BBH

BBH (BIG-Bench Hard) is a challenging reasoning benchmark consisting of 23 demanding tasks selected from the original BIG-Bench suite where prior language models failed to surpass average human performance. The benchmark focuses on multi-step reasoning tasks and evaluates chain-of-thought prompting effectiveness. BBH tests complex reasoning abilities across diverse domains, serving as a rigorous assessment of advanced language model capabilities in scenarios requiring sophisticated logical thinking.