ViSTR-Bench

Researchers introduced ViSTR-Bench, a benchmark to evaluate whether multimodal large language models (MLLMs) can reason from continuous visual cues in dynamic scenes, finding that current models remai…