🎬 Video Streaming Thinking (VST-7B)

VideoLLMs that watch and think simultaneously. Instead of waiting for the whole video before reasoning, VST consumes the clip in segments, emitting intermediate streaming thoughts as it watches, then delivers a grounded final answer — amortizing test-time reasoning for real-time responsiveness.

Built on Catalan258/VST-7B · paper · code · ECCV 2026