Code generation · contests · reference
LiveCodeBench
Competitive-programming problems collected after models were trained, so answers can’t have been memorised. Scored on the first attempt.
- Models
- 20
- By makers
- 20
- Latest
- Aug 7
pass@1, %. v marks a score the model’s maker reported. Prices are FullsendAI’s per million tokens, blended 3 input : 1 output.
About the data
From LiveCodeBench. No licence on the leaderboard site; the benchmark code is MIT. Window ends May 2025.