rrchen2026/Theory_of_Mind_CoMMET

CoMMET CoMMET is a multi-turn, multimodal benchmark designed to evaluate the Theory of Mind (ToM) capabilities of multimodal large language models (MLLMs). Unlike conventional single-turn Theory of Mind benchmarks, CoMMET represents each scenario as a sequence of interconnected turns. Models are required to reason about stories, previous interactions, feedback, questions, and, when necessary, visual information. The benchmark covers multiple types of mental-state reasoning and… See the full description on the dataset page: https://huggingface.co/datasets/rrchen2026/Theory_of_Mind_CoMMET.