Benchmark
MLE-bench
Compare agents building machine-learning solutions for data-science competitions.
Published leaderboard
Explore scores and evaluation settings on the publisher’s leaderboard.
Open leaderboardCoverage
75 Kaggle competitions; MLE-bench Lite is a separate 22-competition subset.
Methodology
Agents train models and submit predictions on held-out splits. The leaderboard compares medal achievement against competition thresholds and reports repeated-run uncertainty.