Inference
Epoch AI

FrontierMath-Tiers-1-3-v2-Private

Evaluated by Epoch AI on their own harness.

ModelsModels with a result.71
Top resultBest result on this test.93.7%GPT-6 Astra
Top-3 spreadPoints from first to third.4.6 pts
YearYear of release.N/A

Result and price

020406080100$0.1$0.3$1$3$10$30$100$300RESULTOUTPUT PRICE PER 1M TOKENS · LOG SCALE

Results

71 results
#ModelResultThe score from the source.
01GPT-6 Astramax effort93.7% ±1.4
02Claude Fable 5.1max effort90.2% ±1.8
03GPT-5.6 Solmax effort89.1% ±1.8
04GPT-5.5 Proxhigh effort87.7% ±1.9
05Claude Fable 5max effort87.0% ±2.0
06GPT-5.6 Terramax effort86.0% ±2.1
07Claude Opus 5max effort85.6% ±2.1
08GPT-5.5xhigh effort85.3% ±2.1
09GPT-5.4 Proxhigh effort82.5% ±2.3
10GPT-5.6 Lunamax effort82.1% ±2.3
11Claude Opus 4.8max effort80.0% ±2.4
12GPT-5.4xhigh effort78.6% ±2.4
13Qwen3.8 Maxxhigh effort74.7% ±2.6
14Muse Spark 1.3max effort74.0% ±2.6
15GPT-5.2 Proxhigh effort74.0% ±2.6
16Gemini 3.7 Flashhigh effort71.6% ±2.7
17Claude Opus 4.7max effort70.2% ±2.7
18GLM-5.3max effort68.8% ±2.7
19Gemini 3.8 Flashhigh effort68.4% ±2.8
20GPT-5.2xhigh effort67.4% ±2.8
21Grok 4.6xhigh effort66.0% ±2.8
21Claude Opus 4.6max effort66.0% ±2.8
23Claude Sonnet 5max effort65.6% ±2.8
23Qwen3.8 Max (0902)xhigh effort65.6% ±2.8
25Qwen3.7 Max64.6% ±2.8
26Gemini 3.5 Flashhigh effort62.8% ±2.9
27Gemini 3.1 Pro59.6% ±2.9
28GLM-5.2max effort59.2% ±3.0
29Gemini 3.6 Flashhigh effort58.9% ±2.9
30Grok 4.5high effort57.2% ±2.9
31GLM-5.3-Flashmax effort55.8% ±2.9
31GPT-5 Prohigh effort55.8% ±2.9
33GPT-5high effort55.4% ±2.9
34Kimi K2.7 Code54.0% ±3.0
35GPT-5.4 minixhigh effort51.2% ±3.0
35Gemini 3 Flash51.2% ±3.0
37GPT-5 minihigh effort46.7% ±3.0
38Inkling-Smallxhigh effort46.3% ±3.0
39DeepSeek V4 Pro 0423max effort45.3% ±3.0
40GPT-5.4 nanohigh effort44.9% ±3.0
40Grok 4.2044.9% ±3.0
42Grok 4.3 Betahigh effort42.8% ±2.9
43o4-mini (high)high effort36.1% ±2.9
44Qwen3.7 Plusnone effort34.4% ±2.8
44Claude Opus 4.534.4% ±2.8
46Qwen3.6-27Bnone effort34.0% ±2.8
47Inklingxhigh effort33.3% ±2.8
47o3high effort33.3% ±2.8
49Qwen3.6 Plusnone effort32.3% ±2.8
50Qwen3.5 397B A17Bnone effort31.2% ±2.7
51Gemini 3.1 Flash-Litehigh effort27.7% ±2.7
52GPT-5.5 Instant26.3% ±2.6
53Gemini 3.5 Flash-Litehigh effort26.0% ±2.6
54GLM-5.1none effort24.9% ±2.6
55Gemini 2.5 Pro (Jun 2025)24.6% ±2.6
56Claude Sonnet 4.523.9% ±2.5
57Qwen3.6 Flashnone effort22.5% ±2.5
58Qwen3.6-35B-A3Bnone effort20.4% ±2.4
59GPT-5 nanohigh effort20.0% ±2.4
60Qwen3.7 Flashnone effort19.3% ±2.3
61Qwen3 Max18.9% ±2.3
62o3-minihigh effort18.6% ±2.3
63Qwen 3.5 Flash (hosted 35B-A3B)none effort18.2% ±2.3
64o1high effort14.7% ±2.1
65Claude Opus 4.112.6% ±2.0
66GPT-4.1 mini6.7% ±1.5
67GPT-4.16.0% ±1.4
68GPT-4o mini0.7% ±0.5
68GPT-4 Turbo (Apr 2024)0.7% ±0.5
70GPT-4o0.4% ±0.4
71GPT-3.5 Turbo0.0%