LiveCodeBench v6
CodingContamination-free competitive programming: problems are continuously collected from LeetCode, AtCoder and Codeforces after model cutoffs and scored as pass@1. Higher is better.
LiveCodeBench continuously scrapes new problems from LeetCode, AtCoder and Codeforces contests, tagging each with its publication date so models can be evaluated only on problems released after their training cutoff. Release v6 contains 1,055 problems spanning May 2023 through April 2025. The headline metric is pass@1 on code generation, computed with test-case checkers, and the official leaderboard exposes a date-range slider that recomputes scores for the selected window while flagging potentially contaminated models. Most 2026 frontier scores on this board are vendor-reported on the full v6 set via aggregators, since the official site's own table has not been refreshed past the o3/Gemini 2.5 era.
- 1DeepSeek V4Pro Max93.5%DeepSeek
- 2Gemini 3 ProHigh91.7%Google DeepMind
- 391.6%Qwen
- 491.6%DeepSeek
- 5Gemini 3 FlashReasoning90.8%Google DeepMind
- 689.6%Moonshot AI
- 787.1%Qwen
- 886.4%StepFun
- 985%Moonshot AI
- 1084.9%Z.AI
- 1184.8%Anthropic
- 1283.1%Moonshot AI
- 1382.8%Z.AI
- 1481.7%ByteDance Seed
- 1580.2%OpenAI
- 16o3High75.8%OpenAI
- 1773.6%Google DeepMind
Category
CodingLCreated by
LiveCodeBenchModels tested
17Leader
DeepSeek V4Top score
93.5%
Updated June 2026