@BenchBench

A new benchmark called BenchBench tests AI models on their ability to create benchmarks for other models, revealing that only GPT 5.2 succeeded in generating a practically solvable yet challenging eva…