Skip to content

ResearchClawBench

An end-to-end autonomous scientific research benchmark with 40 tasks across 10 scientific domains, where agents receive related literature and raw data, then attempt to rediscover the hidden target paper.

19 models

Higher is better.

  1. 1Claude Opus 4.8Anthropic21.1
  2. 2Claude Opus 4.7Anthropic20.7
  3. 3GLM-5.2Z.AI20.7
  4. 4Claude Opus 4.6Anthropic19.9
  5. 5MiniMax M3MiniMax19.8
  6. 6Qwen3.7 MaxAlibaba18.7
  7. 7GLM-5.1Z.AI18.2
  8. 8Gemini 3.5 FlashGoogle18.0
  9. 9Kimi K2.6Moonshot AI18.0
  10. 10Qwen3.6 PlusAlibaba18.0
  11. 11GPT-5.5OpenAI17.0
  12. 12MiMo-V2.5Xiaomi16.9
  13. 13GPT-5.4OpenAI15.3
  14. 14MiMo-V2-ProXiaomi15.3
  15. 15Qwen3.5 397BAlibaba14.2
  16. 16Kimi K2.5Moonshot AI14.0
  17. 17Grok 4.1xAI13.5
  18. 18Gemini 3.1 ProGoogle13.3
  19. 19Grok 4.3xAI12.4
Higher is better. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Metric
End-to-end autonomous research evaluation with RADS scoring
Tasks
40 tasks across 10 scientific domains
Authors
InternScience (2026)

Scores are as reported by model developers or evaluators and may use different settings (for example reasoning effort or tools). Compare with care.