MoveBench: A Rubric-Based Benchmark for Evaluating Large Language Models in Sports and Health Coaching
Summary
MoveBench introduces the first rubric-based benchmark for evaluating large language models in sports and health coaching, covering safety, scientific validity, and personalization across multiple-choice and open-ended tasks. An evaluation of 15 LLMs on 809 questions with 5,012 rubric criteria reveals significant gaps in safety awareness and personalized program design.
View Cached Full Text
Cached at: 10/28/26, 05:25 AM
Paper page - MoveBench: A Rubric-Based Benchmark for Evaluating Large Language Models in Sports and Health Coaching
Source: https://huggingface.co/papers/2610.26185
Abstract
MoveBench introduces a specialized benchmark for evaluating large language models on sports and health coaching, emphasizing safety, scientific validity, and personalization across multiple-choice and open-ended tasks.
Despite rapid advances inlarge language models(LLMs), their deployment in the sports and health domain remains high-risk due to the absence of domain-specific evaluation benchmarks. Existing general-purpose benchmarks (e.g., MMLU and HELM) primarily assess broad knowledge and reasoning capabilities, but fail to capture critical domain requirements such as exercise safety, physiological appropriateness, personalization, and actionable coaching quality. This gap is particularly concerning in fitness and health applications, where incorrect or unsafe recommendations can lead to physical harm. To address this limitation, we introduceMoveBench, the first comprehensive benchmarking framework specifically designed for evaluating LLMs in sports and health coaching. The proposed benchmark integrates heterogeneous task formats, includingmultiple-choice questions(MCQs) andopen-ended question answering(OpenQA), grounded in real-world user scenarios and paired with structured rubrics emphasizingsafety compliance,scientific validity, andtraining logic. Through large-scale evaluation of 15 diverse LLMs on 809 questions with 5,012rubric-level criteria, our results reveal substantial performance variability, particularly in safety awareness, structured program design, and personalization capability, highlighting a critical gap between knowing and doing. Our work establishes a foundational evaluation standard for sports and health LLM applications and provides a reproducible framework for advancing safe, reliable, and context-aware AI-driven coaching systems. Open resources are available at https://github.com/SatoSakula/MoveBench.
View arXiv pageView PDFProject pageGitHub17Add to collection
Get this paper in your agent:
hf papers read 2610\.26185
Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash
Models citing this paper0
No model linking this paper
Cite arxiv.org/abs/2610.26185 in a model README.md to link it from this page.
Datasets citing this paper1
#### KeepaceAI/MoveBench Viewer• Updatedabout 2 hours ago • 80 • 535 • 2
Spaces citing this paper0
No Space linking this paper
Cite arxiv.org/abs/2610.26185 in a Space README.md to link it from this page.
Collections including this paper0
No Collection including this paper
Add this paper to acollectionto link it from this page.
Similar Articles
BrainBench: Benchmarking Large Language Models for Comprehensive EEG Understanding
BrainBench is a new unified benchmark for evaluating large language models on comprehensive, instruction-conditioned EEG understanding, covering 17 datasets, 172 tasks, and over 4K real-data instances. The paper evaluates 13 LLMs across two execution paradigms, showing that EEG competence varies by model and operationalization.
Benchmarking the Benchmarks: Evaluating Automated Safety Benchmarks for Small Language Models
This paper evaluates automated safety benchmarks for small language models, finding high ambiguity in judgments that compromises reliability and reveals a capability-safety confound.
MCBench: A Multicontext Safety Assessment Benchmark for Omni Large Language Models
MCBench is a new benchmark for assessing the safety of omnimodal large language models across vision, audio, and text modalities. It includes 1196 scenarios and finds current models struggle with cross-modal safety reasoning.
A Human-Centred Approach to Benchmarking LLMs for Parenting Advice
This paper presents a human-centred benchmarking approach for evaluating large language models in parenting advice, assessing 15 LLMs across 100 scenarios in English and Chinese using expert-derived rubrics.
MyoCardBench: A Real-World Data Benchmark for Evaluating Large Language Models in Clinically Authentic Cardiovascular Care Scenarios
MyoCardBench is a real-world benchmark for evaluating large language models in cardiovascular care, comprising 2,263 items across 13 tasks. GPT-5.4 achieved the highest overall score, demonstrating strengths in full-cycle care and multimodal interpretation.