LiveBench
livebench.ai
An LLM benchmark that continuously updates with new questions sourced from recent news, research, and competitions — designed so models cannot be trained on the answers and results cannot be gamed.
Why it is useful
Solves the contamination problem with static benchmarks: once a benchmark is published, model developers can train on it or cherry-pick results. LiveBench uses questions from sources published after most model training cutoffs, making it impossible to inflate scores by memorizing answers. The leaderboard updates frequently and covers a wider range of skills than most benchmarks.