AdaThinkV: Adaptive Thinking for Token-Efficient Video Reasoning
2026-08-03 • Computer Vision and Pattern Recognition
Computer Vision and Pattern RecognitionArtificial Intelligence
AI summaryⓘ
The authors developed AdaThinkV, a system that helps a video-based language model decide when to think carefully before answering questions and when to respond quickly. It learns this by trying both detailed thinking and simple answering during training, without needing extra labels or hard rules. They introduced a new method called VRPO to help the model learn better from hard but solvable questions. When tested, AdaThinkV answered more accurately while using fewer words compared to other systems. This shows it can smartly balance effort and efficiency when answering video questions.
Chain-of-thought reasoningvideo multimodal large language modelreinforcement learningadaptive reasoningrolloutaccuracy gainautoregressive sequencepolicy optimizationvariance recovery
Authors
Jingqi Tian, Haoji Zhang, Lin Chen, Hongbo Jin, Haonan Xu, Tianrui Zhu, Xingming Shui, Shilin Ma, Wenjing Yang, Yansong Tang
Abstract
Chain-of-thought (CoT) reasoning can improve performance on difficult video questions but often wastes decoding tokens on simple ones. We study whether a video multimodal large language model can adapt its reasoning effort to each question. We propose AdaThinkV, an adaptive framework for video reasoning that learns whether to reason explicitly without offline difficulty labels, manually tuned confidence thresholds, or an external router. During reinforcement learning, AdaThinkV samples matched rollouts in explicit reasoning and direct answering modes for each prompt. ThinkGain estimates the prompt-level utility of explicit reasoning by balancing its accuracy gain against additional response length, providing supervision for both conditional response generation and autonomous mode selection. For difficult prompts, limited rollout exploration can yield groups in which every response is unsuccessful and accuracy rewards show little variation, providing insufficient signal for learning. We therefore introduce Variance Recovery Policy Optimization (VRPO), which retains and progressively expands these groups to recover informative signals from prompts that are difficult yet solvable. At inference, AdaThinkV selects a response mode and generates the response in a single autoregressive sequence. Across a unified suite of video reasoning evaluations, AdaThinkV achieves a mean accuracy of 40.79 with an average of 257.20 output tokens, outperforming the strongest evaluated adaptive baseline by 2.98 points while using 22.7% fewer tokens. Project page: https://trilarflagz.github.io/AdaThinkV/