Paper 2511.04570
Thinking with Video: Video Generation as a Promising Multimodal Reasoning Paradigm
- Published
- Nov 2025
- Research lab
- Independent
- Citations
- 27
- GitHub
- 316 stars
01 In brief
Summary
The paper proposes 'Thinking with Video', a new paradigm using video generation models like Sora-2 for multimodal reasoning, addressing limitations of text- and image-based paradigms.
The authors introduce VideoThinkBench, a benchmark with vision-centric tasks (eyeballing puzzles, visual puzzles, ARC-AGI-2, mazes) and text-centric tasks (GSM8K, MATH, MMLU, etc.).
Sora-2 achieves 40.4% average on vision-centric tasks, comparable to SOTA VLMs, and surpasses GPT-5 by 10% on eyeballing puzzles.
On text-centric tasks, Sora-2 reaches 98.9% on GSM8K, 92.0% on MATH, and 69.2% on MMMU via audio evaluation.
Analysis shows Sora-2 is a few-shot learner, and self-consistency improves performance (e.g., Arc Connect accuracy rises from 68% to 90% with majority voting).
Text-centric abilities likely stem from an internal prompt rewriter, as Wan 2.5's performance drops to near zero without it.
The findings suggest video generation models can unify multimodal understanding and generation, positioning 'Thinking with Video' as a potential unified reasoning paradigm.
02 From the paper
Abstract
The "Thinking with Text" and "Thinking with Images" paradigms significantly improve the reasoning abilities of large language models (LLMs) and Vision-Language Models (VLMs). However, these paradigms have inherent limitations. (1) Images capture only single moments and fail to represent dynamic processes or continuous changes, and (2) The separation of text and vision as distinct modalities, which hinders unified multimodal understanding and generation. Therefore, we propose "Thinking with Video", a new paradigm that leverages video generation models such as Sora-2 to use video frames as a unified medium for multimodal reasoning. To support this exploration, we developed the Video Thinking Benchmark (VideoThinkBench), which covers both vision-centric tasks (e.g., Eyeballing Puzzles) and text-centric tasks (e.g., GSM8K and MMMU). Our evaluation on VideoThinkBench establishes Sora-2 as a capable reasoner. On vision-centric tasks, Sora-2 is comparable to state-of-the-art (SOTA) VLMs, and even surpasses GPT-5 by 10% on eyeballing puzzles. On text-centric tasks, Sora-2 achieves 92% accuracy on MATH, and 69.2% accuracy on MMMU. Furthermore, we systematically analyze the source of these abilities. We also find that self-consistency and in-context learning can improve Sora-2's performance. In summary, our findings show that the video generation model is the potential unified multimodal understanding and generation model, positioning "Thinking with Video" as a potential unified multimodal reasoning paradigm.