MoveBench, described in a new arXiv preprint, is a rubric-based benchmark for evaluating large language models in sports and health coaching. The authors argue that this domain is high-risk, yet lacks dedicated evaluation tools. General-purpose benchmarks such as MMLU are not designed to capture the nuanced requirements of coaching tasks.
The benchmark is intended to fill that gap by providing a domain-specific evaluation framework. Because the abstract does not include experimental results, the paper's contribution at this stage is the benchmark design itself rather than findings about model performance.
The key implication is that LLM deployment in coaching contexts needs tailored assessment, not just broad knowledge tests. MoveBench represents an attempt to create that assessment.