Inference
EN
Epoch AI

FrontierMath-Tier-4-v2-Private

Evaluated by Epoch AI on their own harness.

ModelsModels with a result.72
Top resultBest result on this benchmark.100.0%Claude Fable 5.2
Top-3 spreadPoints from first to third.2.4 pts
YearYear of release.—

Result and price

67 of 72 results

Results

72 results
#ModelResultThe score from the source.
01Claude Fable 5.2max effort100.0%
01GPT-6.1 Solmax effort100.0%
03GPT-6 Astramax effort97.6% ±2.4
04Gemini 4 Carbonhigh effort95.1% ±3.4
05Claude Opus 5.5max effort95.0% ±4.1
06GPT-6.1 Terramax effort92.7% ±4.1
07Claude Fable 5max effort90.2% ±4.6
08GPT-6 Solmax effort90.0% ±5.2
09Claude Fable 5.1max effort87.8% ±5.2
10GPT-5.6 Solmax effort82.9% ±5.9
11Claude Sonnet 5.5max effort80.5% ±6.3
12Gemini 4 Argonhigh effort78.0% ±6.5
12GPT-5.5 Proxhigh effort78.0% ±6.5
14AI Co-Mathematician75.6% ±6.7
15Claude Opus 5max effort73.2% ±7.0
16GPT-5.5xhigh effort72.5% ±7.1
17Muse Wondermax effort70.7% ±7.2
17GPT-5.6 Terramax effort70.7% ±7.2
19GPT-5.6 Lunamax effort61.0% ±7.7
20GPT-5.4 Proxhigh effort58.5% ±7.8
21Claude Opus 4.8max effort56.1% ±7.8
21GPT-6 Lunamax effort56.1% ±7.8
23GPT-5.4xhigh effort49.0% ±8.0
24Muse Spark 1.3max effort46.3% ±7.9
24Claude Haiku 5.5max effort46.3% ±7.9
24Qwen3.8 Maxxhigh effort46.3% ±7.9
27GPT-5.2 Proxhigh effort46.0% ±7.8
28Kimi K3max effort39.0% ±7.7
29Gemini 3.7 Flashhigh effort36.6% ±7.6
30Qwen3.8 Max (0902)xhigh effort34.1% ±7.5
30Qwen3.7 Max34.1% ±7.5
32Grok 4.6xhigh effort31.7% ±7.4
32Muse Spark Openxhigh effort31.7% ±7.4
32Claude Opus 4.7max effort31.7% ±7.4
32GPT-5.2xhigh effort31.7% ±7.3
36GLM-5.3max effort29.3% ±7.2
36Claude Sonnet 5max effort29.3% ±7.2
36GLM-5.2max effort29.3% ±7.2
39DeepSeek V4.1 Flash 1017max effort26.8% ±7.0
39DeepSeek V4 Pro 0813max effort26.8% ±7.0
39Gemini 3.5 Flashhigh effort26.8% ±7.0
39Gemini 3.1 Pro26.8% ±7.0
39Mistral Large 4high effort26.8% ±7.0
39Claude Opus 4.6max effort26.8% ±7.0
45Kimi K2.625.6% ±7.1
46Grok 4.5high effort24.4% ±6.8
46DeepSeek V4 Flash 0731max effort24.4% ±6.8
48Gemini 3.8 Flashhigh effort22.0% ±6.5
48Gemini 3.6 Flashhigh effort22.0% ±6.5
48GPT-5high effort22.0% ±6.5
51GPT-5 Prohigh effort19.5% ±6.3
52Grok 4.7xhigh effort17.1% ±5.9
52GLM-5.3-Flashmax effort17.1% ±5.9
52Inkling-Smallxhigh effort17.1% ±5.9
52Gemini 3 Flash17.1% ±5.9
52Grok 4.2017.1% ±5.9
57Grok 4.3 Betahigh effort14.6% ±5.6
58Kimi K2.7 Code12.2% ±5.2
58GPT-5.4 nanohigh effort12.2% ±5.2
58GPT-5 minihigh effort12.2% ±5.2
61GPT-5.4 minixhigh effort9.8% ±4.7
62Inklingxhigh effort4.9% ±3.4
62Claude Opus 4.54.9% ±3.4
62o4-minihigh effort4.9% ±3.4
65DeepSeek V4 Pro 0423max effort2.4% ±2.4
65Claude Sonnet 4.52.4% ±2.4
65Claude Opus 4.12.4% ±2.4
65GPT-5 nanohigh effort2.4% ±2.4
65GPT-5.5 Instant2.4% ±2.4
70Gemini 3.5 Flash-Litehigh effort0.0%
70o3-minihigh effort0.0%
70Gemini 2.5 Pro (Jun 2025)0.0%