Inkling-SmallEpoch: inkling-small_xhigh, Thinking Machines | open | effort xhigh | 17.1%±5.9 | 14 Aug 2026 | Epoch AI log |
| 68 | grok-4.3_highxAI | closed | effort high | 14.6%±5.6 | 17 Jun 2026 | Epoch AI log |
| 69 | gpt-5-mini-2025-08-07_highOpenAI | closed | effort high | 12.2%±5.2 | 12 Jun 2026 | Epoch AI log |
| 70 | gpt-5.4-nano-2026-03-17_highOpenAI | closed | effort high | 12.2%±5.2 | 12 Jun 2026 | Epoch AI log |
| 71 | moonshotai/Kimi-K2.7-CodeEpoch: kimi-k2.7-code, Moonshot | open | default | 12.2%±5.2 | 13 Jun 2026 | Epoch AI log |
| 72 | gpt-5.4-mini-2026-03-17_xhighOpenAI | closed | effort xhigh | 9.8%±4.7 | 12 Jun 2026 | Epoch AI |
| 73 | claude-opus-4-5-20251101_32KAnthropic | closed | default | 4.9%±3.4 | 11 Jun 2026 | Epoch AI log |
| 74 | thinkingmachines/InklingEpoch: Inkling_xhigh, Thinking Machines | open | effort xhigh | 4.9%±3.4 | 6 Aug 2026 | Epoch AI log |
| 75 | o4-mini-2025-04-16_highOpenAI | closed | effort high | 4.9%±3.4 | 11 Jun 2026 | Epoch AI log |
| 76 | claude-opus-4-1-20250805_32KAnthropic | closed | default | 2.4%±2.4 | 11 Jun 2026 | Epoch AI log |
| 77 | claude-sonnet-4-5-20250929_32KAnthropic | closed | default | 2.4%±2.4 | 11 Jun 2026 | Epoch AI log |
| 78 | deepseek-ai/DeepSeek-V4-ProEpoch: deepseek-v4-pro_max, DeepSeek | open | effort max | 2.4%±2.4 | 17 Jun 2026 | Epoch AI log |
| 79 | gpt-5-nano-2025-08-07_highOpenAI | closed | effort high | 2.4%±2.4 | 12 Jun 2026 | Epoch AI log |
| 80 | gpt-5.5-instantOpenAI | closed | default | 2.4%±2.4 | 2 Aug 2026 | Epoch AI log |
| 81 | gemini-2.5-proGoogle DeepMind | closed | default | 0.0% | 11 Jun 2026 | Epoch AI log |
| 82 | gemini-3.5-flash-lite_highGoogle DeepMind | closed | effort high | 0.0% | 2 Aug 2026 | Epoch AI log |
| 83 | o3-mini-2025-01-31_highOpenAI | closed | effort high | 0.0% | 11 Jun 2026 | Epoch AI log |
Open-weight rows are shaded. Dates are when the result was run or evaluated, where the data says. Data from Epoch AI (CC-BY 4.0), fetched 19 Nov 2026: Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from ‘https://epoch.ai/benchmarks’ [online resource]. Epoch AI. Method