ddwang2000/MMSU

[ICLR 2026] MMSU: A Massive Multi-task Spoken Language Understanding and Reasoning Benchmark Overview of MMSU MMSU (Massive Multi-task Spoken Language Understanding and Reasoning Benchmark) is a comprehensive benchmark for evaluating fine-grained spoken language understanding and reasoning in multimodal models. It systematically captures the variance of real-world linguistic phenomena in daily speech through 47 sub-tasks, including phonetics, prosody, rhetoric… See the full description on the dataset page: https://huggingface.co/datasets/ddwang2000/MMSU.