BrowseComp

Citation frequency 2/3
What does it measure?
A test by OpenAI of persistence in web research: finding information that takes repeated searches and piecing scattered clues together.
Who evaluates it, and how?
1,266 hard questions. Answers are short, so they can be checked directly against a reference answer, but they are hard to find.
Reading the score
Scores run from 0 to 100%. Higher is better. There is no official leaderboard, so every value is self-reported by the model makers, and their test settings may differ.

This ranking is built only from scores the model makers published themselves. Test settings such as tool use and reasoning effort differ by maker, so check each score's setting before comparing models directly.

Top score92.2%
Models tested40
Scoring basisHigher is better
Last updated2026.09.20
Model release date

Can it track down hard-to-find information on the web?

BrowseComp

  1. 🥇OpenAIGPT-5.6 SolReasoning effort: Max92.2%
  2. 🥈OpenAIGPT-6 Astra91.5%
  3. 🥉Moonshot AIKimi K3Reasoning effort: Max91.2%
  4. #4AnthropicClaude Opus 5Reasoning effort: Max90.8%
  5. #5OpenAIGPT-5.4 ProReasoning effort: Extra High89.3%
  6. #6StepfunStep-5Reasoning effort: High88.7%
  7. #7AnthropicClaude Opus 4.8Reasoning effort: Max88.5%
  8. #8OpenAIGPT-5.6 Terra87.5%
  9. #9AnthropicClaude Fable 587.4%
  10. #10AnthropicClaude Sonnet 5Reasoning effort: Max86.6%
  11. #11AnthropicClaude Opus 4.6Reasoning effort: Max86.6%
  12. #12Moonshot AIKimi K2.6Reasoning effort: Thinking86.3%
  13. #13GoogleGemini 3.1 ProReasoning effort: High85.9%
  14. #14OpenAIGPT-5.5Reasoning effort: Extra High84.4%
  15. #15MiniMaxMiniMax M383.5%
  16. #16DeepSeekDeepSeek V4 ProReasoning effort: Max83.4%
  17. #17OpenAIGPT-5.6 Luna83.3%
  18. #18OpenAIGPT-5.4Reasoning effort: Extra High82.7%
  19. #19AnthropicClaude Sonnet 4.6Reasoning effort: Max82.1%
  20. #20MeituanLongCat 2.079.9%
  21. #21AnthropicClaude Opus 4.779.8%
  22. #22Moonshot AIKimi K2.5Reasoning effort: Thinking78.4%
  23. #23ByteDanceDola Seed 2.0 Pro77.3%
  24. #24MiniMaxMiniMax M2.776.3%
  25. #24MiniMaxMiniMax M2.576.3%
  26. #26StepfunStep 3.7 Flash75.8%
  27. #27DeepSeekDeepSeek V4 FlashReasoning effort: Max73.2%
  28. #28Z.aiGLM-5.168.0%
  29. #29AnthropicClaude Opus 4.5Reasoning effort: Max (no thinking)67.8%
  30. #30TencentHy367.1%
  31. #31Z.aiGLM-562.0%
  32. #32DeepSeekDeepSeek V3.2Reasoning effort: Thinking51.4%
  33. #33UpstageSolar Pro 449.2%
  34. #34Mistral AIMistral Medium 3.5Reasoning effort: Max48.6%
  35. #35NVIDIANemotron 3 Ultra44.4%
  36. #36AnthropicClaude Sonnet 4.543.9%
  37. #37LG AI ResearchK-EXAONEReasoning effort: None31.4%
  38. #38NVIDIANemotron 3 Super31.3%
  39. #39Mistral AIMistral Small 421.3%
  40. #40SK TelecomA.X K2Reasoning effort: Thinking9.3%
Half of models ≤ 79.9%Measured by: Maker-reported

The "harness" is the agent program the AI used to carry out the task. The same model can score very differently depending on its harness and reasoning effort.

  • Self-reported by each model maker — see the source link on each scoreFigures that model makers announced themselves, checked against the original post or system card and copied over. Conditions differ by maker, and each score links to its original source.
  • BrowseComp