#vision-language tasks

ByteDance introduces VGR, a multimodal large language model that dynamically integrates visual information during reasoning, significantly improving accuracy and efficiency in vision-language benchmarks.