TimeThink: Reasoning with Time for Video LLMs
2026-07-06 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors introduce TimeThink, a new method to help video large language models (Video-LLMs) better understand and reason about events happening over time in long videos. Unlike previous methods that only judge the model based on the final answer, TimeThink gives step-by-step feedback on how well the model identifies important time segments during its reasoning process. They created a new dataset to train their system and tested it on multiple tasks, finding that it improves both locating key moments and reasoning about them. This leads to better performance compared to other open-source video reinforcement learning models.
Video Large Language ModelsReinforcement LearningTemporal LocalizationTemporal EvidenceVideo ReasoningReward FunctionStep-wise OptimizationTemporal GroundingDataset ConstructionProcess-Outcome Optimization
Authors
Handong Li, Longteng Guo, Zikang Liu, Dongze Hao, Yepeng Tang, Zijia Zhao, Jie Jiang, Zhiwei Jin, Chen Chen, Haonan Lu, Jing Liu
Abstract
Video reasoning requires models to identify and verify temporally localized evidence within long video sequences. Recent Video Large Language Models (Video-LLMs) have shown promising reasoning abilities when aligned with reinforcement learning, yet existing approaches typically rely on outcome-based rewards that supervise only the final prediction. Such supervision provides limited guidance on how models should discover the relevant temporal evidence during intermediate reasoning. In this work, we propose TimeThink, a reinforcement learning framework that explicitly guides temporal evidence discovery in Video-LLMs. Our key idea is to treat temporal clue steps as the fundamental optimization primitive of video reasoning, where each reasoning step references a candidate time interval in the video. We introduce a step-wise temporal process reward that provides localized credit assignment for these clues and a joint process--outcome optimization objective that balances reasoning fidelity with task correctness. To enable scalable training, we construct TimeThink-RFT-20K, a dataset with automatically derived temporal evidence segments. Extensive experiments across video reasoning, temporal grounding, and general video understanding benchmarks show that TimeThink consistently improves both temporal localization and reasoning performance, achieving state-of-the-art results among open-source video RL models.