Pinned
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,
