Skip to content
Weights Not Walls
Menu

FrontierMath-Tier-4-2025-07-01-Private

Re-run by Epoch AI

The hardest tier of FrontierMath: unpublished research-level mathematics problems written for Epoch AI.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
72 in all: 9 for open-weight models, 59 for closed ones, 4 where Epoch does not say. 9 linked to a model page here.

Best open-weight result: moonshotai/Kimi-K2.6, 14.6%, number 22 of 72 results.

FrontierMath-Tier-4-2025-07-01-Private results
#ModelWeightsSettingScoreDateSource
1gdm-ai-co-mathematicianGoogle DeepMindcloseddefault47.9%±7.28 May 2026Epoch AI
2gpt-5.5-pro-pre-release_highOpenAIclosedeffort high39.6%±7.123 Apr 2026Epoch AI
3gpt-5.5-pro-pre-release_xhighOpenAIclosedeffort xhigh39.6%±7.123 Apr 2026Epoch AI
4gpt-5.4-pro-2026-03-05-web-appOpenAIcloseddefault37.5%±7.06 Mar 2026Epoch AI
5gpt-5.5-pre-release_xhighOpenAIclosedeffort xhigh35.4%±6.923 Apr 2026Epoch AI
6gpt-5.2-pro-2025-12-11-webappOpenAIcloseddefault31.3%±6.724 Dec 2025Epoch AI
7claude-opus-4-8_maxAnthropicclosedeffort max31.3%±6.88 Jun 2026Epoch AI log
8gpt-5.4-2026-03-05_xhighOpenAIclosedeffort xhigh27.1%±6.46 Mar 2026Epoch AI
9claude-opus-4-7_xhighAnthropicclosedeffort xhigh22.9%±6.117 Apr 2026Epoch AI
10claude-opus-4-6_maxAnthropicclosedeffort max22.9%±6.112 Feb 2026Epoch AI log
11claude-opus-4-6_32KAnthropiccloseddefault20.8%±5.96 Feb 2026Epoch AI log
12claude-opus-4-6_64KAnthropiccloseddefault20.8%±5.96 Feb 2026Epoch AI log
13gpt-5.2-2025-12-11_highOpenAIclosedeffort high18.8%±5.611 Dec 2025Epoch AI log
14gpt-5.2-2025-12-11_xhighOpenAIclosedeffort xhigh18.8%±5.614 Dec 2025Epoch AI
15gemini-3-pro-previewGoogle DeepMindcloseddefault18.8%±5.721 Nov 2025Epoch AI
16gemini-3.1-pro-previewGoogle DeepMindcloseddefault16.7%±5.419 Feb 2026Epoch AI log
17gpt-5.2-2025-12-11_mediumOpenAIclosedeffort medium16.7%±5.411 Dec 2025Epoch AI log
18gpt-5-pro-2025-10-06_highOpenAIclosedeffort high14.6%±5.19 Oct 2025Epoch AI
19muse-sparkMeta AIcloseddefault14.6%±5.18 Apr 2026Epoch AI
20claude-opus-4-6Anthropiccloseddefault14.6%±5.15 Feb 2026Epoch AI log
21gemini-3.5-flash_highGoogle DeepMindclosedeffort high14.6%±5.125 May 2026Epoch AI
22moonshotai/Kimi-K2.6Epoch: kimi-k2.6, Moonshotopendefault14.6%±5.18 May 2026Epoch AI
23zai-org/GLM-5.1Epoch: glm-5.1, Z.ai (Zhipu AI)opendefault12.5%±4.812 May 2026Epoch AI
24gpt-5-2025-08-07_highOpenAIclosedeffort high12.5%±4.830 Oct 2025Epoch AI log
25gpt-5.1-2025-11-13_highOpenAIclosedeffort high12.5%±4.813 Nov 2025Epoch AI log
26gemini-2.5-deep-think-2025-08-01-webappGoogle,Google DeepMindcloseddefault10.4%±4.4not datedEpoch AI
27qwen3.6-plusAlibabacloseddefault8.3%±4.012 May 2026Epoch AI
28claude-sonnet-4-6_16KAnthropiccloseddefault8.3%±4.021 Feb 2026Epoch AI log
29gpt-5-2025-08-07_mediumOpenAIclosedeffort medium6.3%±3.57 Aug 2025Epoch AI log
30gpt-5-mini-2025-08-07_highOpenAIclosedeffort high6.3%±3.530 Oct 2025Epoch AI log
31gpt-5.2-2025-12-11_lowOpenAIclosedeffort low6.3%±3.511 Dec 2025Epoch AI log
32gpt-5.4-nano-2026-03-17_highOpenAIclosedeffort high6.3%±3.515 Apr 2026Epoch AI
33o4-mini-2025-04-16_highOpenAIclosedeffort high6.3%±3.51 Jul 2025Epoch AI log
34moonshotai/Kimi-K2.5Epoch: fireworks/kimi-k2p5, Moonshotopendefault4.2%±2.92 Feb 2026Epoch AI
35claude-opus-4-1-20250805_27KAnthropiccloseddefault4.2%±2.95 Aug 2025Epoch AI log
36claude-opus-4-20250514_27KAnthropiccloseddefault4.2%±2.91 Jul 2025Epoch AI log
37claude-opus-4-5-20251101Anthropiccloseddefault4.2%±2.925 Nov 2025Epoch AI log
38claude-opus-4-5-20251101_32KAnthropiccloseddefault4.2%±2.925 Nov 2025Epoch AI log
39claude-sonnet-4-5-20250929_32KAnthropiccloseddefault4.2%±2.922 Oct 2025Epoch AI log
40gemini-2.5-flashGoogle DeepMindcloseddefault4.2%±2.918 Dec 2025Epoch AI log
41gemini-2.5-proGoogle DeepMindcloseddefault4.2%±2.93 Jul 2025Epoch AI log
42gemini-3-flash-previewGoogle DeepMindcloseddefault4.2%±2.917 Dec 2025Epoch AI log
43gpt-5-mini-2025-08-07_mediumOpenAIclosedeffort medium4.2%±2.97 Aug 2025Epoch AI log
44gpt-5.1-2025-11-13_mediumOpenAIclosedeffort medium4.2%±2.917 Nov 2025Epoch AI log
45o3-mini-2025-01-31_highOpenAIclosedeffort high4.2%±2.91 Jul 2025Epoch AI log
46qwen3.6-max-previewAlibabanot stateddefault4.2%±2.928 May 2026Epoch AI
47zai-org/GLM-4.6Epoch: zai-org/GLM-4.6, Z.ai (Zhipu AI),Tsinghua Universityopendefault2.1%±2.18 Dec 2025Epoch AI log
48deepseek-ai/DeepSeek-V3.2Epoch: fireworks/deepseek-v3p2, DeepSeekopendefault2.1%±2.116 Dec 2025Epoch AI
49zai-org/GLM-5Epoch: glm-5, Z.ai (Zhipu AI)opendefault2.1%±2.119 Feb 2026Epoch AI log
50claude-haiku-4-5-20251001_32KAnthropiccloseddefault2.1%±2.122 Oct 2025Epoch AI log
51claude-opus-4-5-20251101_16KAnthropiccloseddefault2.1%±2.125 Nov 2025Epoch AI log
52claude-sonnet-4-5-20250929Anthropiccloseddefault2.1%±2.129 Sep 2025Epoch AI log
53gemini-2.5-pro-preview-06-05Google DeepMindcloseddefault2.1%±2.13 Jul 2025Epoch AI log
54gpt-5-nano-2025-08-07_mediumOpenAIclosedeffort medium2.1%±2.17 Aug 2025Epoch AI log
55grok-4-0709xAIcloseddefault2.1%±2.111 Aug 2025Epoch AI log
56o3-2025-04-16_highOpenAIclosedeffort high2.1%±2.11 Jul 2025Epoch AI log
57o4-mini-2025-04-16_mediumOpenAIclosedeffort medium2.1%±2.17 Aug 2025Epoch AI log
58qwen3.5-plusAlibabanot stateddefault2.1%±2.115 May 2026Epoch AI
59gpt-5.4-mini-2026-03-17_highOpenAIclosedeffort high2.1%±2.115 Apr 2026Epoch AI
60grok-4-heavy-web-appxAIcloseddefault2.1%±2.19 Oct 2025Epoch AI
61claude-3-5-sonnet-20240620Anthropiccloseddefault0.0%1 Jul 2025Epoch AI log
62claude-3-5-sonnet-20241022Anthropiccloseddefault0.0%1 Jul 2025Epoch AI log
63claude-opus-4-20250514Anthropiccloseddefault0.0%1 Jul 2025Epoch AI log
64claude-sonnet-4-20250514Anthropiccloseddefault0.0%1 Jul 2025Epoch AI log
65gpt-4.1-2025-04-14OpenAIcloseddefault0.0%1 Jul 2025Epoch AI log
66gpt-5-nano-2025-08-07_highOpenAIclosedeffort high0.0%30 Oct 2025Epoch AI log
67grok-3-betaxAIcloseddefault0.0%1 Jul 2025Epoch AI log
68moonshotai/Kimi-K2-ThinkingEpoch: moonshotai/Kimi-K2-Thinking, Moonshotopendefault0.0%4 Dec 2025Epoch AI log
69Qwen/Qwen3-235B-A22B-Thinking-2507Epoch: Qwen/Qwen3-235B-A22B-Thinking-2507, Alibabaopendefault0.0%11 Dec 2025Epoch AI log
70qwen3.5-flashAlibabanot stateddefault0.0%12 May 2026Epoch AI
71qwen3.6-flashAlibabanot stateddefault0.0%12 May 2026Epoch AI
72zai-org/GLM-4.7Epoch: zai-org/GLM-4.7, Z.ai (Zhipu AI)opendefault0.0%30 Jan 2026Epoch 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