Paper 2511.15065
Reasoning via Video: The First Evaluation of Video Models' Reasoning Abilities through Maze-Solving Tasks
- Published
- Nov 2025
- Research lab
- Independent
- Citations
- 15
- GitHub
- 2 stars
01 In brief
Summary
This paper introduces VR-Bench, a benchmark for evaluating the reasoning abilities of video generation models through maze-solving tasks.
It comprises 7,920 procedurally generated videos across five maze types (Regular, Irregular, 3D, Trapfield, Sokoban) with varying difficulty and textures.
The authors propose a 'reasoning via video' paradigm, where models solve tasks by generating frame sequences rather than text.
They fine-tune the open-source video model Wan2.2-TI2V-5B (resulting in Wan-R1) using supervised fine-tuning (SFT) on VR-Bench.
Results show that Wan-R1 outperforms leading VLMs and closed-source video models on metrics like Exact Match, Success Rate, Precision Rate, and Step Deviation, achieving perfect Success Rate on Trapfield and 3D mazes.
The model also demonstrates strong generalization to unseen difficulties, textures, and maze types.
Additionally, test-time scaling (sampling multiple rollouts) improves performance by 10–20%, analogous to self-consistency in LLMs.
The study highlights that video-based reasoning scales better with task complexity than text-based reasoning, as VLMs degrade more sharply on harder mazes.
02 From the paper
Abstract
Video Models have achieved remarkable success in high-fidelity video generation with coherent motion dynamics. Analogous to the development from text generation to text-based reasoning in language modeling, the development of video models motivates us to ask: Can video models reason via video generation? Compared with the discrete text corpus, video grounds reasoning in explicit spatial layouts and temporal continuity, which serves as an ideal substrate for spatial reasoning. In this work, we explore the reasoning via video paradigm and introduce VR-Bench -- a comprehensive benchmark designed to systematically evaluate video models' reasoning capabilities. Grounded in maze-solving tasks that inherently require spatial planning and multi-step reasoning, VR-Bench contains 7,920 procedurally generated videos across five maze types and diverse visual styles. Our empirical analysis demonstrates that SFT can efficiently elicit the reasoning ability of video model. Video models exhibit stronger spatial perception during reasoning, outperforming leading VLMs and generalizing well across diverse scenarios, tasks, and levels of complexity. We further discover a test-time scaling effect, where diverse sampling during inference improves reasoning reliability by 10--20%. These findings highlight the unique potential and scalability of reasoning via video for spatial reasoning tasks.