MVPBench

MVPBench is a benchmark that contains 24,020 high-quality instances annotated with fine-grained value labels, personalized questions, and demographic metadata, used to evaluate the alignment of large language models with diverse human values across 75 countries.

Papers using MVPBench (3)

Accuracy Without Grounding: Diagnosing Visual Dependency Dissociation in Video LLM Benchmarks2026MVPBench: A Benchmark and Fine-Tuning Framework for Aligning Large Language Models with Diverse Human Values2025ReGATE: Learning Faster and Better with Fewer Tokens in MLLMs2025