Inference
Epoch AI

FrontierMath-Tier-4-v2-Private

Evaluated by Epoch AI on their own harness.

ModelsModels with a result.53
Top resultBest result on this test.97.6%GPT-6 Astra
Top-3 spreadPoints from first to third.9.8 pts
YearYear of release.N/A

Result and price

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

Results

53 results
#ModelResultThe score from the source.
01GPT-6 Astrahigh effort97.6% ±2.4
02Claude Fable 5max effort90.2% ±4.6
03Claude Fable 5.1max effort87.8% ±5.2
04GPT-5.6 Solmax effort82.9% ±5.9
05GPT-5.5 Proxhigh effort78.0% ±6.5
06AI Co-Mathematician75.6% ±6.7
07Claude Opus 5max effort73.2% ±7.0
08GPT-5.5xhigh effort72.5% ±7.1
09GPT-5.6 Terramax effort70.7% ±7.2
10GPT-5.6 Lunamax effort61.0% ±7.7
11GPT-5.4 Proxhigh effort58.5% ±7.8
12Claude Opus 4.8max effort56.1% ±7.8
13GPT-5.4xhigh effort49.0% ±8.0
14Muse Spark 1.3max effort46.3% ±7.9
14Qwen3.8 Maxxhigh effort46.3% ±7.9
16GPT-5.2 Proxhigh effort46.0% ±7.8
17Gemini 3.7 Flashhigh effort36.6% ±7.6
18Qwen3.7 Max34.1% ±7.5
18Qwen3.8 Max (0902)xhigh effort34.1% ±7.5
20Grok 4.6xhigh effort31.7% ±7.4
20Claude Opus 4.7max effort31.7% ±7.4
22GPT-5.2xhigh effort31.7% ±7.3
23GLM-5.3max effort29.3% ±7.2
23Claude Sonnet 5max effort29.3% ±7.2
23GLM-5.2max effort29.3% ±7.2
26Gemini 3.5 Flashhigh effort26.8% ±7.0
26Gemini 3.1 Pro26.8% ±7.0
26Claude Opus 4.6max effort26.8% ±7.0
29Grok 4.5high effort24.4% ±6.8
30Gemini 3.8 Flashhigh effort22.0% ±6.5
30Gemini 3.6 Flashhigh effort22.0% ±6.5
30GPT-5high effort22.0% ±6.5
33GPT-5 Prohigh effort19.5% ±6.3
34GLM-5.3-Flashmax effort17.1% ±5.9
34Inkling-Smallxhigh effort17.1% ±5.9
34Gemini 3 Flash17.1% ±5.9
34Grok 4.2017.1% ±5.9
38Grok 4.3 Betahigh effort14.6% ±5.6
39Kimi K2.7 Code12.2% ±5.2
39GPT-5.4 nanohigh effort12.2% ±5.2
39GPT-5 minihigh effort12.2% ±5.2
42GPT-5.4 minixhigh effort9.8% ±4.7
43Inklingxhigh effort4.9% ±3.4
43Claude Opus 4.54.9% ±3.4
43o4-mini (high)high effort4.9% ±3.4
46DeepSeek V4 Pro 0423max effort2.4% ±2.4
46Claude Sonnet 4.52.4% ±2.4
46Claude Opus 4.12.4% ±2.4
46GPT-5 nanohigh effort2.4% ±2.4
46GPT-5.5 Instant2.4% ±2.4
51Gemini 3.5 Flash-Litehigh effort0.0%
51Gemini 2.5 Pro (Jun 2025)0.0%
51o3-minihigh effort0.0%