ParaVT: Taming the Tool Prior Paradox for Parallel Tool Use in Agentic Video Reinforcement Learning

Benchmark Model Rank Results
video-question-answering-on-mmvuParaVT (Qwen3-VL-8B)#14Accuracy: 68.6
video-question-answering-on-video-mme-with-subtitlesParaVT (Qwen3-VL-8B)#27Accuracy: 69.4
moment-retrieval-on-charades-staParaVT (Qwen3-VL-8B)#77mIoU: 50.1