Thoughtful

An independent research lab developing benchmarks and tools for language-model post-training. Its PostTrainBench tests whether frontier agents can autonomously improve open models under fixed compute limits.