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

GPQA Diamond

Graduate-level science questions designed to be Google-proof.

Items
198
Revision
2024.11
Contamination
watched
Scorer
scorer.choice.v3
Dataset revision
[email protected]
Scorer
exact-choice · scorer.choice.v3
License
CC BY 4.0
Kind
Public reproduction suite
ModelScore$/taskMedian sToolsEvidence
o4-pro90.4$0.63838.20.2LIVE_MEASURED
Grok 4 Heavy88.1$0.13222.60.2LIVE_MEASURED
Grok 484.8$0.03810.40.2LIVE_MEASURED
Gemini 2.5 Pro83.4$0.0258.70.2LIVE_MEASURED
GPT-582.6$0.06312.90.2LIVE_MEASURED
Claude 4.1 Opus81.2$0.26716.80.2LIVE_MEASURED
DeepSeek V3.179.6$0.00428.20.2REPRODUCED
Qwen3-Max78.2$0.00569.70.2LIVE_MEASURED
Claude 4.1 Sonnet76.8$0.04711.40.2LIVE_MEASURED
Mistral Large 374.6$0.0159.90.2LIVE_MEASURED
Gemini 2.5 Flash72.1$0.00113.30.2LIVE_MEASURED
GPT-5 Mini71.4$0.00315.10.2LIVE_MEASURED
Llama 4 405B70.8$0.00439.10.2REPRODUCED
Grok 3 Mini68.2$0.00164.00.2LIVE_MEASURED
Llama 4 70B61.4$0.000941.90.2REPRODUCED