Benchmarks / Math

FrontierMath Tier 4 (v2)

Epoch AI's FrontierMath Tier 4, version 2: expert-written mathematics problems evaluated with Python tools. Accuracy on the private set, not the public sample questions. Data: Epoch AI, CC BY.

66models
–Open Index weight
–last updated
Epoch AIpublisher

FrontierMath Tier 4 (v2)

Higher is better · best effort per model

Chart settings
66 of 66 selected

full table below

#ModelEffortScoreCost / taskHarness
1GPT-6.1 Solmax100.0%–Epoch AI
2GPT-6 Astrahigh97.6%–Epoch AI
3Opus 5.5max95.0%–Epoch AI
4Fable 5max90.2%–Epoch AI
5GPT-6 Solmax90.0%–Epoch AI
6Fable 5.1max87.8%–Epoch AI
7GPT-5.6 Solmax82.9%–Epoch AI
8Sonnet 5.5max80.5%–Epoch AI
9GPT-5.6 Sol Promax80.5%–Epoch AI
10Gemini 4high78.0%–Epoch AI
11GPT-5.5 Proxhigh78.0%–Epoch AI
12Opus 5max73.2%–Epoch AI
13GPT-5.5xhigh72.5%–Epoch AI
14GPT-5.6 Terramax70.7%–Epoch AI
15GPT-5.6 Lunamax61.0%–Epoch AI
16GPT-5.4 Proxhigh58.5%–Epoch AI
17GPT-6 Lunamax56.1%–Epoch AI
18Opus 4.8max56.1%–Epoch AI
19GPT-5.4xhigh49.0%–Epoch AI
20Haiku 5.5max46.3%–Epoch AI
21Muse Spark 1.3max46.3%–Epoch AI
22GPT-5.2 Proxhigh46.0%–Epoch AI
23Kimi K3max39.0%–Epoch AI
24Gemini 3.7 Flashhigh36.6%–Epoch AI
25Qwen 3.8 Maxxhigh34.1%–Epoch AI
26Qwen 3.7 Maxmax34.1%–Epoch AI
27Grok 4.6xhigh31.7%–Epoch AI
28Opus 4.7max31.7%–Epoch AI
29GPT-5.2xhigh31.7%–Epoch AI
30GLM 5.3max29.3%–Epoch AI
31Sonnet 5max29.3%–Epoch AI
32GLM 5.2max29.3%–Epoch AI
33Mistral Large 4high26.8%–Epoch AI
34DeepSeek V4.1 Flashmax26.8%–Epoch AI
35DeepSeek V4 Pro 0813max26.8%–Epoch AI
36Gemini 3.1 Prohigh26.8%–Epoch AI
37Opus 4.6max26.8%–Epoch AI
38Gemini 3.5 Flashhigh26.8%–Epoch AI
39Kimi K2.6max25.6%–Epoch AI
40DeepSeek V4 Flash 0731max24.4%–Epoch AI
41Grok 4.5high24.4%–Epoch AI
42Gemini 3.8 Flashhigh22.0%–Epoch AI
43Gemini 3.6 Flashhigh22.0%–Epoch AI
44GPT-5high22.0%–Epoch AI
45GPT-5 Prohigh19.5%–Epoch AI
46Grok 4.7xhigh17.1%–Epoch AI
47GLM 5.3 Flashmax17.1%–Epoch AI
48Inkling Smallxhigh17.1%–Epoch AI
49Grok 4.20xhigh17.1%–Epoch AI
50Gemini 3 Flashhigh17.1%–Epoch AI
51Grok 4.3high14.6%–Epoch AI
52Kimi K2.7 Codemax12.2%–Epoch AI
53GPT-5 Minihigh12.2%–Epoch AI
54GPT-5.4 Nanohigh12.2%–Epoch AI
55GPT-5.4 Minixhigh9.8%–Epoch AI
56Inklingxhigh4.9%–Epoch AI
57o4 minihigh4.9%–Epoch AI
58Opus 4.5high4.9%–Epoch AI
59GPT-5.5 Instant–2.4%–Epoch AI
60DeepSeek V4 Promax2.4%–Epoch AI
61GPT-5 Nanohigh2.4%–Epoch AI
62Opus 4.1high2.4%–Epoch AI
63Sonnet 4.5high2.4%–Epoch AI
64Gemini 3.5 Flash-Litehigh0.0%–Epoch AI
65Gemini 2.5 Prohigh0.0%–Epoch AI
66o3 minihigh0.0%–Epoch AI