Wikibench

GPQA

reasoning Main

A challenging dataset of 448 multiple-choice questions written by domain experts in biology, physics, and chemistry. Questions are Google-proof and extremely difficult, with PhD experts reaching 65% accuracy.

Leaderboard

Showing 20 of 224 results

  1. Claude Mythos Preview
    94.6%
    i
  2. Gemini 3.1 Pro
    94.3%
    i
  3. Claude Opus 4.7
    94.2%
    i
  4. Claude Opus 4.8 max
    93.6%
    i
  5. GPT-5.5
    93.6%
    i
  6. GPT-5.2 Pro
    93.2%
    i
  7. GPT-5.4
    92.8%
    i
  8. GPT-5.2
    92.4%
    i
  9. Qwen3.7 Max
    92.4%
    i
  10. Gemini 3 Pro
    91.9%
    i
  11. Claude Opus 4.6
    91.3%
    i
  12. Kimi K2.6
    90.5%
    i
  13. Gemini 3 Flash
    90.4%
    i
  14. Qwen3.6 Plus
    90.4%
    i
  15. DeepSeek-V4-Pro-Max
    90.1%
    i
  16. Claude Sonnet 4.6
    89.9%
    i
  17. Muse Spark
    89.5%
    i
  18. Seed 2.0 Pro
    88.9%
    i
  19. Grok-4 Heavy
    88.4%
    i
  20. Qwen3.5-397B-A17B
    88.4%
    i