On May 13th, Tencent Hunyuan, in partnership with the Shanghai AI Lab, Fudan University, and the Shanghai Institute of Intelligent Science and Technology, introduced a groundbreaking research project titled UnifiedReward-Think. This initiative has successfully developed the first-ever unified multimodal reward model, endowed with advanced long-chain reasoning capabilities. This model is capable of "thinking" across a spectrum of visual tasks, markedly improving the evaluation accuracy, cross-task generalization, and interpretability of reasoning for intricate visual generation and comprehension assignments. The entire project, including the model, datasets, training scripts, and evaluation tools, has now been fully open-sourced.
