Inference
Epoch AI

FrontierMath-Tier-4-2025-07-01-Private

Evaluated by Epoch AI on their own harness.

ModelsModels with a result.54
Top resultBest result on this test.47.9%AI Co-Mathematician
Top-3 spreadPoints from first to third.10.4 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

54 results
#ModelResultThe score from the source.
01AI Co-Mathematician47.9% ±7.2
02GPT-5.5 Proxhigh effort39.6% ±7.1
03GPT-5.4 Pro37.5% ±7.0
04GPT-5.5xhigh effort35.4% ±6.9
05GPT-5.2 Pro31.3% ±6.7
06Claude Opus 4.8max effort31.3% ±6.8
07GPT-5.4xhigh effort27.1% ±6.4
08Claude Opus 4.7xhigh effort22.9% ±6.1
09Claude Opus 4.6max effort22.9% ±6.1
10GPT-5.2xhigh effort18.8% ±5.6
11Gemini 3 Pro18.8% ±5.7
12Gemini 3.1 Pro16.7% ±5.4
13Muse Spark14.6% ±5.1
13GPT-5 Prohigh effort14.6% ±5.1
15Gemini 3.5 Flashhigh effort14.6% ±5.1
16GLM-5.112.5% ±4.8
16GPT-5.1high effort12.5% ±4.8
16GPT-5high effort12.5% ±4.8
19Gemini 2.5 Deep Think10.4% ±4.4
20Qwen3.6 Plus8.3% ±4.0
21Claude Sonnet 4.68.3% ±4.0
22GPT-5.4 nanohigh effort6.3% ±3.5
22GPT-5 minihigh effort6.3% ±3.5
22o4-mini (high)high effort6.3% ±3.5
25Kimi K2.54.2% ±2.9
26Qwen 3.6 Max (preview)4.2% ±2.9
26Gemini 3 Flash4.2% ±2.9
26Claude Opus 4.54.2% ±2.9
26Claude Opus 4.14.2% ±2.9
26o3-minihigh effort4.2% ±2.9
26Gemini 2.5 Flash4.2% ±2.9
32Gemini 2.5 Pro (Jun 2025)3.1% ±2.9
32Claude Sonnet 4.53.1% ±2.9
34GLM-4.62.1% ±2.1
35GLM-52.1% ±2.1
35DeepSeek V3.22.1% ±2.1
37o3high effort2.1% ±2.1
37Grok 42.1% ±2.1
37Claude Opus 42.1%
37Claude Haiku 4.52.1% ±2.1
37Qwen 3.5 Plus (hosted 397B-A17B)2.1% ±2.1
42GPT-5.4 minihigh effort2.1% ±2.1
42Grok 4 Heavy2.1% ±2.1
44GLM-4.70.0%
44Kimi K2 Thinking0.0%
44Claude Sonnet 40.0%
44GPT-5 nanohigh effort0.0%
44Grok 3 [Beta]0.0%
44GPT-4.10.0%
44Claude 3.5 Sonnet0.0%
44Claude 3.5 Sonnet (October 2024)0.0%
44Qwen 3.5 Flash (hosted 35B-A3B)0.0%
44Qwen3-235B-A22B (Jul 2025)0.0%
44Qwen3.6 Flash0.0%