LiveCodeBench

What does it measure?
Artificial Analysis now classifies this as 'legacy' and rarely measures new models, so recent ones may be missing. Coding ability measured with competitive-programming problems: designing an algorithm and writing code that works correctly within the limits.
Who evaluates it, and how?
Researchers at UC Berkeley, MIT and Cornell continuously collect new contest problems from LeetCode, AtCoder and Codeforces. Each problem is dated, so models can be tested only on problems released after their training cut-off (to prevent memorisation). Submitted code is executed and scored by first-attempt pass rate (pass@1).
Reading the score
0–100%. Scores depend on which time window of problems was used, so check the window before comparing with LiveCodeBench numbers elsewhere.

Rank

Top score93.2%
Models tested29
Scoring basisHigher is better
Last updated2026.11.21
Model release date

Can it solve newly released programming problems?

LiveCodeBench

  1. 🥇Sakana AIFugu Ultra93.2%
  2. 🥈Sakana AIFugu92.9%
  3. 🥉UpstageSolar Pro 492.4%
  4. #4GoogleGemini 3 Flash90.8%
  5. #5OpenAIGPT OSS 120B87.8%
  6. #6AnthropicClaude Opus 4.587.1%
  7. #7DeepSeekDeepSeek V3.286.2%
  8. #8OpenAIGPT-584.6%
  9. #9OpenAIGPT-5 Mini83.8%
  10. #10SpaceXSIGrok 4.1 Fast (Reasoning)82.2%
  11. #11BaiduERNIE 5.0 Thinking81.2%
  12. #12GoogleGemini 2.5 Pro80.1%
  13. #13OpenAIGPT-5 Nano78.9%
  14. #14LG AI ResearchK-EXAONE76.8%
  15. #15NVIDIANemotron 3 Nano 30B A3B74.1%
  16. #16AnthropicClaude Sonnet 4.571.4%
  17. #17AmazonNova 2 Lite71.1%
  18. #18GoogleGemini 2.5 Flash69.5%
  19. #19GoogleGemini 2.5 Flash Lite68.8%
  20. #20AnthropicClaude Sonnet 465.5%
  21. #21AnthropicClaude Opus 4.165.4%
  22. #22AnthropicClaude Opus 463.6%
  23. #23AnthropicClaude Haiku 4.561.5%
  24. #24OpenAIGPT-4.145.7%
  25. #25SpaceXSIGrok 4.1 Fast39.9%
  26. #26MetaLlama 4 Maverick39.7%
  27. #27MetaLlama 4 Scout29.9%
  28. #28PerplexitySonar29.5%
  29. #29Mistral AIMistral Small 411.1%
Half of models ≤ 74.1%

59 of the 59 AIs released in the last 60 days have no score here yet · Gemini 4 Neon, GPT-6.2 Astra, Microsoft-Decision-1.1 and 56 more

This score comes from a fixed, predefined evaluation, so it may differ from what you get with your own question.

Scores labelled 'Self-reported' were published by the AI maker itself, not measured by a third party.

Related news