Static benchmarks are dying — they tend to get saturated quickly.

Evaluation and training data should co-evolve with frontier models.

We released BenchEvolver — a framework that automatically evolves saturated problems into harder, verified tasks for evaluating frontier models, which can also serve as useful self-improvement signals for RL.

New work from UC Berkeley @berkeley_ai @BerkeleyRDI @BerkeleySky

Project Page: http://benchevolver.github.io Paper: https://arxiv.org/abs/2606.01286