Vero: An Open RL Recipe for General Visual Reasoning

Vero introduces a family of open vision-language models (VLMs) designed for general visual reasoning, including Vero-600K, a dataset comprising 600,000 samples sourced from 59 datasets. The models demonstrate significant performance improvements, with Vero-Qwen3I-8B achieving an average gain of 3.8 points over existing models without additional distillation. This release is crucial for AI practitioners as it provides fully open resources for studying and extending visual reasoning capabilities across diverse tasks, facilitating reproducibility and innovation in the field.