MLE-Bench Lite
An evaluation benchmark for machine learning engineering capability, built on Kaggle competition tasks, that measures a model's ability to autonomously complete MLE experiments through medal rates.
Timeline (last 90 days)
MLExpert raised the base model's average medal rate on MLE-Bench Lite from 37.9% to 72.7%.
The paper evaluates Qwen3.6-27B on five competition tasks from MLE-Bench Lite and Qwen3-4B on the text adventure game Zork I (Jericho framework).
On MLE-Bench, RL does not yield task performance improvements over the untrained environment-modification approach.
All Facts (3)
MLExpert raised the base model's average medal rate on MLE-Bench Lite from 37.9% to 72.7%.
50%UnverifiedThe paper evaluates Qwen3.6-27B on five competition tasks from MLE-Bench Lite and Qwen3-4B on the text adventure game Zork I (Jericho framework).
50%UnverifiedOn MLE-Bench, RL does not yield task performance improvements over the untrained environment-modification approach.
50%