PlanningBench: Generating Scalable and Verifiable Planning Data for Evaluating and Training Large Language Models

In three linesPlanningBench is a framework for generating scalable and verifiable planning data. It abstracts 30+ task types and difficulty factors from real scenarios, then synthesizes problems with adaptive control and automatic verification. RL training on verified data improves performance on unseen benchmarks.

Summary generated by Claude — human-verified