Learn
What is LiveCodeBench?
Last reviewed July 16, 2026
LiveCodeBench is a coding benchmark built from a continuously updated pool of 1000+ problems collected from programming contest sites, each stamped with its publication date. Evaluators can score a model only on problems released after its training cutoff, which removes the classic failure of static coding benchmarks: models reciting solutions they saw during training. Scores therefore track genuine problem-solving.
The contamination problem it solves
Static benchmarks leak into training data over time, quietly inflating scores: a model can reproduce a memorized solution without being able to solve a new problem. Date-stamped rolling collection makes the leak detectable and avoidable, and a large gap between a model's scores on pre-cutoff and post-cutoff problems is itself evidence of contamination.
What it covers
Problems span code generation, self-repair from execution feedback, test-output prediction, and code execution, drawn from live contest difficulty tiers. That makes it a strong signal for algorithmic problem-solving; it does not cover repository-scale engineering, which is what SWE-bench style suites measure.
idapt's coding rankings include a LiveCodeBench lane, and you can run the same coding task across models in one chat, executing the code to check the answers rather than trusting scores.
Frequently asked
See it in practice
The fastest way to understand it is to use it. Start free, no credit card.
Try idapt free