SPACEXAI
4/5 workers samplingPOSSIBLE_MODEL_CHANGE · gpt-5
public · math

FrontierMath

Original research-level math. Licensed subset only.

Items
300
Revision
2026.01
Contamination
clean
Scorer
scorer.fm.v2
Dataset revision
[email protected]
Scorer
equiv-numeric · scorer.fm.v2
License
Epoch / FrontierMath license
Kind
Public reproduction suite
ModelScore$/taskMedian sToolsEvidence
o4-pro48.8$0.63881.30.2LIVE_MEASURED
Grok 4 Heavy36.2$0.13246.70.2LIVE_MEASURED
DeepSeek V3.134.1$0.004217.10.2REPRODUCED
GPT-531.2$0.06326.50.2LIVE_MEASURED
Qwen3-Max29.6$0.005620.10.2LIVE_MEASURED
Grok 428.4$0.03821.10.2LIVE_MEASURED
Gemini 2.5 Pro27.4$0.02517.70.2LIVE_MEASURED
Claude 4.1 Opus26.8$0.26734.20.2LIVE_MEASURED
Claude 4.1 Sonnet19.6$0.04723.00.2LIVE_MEASURED
Mistral Large 316.2$0.0158.30.2LIVE_MEASURED
Llama 4 405B14.8$0.00437.70.2REPRODUCED
GPT-5 Mini12.4$0.00314.20.2LIVE_MEASURED
Gemini 2.5 Flash11.2$0.00112.80.2LIVE_MEASURED
Grok 3 Mini8.6$0.00163.30.2LIVE_MEASURED
Llama 4 70B7.4$0.000941.60.2REPRODUCED