GPQA-DiamondReasoning

GPQA (Graduate-Level Google-Proof Q&A) was released in 2023 by New York University and other institutions. It contains 448 difficult multiple-choice questions written by PhD-level experts in biology, physics, and chemistry. Even with web search, non-experts achieve 34% accuracy versus 65% for experts, providing human baselines for evaluating scientific knowledge and reasoning depth.

Ranking Sources

#ModelScore

1

GPT-6 AstraOpenAI

96.0%

2

Gemini 3.8 FlashGoogle

95.3%

3

GPT-5.6 SolOpenAI

94.6%

4

Claude Mythos PreviewAnthropic

94.6%

5

GPT-5.4 ProOpenAI

94.4%

6

Gemini 3.1 ProGoogle

94.3%

7

Claude Opus 4.7Anthropic

94.2%

8

Claude Fable 5.1Anthropic

93.7%

9

Claude Opus 5Anthropic

93.7%

10

Claude Opus 4.8Anthropic

93.6%

138 models total

Back to benchmarks

GPQA-DiamondReasoning

Official website

GPQA (Graduate-Level Google-Proof Q&A) was released in 2023 by New York University and other institutions. It contains 448 difficult multiple-choice questions written by PhD-level experts in biology, physics, and chemistry. Even with web search, non-experts achieve 34% accuracy versus 65% for experts, providing human baselines for evaluating scientific knowledge and reasoning depth.

Ranking Sources

#ModelScore

1

GPT-6 AstraOpenAI

96.0%

2

Gemini 3.8 FlashGoogle

95.3%

3

GPT-5.6 SolOpenAI

94.6%

4