Inference
EN
Epoch AI

FrontierMath-Tiers-1-3-v2-Private

Evaluated by Epoch AI on their own harness.

ModelsModels with a result.90
Top resultBest result on this benchmark.95.1%Claude Fable 5.2
Top-3 spreadPoints from first to third.1.4 pts
YearYear of release.—

Result and price

84 of 90 results

Results

90 results
#ModelResultThe score from the source.
01Claude Fable 5.2max effort95.1% ±1.3
02GPT-6 Astramax effort93.7% ±1.4
02GPT-6.1 Solmax effort93.7% ±1.4
04Claude Opus 5.5max effort91.2% ±1.7
05GPT-6.1 Terramax effort90.9% ±1.7
06Gemini 4 Carbonhigh effort90.5% ±1.7
07Claude Fable 5.1max effort90.2% ±1.8
08GPT-6 Solmax effort89.8% ±1.8
09GPT-5.6 Solmax effort89.1% ±1.8
10Claude Sonnet 5.5max effort88.8% ±1.9
11GPT-5.5 Proxhigh effort87.7% ±1.9
12Claude Fable 5max effort87.0% ±2.0
13Gemini 4 Argonhigh effort86.0% ±2.1
13GPT-5.6 Terramax effort86.0% ±2.1
15Claude Opus 5max effort85.6% ±2.1
16GPT-5.5xhigh effort85.3% ±2.1
17GPT-5.4 Proxhigh effort82.5% ±2.3
18Muse Wondermax effort82.1% ±2.3
18GPT-5.6 Lunamax effort82.1% ±2.3
20Claude Opus 4.8max effort80.0% ±2.4
21GPT-6 Lunamax effort78.9% ±2.4
22GPT-5.4xhigh effort78.6% ±2.4
23Claude Haiku 5.5max effort75.1% ±2.6
24Qwen3.8 Maxxhigh effort74.7% ±2.6
25Muse Spark 1.3max effort74.0% ±2.6
25GPT-5.2 Proxhigh effort74.0% ±2.6
27Kimi K3max effort72.2% ±2.7
28Gemini 3.7 Flashhigh effort71.6% ±2.7
29Claude Opus 4.7max effort70.2% ±2.7
30Muse Spark Openxhigh effort69.8% ±2.7
31GLM-5.3max effort68.8% ±2.7
32Gemini 3.8 Flashhigh effort68.4% ±2.8
33GPT-5.2xhigh effort67.4% ±2.8
33DeepSeek V4.1 Flash 1017max effort67.4% ±2.8
35Grok 4.6xhigh effort66.0% ±2.8
35Claude Opus 4.6max effort66.0% ±2.8
37Claude Sonnet 5max effort65.6% ±2.8
37Qwen3.8 Max (0902)xhigh effort65.6% ±2.8
39DeepSeek V4 Pro 0813max effort64.6% ±2.8
39Qwen3.7 Max64.6% ±2.8
41Gemini 3.5 Flashhigh effort62.8% ±2.9
42Gemini 3.1 Pro59.6% ±2.9
43GLM-5.2max effort59.2% ±3.0
44Gemini 3.6 Flashhigh effort58.9% ±2.9
45Mistral Large 4high effort58.6% ±2.9
46DeepSeek V4 Flash 0731max effort57.5% ±2.9
47Grok 4.5high effort57.2% ±2.9
47Kimi K2.657.2% ±2.9
49GLM-5.3-Flashmax effort55.8% ±2.9
49GPT-5 Prohigh effort55.8% ±2.9
51GPT-5high effort55.4% ±2.9
52Kimi K2.7 Code54.0% ±3.0
53Grok 4.7xhigh effort53.0% ±3.0
54Gemini 3 Flash51.2% ±3.0
54GPT-5.4 minixhigh effort51.2% ±3.0
56GPT-5 minihigh effort46.7% ±3.0
57Inkling-Smallxhigh effort46.3% ±3.0
58DeepSeek V4 Pro 0423max effort45.3% ±3.0
59Grok 4.2044.9% ±3.0
59GPT-5.4 nanohigh effort44.9% ±3.0
61Grok 4.3 Betahigh effort42.8% ±2.9
62o4-minihigh effort36.1% ±2.9
63Qwen3.7 Plusnone effort34.4% ±2.8
63Claude Opus 4.534.4% ±2.8
65Qwen3.6-27Bnone effort34.0% ±2.8
66Inklingxhigh effort33.3% ±2.8
66o3high effort33.3% ±2.8
68Qwen3.6 Plusnone effort32.3% ±2.8
69Qwen3.5 397B A17Bnone effort31.2% ±2.7
70Gemini 3.1 Flash-Litehigh effort27.7% ±2.7
71GPT-5.5 Instant26.3% ±2.6
72Gemini 3.5 Flash-Litehigh effort26.0% ±2.6
73GLM-5.1none effort24.9% ±2.6
74Gemini 2.5 Pro (Jun 2025)24.6% ±2.6
75Claude Sonnet 4.523.9% ±2.5
76Qwen3.6 Flashnone effort22.5% ±2.5
77Qwen3.6-35B-A3Bnone effort20.4% ±2.4
78GPT-5 nanohigh effort20.0% ±2.4
79Qwen3.7 Flashnone effort19.3% ±2.3
80Qwen3 Max18.9% ±2.3
81o3-minihigh effort18.6% ±2.3
82Qwen 3.5 Flash (hosted 35B-A3B)none effort18.2% ±2.3
83o1high effort14.7% ±2.1
84Claude Opus 4.112.6% ±2.0
85GPT-4.1 mini6.7% ±1.5
86GPT-4.16.0% ±1.4
87GPT-4o mini0.7% ±0.5
87GPT-4 Turbo (Apr 2024)0.7% ±0.5
89GPT-4o0.4% ±0.4
90GPT-3.5 Turbo0.0%