Paper 2602.20159
A Very Big Video Reasoning Suite
- Published
- Feb 2026
- Research lab
- Independent
- Citations
- 19
- GitHub
- 225 stars
01 In brief
Summary
The paper introduces the Very Big Video Reasoning (VBVR) suite, comprising a large-scale dataset and benchmark for video reasoning.
VBVR-Dataset contains 2,015,000 images and 1,007,500 video clips across 200 tasks, organized by five cognitive faculties (Abstraction, Knowledge, Perception, Spatiality, Transformation), and is approximately 1000x larger than existing datasets.
VBVR-Bench uses rule-based, human-aligned scorers (Spearman's rho > 0.9) for reproducible evaluation.
Benchmarking shows proprietary models (Sora 2: 0.546, Veo 3.1: 0.480) outperform open-source ones (Wan2.2: 0.371), but all fall short of human performance (0.974).
Fine-tuning Wan2.2 on VBVR-Dataset yields VBVR-Wan2.2, achieving 0.685 overall (84.6% relative improvement).
Scaling studies show performance plateaus, with a persistent 15% gap between in-domain and out-of-domain performance.
Capability correlation analysis reveals positive coupling between Knowledge and Spatiality (rho=0.461) and negative correlations between Knowledge and Perception (rho=-0.757) and Abstraction and Transformation (rho=-0.641).
Qualitative analysis highlights emergent behaviors like controllable execution and multi-step strategies, but also limitations in long-horizon identity stability and process faithfulness.
The data, benchmark toolkit, and models are released at video-reason.com.
02 From the paper
Abstract
Rapid progress in video models has largely focused on visual quality, leaving their reasoning capabilities underexplored. Video reasoning grounds intelligence in spatiotemporally consistent visual environments that go beyond what text can naturally capture, enabling intuitive reasoning over spatiotemporal structure such as continuity, interaction, and causality. However, systematically studying video reasoning and its scaling behavior is hindered by the lack of large-scale training data. To address this gap, we introduce the Very Big Video Reasoning (VBVR) Dataset, an unprecedentedly large-scale resource spanning 200 curated reasoning tasks following a principled taxonomy and over one million video clips, approximately three orders of magnitude larger than existing datasets. We further present VBVR-Bench, a verifiable evaluation framework that moves beyond model-based judging by incorporating rule-based, human-aligned scorers, enabling reproducible and interpretable diagnosis of video reasoning capabilities. Leveraging the VBVR suite, we conduct one of the first large-scale scaling studies of video reasoning and observe early signs of emergent generalization to unseen reasoning tasks. Together, VBVR lays a foundation for the next stage of research in generalizable video reasoning. The data, benchmark toolkit, and models are publicly available at https://video-reason.com/ .