mlsanityData to 14 Oct 2026

Video models are zero-shot learners and reasoners

arXiv 2509.20328 ↗First posted 24 Sep 2025cs.LG cs.AI cs.CVcited by 220

On the map: LLM reasoning → Multimodal reasoning

Abstract (arXiv)

The remarkable zero-shot capabilities of Large Language Models (LLMs) have propelled natural language processing from task-specific models to unified, generalist foundation models. This transformation emerged from simple primitives: large, generative models trained on web-scale data. Curiously, the same primitives apply to today's generative video models. Could video models be on a trajectory towards general-purpose vision understanding, much like LLMs developed general-purpose language understanding? We demonstrate that Veo 3 can solve a broad variety of tasks it wasn't explicitly trained for: segmenting objects, detecting edges, editing images, understanding physical properties, recognizing object affordances, simulating tool use, and more. These abilities to perceive, model, and manipulate the visual world enable early forms of visual reasoning like maze and symmetry solving. Veo's emergent zero-shot capabilities indicate that video models are on a path to becoming unified, generalist vision foundation models.

Key facts. Cited by 220 papers in the map (data through 14 Oct 2026). Placed in Multimodal reasoning, one of 17 directions in LLM reasoning. Builds on 66 of its 88 references that are in the map. First posted on arXiv 24 Sep 2025.

Builds on

66 of 88 references.

Closest ancestors · 2

Same direction, newest first

  1. 2025VEGGIE: Instructional Editing and Reasoning Video Concepts with Grounded Generation18 Mar
  2. 2024KiVA: Kid-inspired Visual Analogies for Testing Large Multimodal Models25 Jul

Other references · 64

  1. 2020Language Models are Few-Shot Learnerscited by 32,419
  2. 2022Training language models to follow instructions with human feedbackcited by 15,971
  3. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,459
  4. 2023Segment Anythingcited by 8,234
  5. 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,708
  6. 2015You Only Look Once: Unified, Real-Time Object Detectioncited by 4,852
  7. 2022Large Language Models are Zero-Shot Reasonerscited by 4,374
  8. 2025Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilitiescited by 3,796
  9. 2023Self-Refine: Iterative Refinement with Self-Feedbackcited by 3,442
  10. 2024SAM 2: Segment Anything in Images and Videoscited by 2,863
  11. 2022Emergent Abilities of Large Language Modelscited by 2,512
  12. 2021Pre-train, Prompt, and Predict: A Systematic Survey of Prompting Methods in Natural Language Processingcited by 2,234
38 more
  1. 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameterscited by 1,986
  2. 2020Shortcut Learning in Deep Neural Networkscited by 1,969
  3. 2018ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustnesscited by 1,432
  4. 2019LVIS: A Dataset for Large Vocabulary Instance Segmentationcited by 1,217
  5. 2024The AI Scientist: Towards Fully Automated Open-Ended Scientific Discoverycited by 1,071
  6. 2022Large Language Models Are Human-Level Prompt Engineerscited by 1,045
  7. 2025V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planningcited by 809
  8. 2023Objaverse-XL: A Universe of 10M+ 3D Objectscited by 774
  9. 2023Instruction Tuning with GPT-4cited by 686
  10. 2023Text2Video-Zero: Text-to-Image Diffusion Models are Zero-Shot Video Generatorscited by 670
  11. 2024A Survey on Large Language Models for Code Generationcited by 658
  12. 2018Taskonomy: Disentangling Task Transfer Learningcited by 634
  13. 2015Holistically-Nested Edge Detectioncited by 592
  14. 2024YOLOv11: An Overview of the Key Architectural Enhancementscited by 425
  15. 2025Agent Laboratory: Using LLM Agents as Research Assistantscited by 425
  16. 2024OmniGen: Unified Image Generationcited by 394
  17. 2023Emu Edit: Precise Image Editing via Recognition and Generation Taskscited by 324
  18. 2022Images Speak in Images: A Generalist Painter for In-Context Visual Learningcited by 318
  19. 2024How Far is Video Generation from World Model: A Physical Law Perspectivecited by 212
  20. 2023Dreamix: Video Diffusion Models are General Video Editorscited by 191
  21. 2024Motion Prompting: Controlling Video Generation with Motion Trajectoriescited by 178
  22. 2025Do generative video models understand physical principles?cited by 161
  23. 2023Multilingual Machine Translation with Large Language Models: Empirical Results and Analysiscited by 160
  24. 2023Text-to-Image Diffusion Models are Zero-Shot Classifierscited by 143
  25. 2019Dense Extreme Inception Network: Towards a Robust CNN Model for Edge Detectioncited by 104
  26. 2024Video as the New Language for Real-World Decision Makingcited by 87
  27. 2024Understanding the Limits of Vision Language Models Through the Lens of the Binding Problemcited by 75
  28. 2025Intuitive physics understanding emerges from self-supervised pretraining on natural videoscited by 70
  29. 2014Fast Edge Detection Using Structured Forestscited by 65
  30. 2023Physion++: Evaluating Physical Scene Understanding that Requires Online Inference of Different Physical Propertiescited by 54
  31. 2025Force Prompting: Video Generation Models Can Learn and Generalize Physics-based Control Signalscited by 53
  32. 2023Intriguing properties of generative classifierscited by 52
  33. 2025Evaluating Newtonian Mechanics in Video Generative Models with Real Physical Systemscited by 48
  34. 2024Diffusion Model-Based Video Editing: A Surveycited by 44
  35. 2024One Diffusion to Generate Them Allcited by 41
  36. 2021Dense Extreme Inception Network for Edge Detectioncited by 39
  37. 2025VisualCloze: A Universal Image Generation Framework via Visual In-Context Learningcited by 39
  38. 2023Visual cognition in multimodal large language modelscited by 38

50 most cited of 64 All 64 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 57; 2026: 163.

220 papers.

Sort:

By area:

  1. 2025WorldPlay: Towards Long-Term Geometric Consistency for Real-Time Interactive World Modelingcited by 170
  2. 2026Cosmos 3: Omnimodal World Models for Physical AIcited by 165
  3. 2025mimic-video: Video-Action Models for Generalizable Robot Control Beyond VLAscited by 142
  4. 2026DreamDojo: A Generalist Robot World Model from Large-Scale Human Videoscited by 135
  5. 2025RELIC: Interactive Video World Model with Long-Horizon Memorycited by 96
  6. 2025Reward Forcing: Efficient Streaming Video Generation with Rewarded Distribution Matching Distillationcited by 86
  7. 2025Kling-Omni Technical Reportcited by 71
  8. 2025Hunyuan-GameCraft-2: Instruction-following Interactive Game World Modelcited by 61
  9. 2025End-to-End Training for Autoregressive Video Diffusion via Self-Resamplingcited by 52
  10. 2026Zooming without Zooming: Region-to-Image Distillation for Fine-Grained Multimodal Perceptioncited by 50
  11. 2025DriveLaW:Unifying Planning and Video Generation in a Latent Driving Worldcited by 47
  12. 2025Are Video Models Ready as Zero-Shot Reasoners? An Empirical Study with the MME-CoF Benchmarkcited by 43
38 more
  1. 2026MOVA: Towards Scalable and Synchronized Video-Audio Generationcited by 43
  2. 2026Image Generators are Generalist Vision Learnerscited by 43
  3. 2025Thinking with Video: Video Generation as a Promising Multimodal Reasoning Paradigmcited by 42
  4. 2025NovaFlow: Zero-Shot Manipulation via Actionable Flow from Generated Videoscited by 40
  5. 2026Beyond Language Modeling: An Exploration of Multimodal Pretrainingcited by 37
  6. 2025PAI-Bench: A Comprehensive Benchmark For Physical AIcited by 34
  7. 2026A Very Big Video Reasoning Suitecited by 34
  8. 2026Rethinking Video Generation Model for the Embodied Worldcited by 33
  9. 2025VChain: Chain-of-Visual-Thought for Reasoning in Video Generationcited by 32
  10. 2026Interactive World Simulator for Robot Policy Training and Evaluationcited by 31
  11. 2026LLM-as-a-Verifier: A General-Purpose Verification Frameworkcited by 30
  12. 2025ChronoEdit: Towards Temporal Reasoning for Image Editing and World Simulationcited by 29
  13. 2026VINO: A Unified Visual Generator with Interleaved OmniModal Contextcited by 26
  14. 2025What about gravity in video generation? Post-Training Newton's Laws with Verifiable Rewardscited by 25
  15. 2026VGGRPO: Towards World-Consistent Video Generation with 4D Latent Rewardcited by 24
  16. 2025Reasoning via Video: The First Evaluation of Video Models' Reasoning Abilities through Maze-Solving Taskscited by 22
  17. 2025V-ReasonBench: Toward Unified Reasoning Benchmark Suite for Video Generation Modelscited by 22
  18. 2025TiViBench: Benchmarking Think-in-Video Reasoning for Video Generative Modelscited by 19
  19. 2025Can World Simulators Reason? Gen-ViRe: A Generative Visual Reasoning Benchmarkcited by 19
  20. 2025DiffThinker: Towards Generative Multimodal Reasoning with Diffusion Modelscited by 19
  21. 2026OmniForcing: Unleashing Real-time Joint Audio-Visual Generationcited by 16
  22. 2025A Practical Guide for Designing, Developing, and Deploying Production-Grade Agentic AI Workflowscited by 15
  23. 2025Vidarc: Embodied Video Diffusion Model for Closed-loop Controlcited by 15
  24. 2026Self-Refining Video Samplingcited by 15
  25. 2026Gamma-World: Generative Multi-Agent World Modeling Beyond Two Playerscited by 15
  26. 2026Walk through Paintings: Egocentric World Models from Internet Priorscited by 14
  27. 2026Dex4D: Task-Agnostic Point Track Policy for Zero-Shot Dexterous Manipulationcited by 14
  28. 2026Demystifying Video Reasoningcited by 14
  29. 2026Video Generation Models are General-Purpose Vision Learnerscited by 14
  30. 2025Time-to-Move: Training-Free Motion Controlled Video Generation via Dual-Clock Denoisingcited by 13
  31. 2025RULER-Bench: Probing Rule-based Reasoning Abilities of Next-level Video Generation Models for Vision Foundation Intelligencecited by 13
  32. 2025In Pursuit of Pixel Supervision for Visual Pre-trainingcited by 13
  33. 2026Robotic Video World Models: A Survey of Applications, Research Challenges, Future Directionscited by 13
  34. 2026LOME: Learning Human-Object Manipulation with Action-Conditioned Egocentric World Modelcited by 13
  35. 2025First Frame Is the Place to Go for Video Content Customizationcited by 12
  36. 2025MMGR: Multi-Modal Generative Reasoning Benchmark and Evaluationcited by 12
  37. 2025How Far Are Surgeons from Surgical World Models? A Pilot Study on Zero-shot Surgical Video Generation with Expert Assessmentcited by 11
  38. 2026Scaling Behavior Cloning Improves Causal Reasoning: An Open Model for Real-Time Video Game Playingcited by 11

50 most cited of 220 All 220 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2023Analyzing Zero-Shot Abilities of Vision-Language Models on Video Understanding Tasks
  2. 2025Vision Language Models: A Survey of 26K Papers
  3. 2025LAST: LeArning to Think in Space and Time for Generalist Vision-Language Models
  4. 2024Fewer Tokens and Fewer Videos: Extending Video Understanding Abilities in Large Vision-Language Models
  5. 2025PerceptionLM: Open-Access Data and Models for Detailed Visual Understanding
  6. 2024Explanatory Instructions: Towards Unified Vision Tasks Understanding and Zero-shot Generalization
  7. 2024Zero-Shot Visual Reasoning by Vision-Language Models: Benchmarking and Analysis
  8. 2025Learning to See Before Seeing: Demystifying LLM Visual Priors from Language Pre-training