The R1-Onevision dataset was released by Zhejiang University in 2025. It aims to give the model advanced multimodal reasoning capabilities. The relevant blog is "R1-Onevision: Open-Source Multimodal Large Language Model with Reasoning AbilityIt bridges the gap between visual and textual understanding by enabling rich, context-aware reasoning tasks in multiple domains, including natural scenes, science, math problems, OCR-based content, and complex diagrams.