Region-Level Policy Optimization for Fine-grained MLLM Perception

Benchmark Model Rank Results
visual-question-answering-on-v-benchVision-RL² (Qwen3.5-9B)#1Accuracy: 95.3
visual-question-answering-vqa-on-hr-bench-4kVision-RL 2#1Overall: 78.8
visual-question-answering-vqa-on-hr-bench-8kVision-RL 2#1Overall: 75.0