// benchmark explainer
LiveCodeBench, explained
By the TrueIQ editorial team at Ahzy Capital · Last reviewed Oct 15, 2026 · How we work
LiveCodeBench keeps collecting new programming contest problems over time, so you can test models only on problems published after their training cutoff.
| Made by | UC Berkeley, MIT and Cornell researchers (2024) |
|---|---|
| Sources | LeetCode, AtCoder and Codeforces contests |
| Format | Write code that passes test cases; also self repair and test output prediction |
| Paper | arXiv 2403.07974 |
What it measures
Every problem has a release date, so results can be filtered to a time window. That helps answer whether a model can solve problems it couldn't have memorised.
Where it falls short
- Contest puzzles aren't everyday software work.
- Scores change with the chosen time window, so two 'LiveCodeBench' numbers may cover different problems.
- Mostly tests short, self contained programs.
How to read a score
Check which date window a score uses. Only compare models on the same window.
Where to see current scores
We don't republish scores. Leaderboards change weekly, and numbers depend on how the test was run. Check these instead:
- Artificial Analysis: LiveCodeBench results (independently run)
- LiveCodeBench leaderboard
- Paper