ProcessThinker: Enhancing Multi-modal Large Language Models Reasoning via Rollout-based Process Reward

In three linesProcessThinker enhances multimodal reasoning by providing step-level process rewards without training an explicit process reward model. The method rewrites reasoning traces, applies GRPO with rollout-based rewards (empirical success rate), and improves Qwen3-VL-8B-Instruct across four video benchmarks (Video-MMMU, MMVU, VideoMathQA, LongVideoBench).

Summary generated by Claude — human-verified