mlsanityData to 7 Oct 2026

Thinking With Videos: Multimodal Tool-Augmented Reinforcement Learning for Long Video Reasoning

arXiv 2508.04416 ↗First posted 6 Aug 2025cs.CVcited by 88

On the map: LLM reasoning → Multimodal reasoning

also in 1 more
Abstract (arXiv)

The video reasoning ability of multimodal large language models (MLLMs) is crucial for downstream tasks like video question answering and temporal grounding. While recent approaches have explored text-based chain-of-thought (CoT) reasoning for MLLMs, these methods often suffer from limited cross-modal interaction and increased hallucination, especially with longer videos or reasoning chains. To address these challenges, we propose Video Intelligence via Tool-Augmented Learning (VITAL), a novel end-to-end agentic video reasoning framework. With a visual toolbox, the model can densely sample new video frames on demand and generate multimodal CoT for precise long video reasoning. We observe that temporal grounding and question answering are mutually beneficial for video understanding tasks. Therefore, we construct two high-quality multi-task video reasoning datasets MTVR-CoT-72k for supervised fine-tuning and MTVR-RL-110k for reinforcement learning. Moreover, we propose a Difficulty-aware Group Relative Policy Optimization algorithm (DGRPO) to mitigate difficulty imbalance in multi-task reinforcement learning. Extensive experiments on 11 challenging video understanding benchmarks demonstrate the advanced reasoning ability of VITAL, outperforming existing methods in video question answering and temporal grounding tasks, especially in long video scenarios. Code is available at https://zhang9302002.github.io/thinkingwithvideos-page/.

Key facts. Cited by 88 papers in the map (data through 7 Oct 2026). Placed in Multimodal reasoning, one of 17 directions in LLM reasoning. Builds on 81 of its 95 references that are in the map. First posted on arXiv 6 Aug 2025.

Builds on

81 of 95 references.

Closest ancestors · 23

Same direction, newest first

  1. 2025Scaling RL to Long Videos10 Jul
  2. 2025VGR: Visual Grounded Reasoning13 Jun
  3. 2025Reinforcement Learning Tuning for VideoLLMs: Reward Design and Data Efficiency2 Jun
  4. 2025Video-Holmes: Can MLLM Think Like Holmes for Complex Video Reasoning?27 May
  5. 2025UniVG-R1: Reasoning Guided Universal Visual Grounding with Reinforcement Learning20 May
  6. 2025VideoRFT: Incentivizing Video Reasoning Capability in MLLMs via Reinforced Fine-Tuning18 May
  7. 2025OpenThinkIMG: Learning to Think with Images via Visual Tool Reinforcement Learning13 May
  8. 2025VLM-R1: A Stable and Generalizable R1-style Large Vision-Language Model10 Apr
  9. 2025Kimi-VL Technical Report10 Apr
  10. 2025VideoChat-R1: Enhancing Spatio-Temporal Perception via Reinforcement Fine-Tuning9 Apr
  11. 2025Video-R1: Reinforcing Video Reasoning in MLLMs27 Mar
  12. 2025LMM-R1: Empowering 3B LMMs with Strong Reasoning Abilities Through Two-Stage Rule-Based RL10 Mar
11 more
  1. 2025Vision-R1: Incentivizing Reasoning Capability in Multimodal Large Language Models9 Mar
  2. 2025Seg-Zero: Reasoning-Chain Guided Segmentation via Cognitive Reinforcement9 Mar
  3. 2025Visual-RFT: Visual Reinforcement Fine-Tuning3 Mar
  4. 2025Imagine while Reasoning in Space: Multimodal Visualization-of-Thought13 Jan
  5. 2025ReFocus: Visual Editing as a Chain of Thought for Structured Image Understanding9 Jan
  6. 2024Enhancing Video-LLM Reasoning via Agent-of-Thoughts Distillation2 Dec
  7. 2024Visual Agents as Fast and Slow Thinkers16 Aug
  8. 2024ReXTime: A Benchmark Suite for Reasoning-Across-Time in Videos27 Jun
  9. 2024Visual Sketchpad: Sketching as a Visual Chain of Thought for Multimodal Language Models13 Jun
  10. 2024Video-of-Thought: Step-by-Step Video Reasoning from Perception to Cognition7 May
  11. 2023Visual ChatGPT: Talking, Drawing and Editing with Visual Foundation Models8 Mar

Other references · 58

  1. 2025Qwen3 Technical Reportcited by 8,195
  2. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,146
  3. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,767
  4. 2025Qwen2.5-VL Technical Reportcited by 5,418
  5. 2024GPT-4o System Cardcited by 4,467
  6. 2025Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilitiescited by 3,655
  7. 2024LLaVA-OneVision: Easy Visual Task Transfercited by 2,669
  8. 2024HybridFlow: A Flexible and Efficient RLHF Frameworkcited by 1,980
  9. 2024Video-MME: The First-Ever Comprehensive Evaluation Benchmark of Multi-modal LLMs in Video Analysiscited by 1,375
  10. 2025Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learningcited by 1,371
  11. 2017Dense-Captioning Events in Videoscited by 789
  12. 2017TALL: Temporal Activity Localization via Language Querycited by 581
38 more
  1. 2024Thinking in Space: How Multimodal Large Language Models See, Remember, and Recall Spacescited by 580
  2. 2023LLaMA-VID: An Image is Worth 2 Tokens in Large Language Modelscited by 539
  3. 2024Executable Code Actions Elicit Better LLM Agentscited by 537
  4. 2023MovieChat: From Dense Token to Sparse Memory for Long Video Understandingcited by 536
  5. 2024Long Context Transfer from Language to Visioncited by 502
  6. 2019CLEVRER: CoLlision Events for Video REpresentation and Reasoningcited by 475
  7. 2023V*: Guided Visual Search as a Core Mechanism in Multimodal LLMscited by 459
  8. 2025ReTool: Reinforcement Learning for Strategic Tool Use in LLMscited by 377
  9. 2017Temporal Relational Reasoning in Videoscited by 324
  10. 2023Driving into the Future: Multiview Visual Forecasting and Planning with World Model for Autonomous Drivingcited by 309
  11. 2024LongVILA: Scaling Long-Context Visual Language Models for Long Videoscited by 303
  12. 2025DeepResearcher: Scaling Deep Research via Reinforcement Learning in Real-world Environmentscited by 268
  13. 2023GPT4Tools: Teaching Large Language Model to Use Tools via Self-instructioncited by 245
  14. 2025Video-MMMU: Evaluating Knowledge Acquisition from Multi-Discipline Professional Videoscited by 225
  15. 2023Self-Chained Image-Language Model for Video Localization and Question Answeringcited by 200
  16. 2023LLaVA-Plus: Learning to Use Tools for Creating Multimodal Agentscited by 188
  17. 2025ToRL: Scaling Tool-Integrated RLcited by 166
  18. 2024MotionLCM: Real-time Controllable Motion Generation via Latent Consistency Modelcited by 164
  19. 2024GenAD: Generalized Predictive Model for Autonomous Drivingcited by 150
  20. 2021Towards Long-Form Video Understandingcited by 146
  21. 2024LongVLM: Efficient Long Video Understanding via Large Language Modelscited by 146
  22. 2025Time-R1: Post-Training Large Vision Language Model for Temporal Video Groundingcited by 138
  23. 2024AuroraCap: Efficient, Performant Video Detailed Captioning and a New Benchmarkcited by 137
  24. 2023Can I Trust Your Answer? Visually Grounded Video Question Answeringcited by 133
  25. 2025MMVU: Measuring Expert-Level Multi-Discipline Video Understandingcited by 131
  26. 2025Flash-VStream: Efficient Real-Time Understanding for Long Video Streamscited by 78
  27. 2023Vista-LLaMA: Reducing Hallucination in Video Language Models via Equal Distance to Visual Tokenscited by 70
  28. 2025Traceable Evidence Enhanced Visual Grounded Reasoning: Evaluation and Methodologycited by 67
  29. 2024Number it: Temporal Grounding Videos like Flipping Mangacited by 65
  30. 2025DeepVideo-R1: Video Reinforcement Fine-Tuning via Difficulty-aware Regressive GRPOcited by 57
  31. 2022MIST: Multi-modal Iterative Spatial-Temporal Transformer for Long-form Video Question Answeringcited by 55
  32. 2025START: Self-taught Reasoner with Toolscited by 47
  33. 2025DyFo: A Training-Free Dynamic Focus Visual Search for Enhancing LMMs in Fine-Grained Visual Understandingcited by 47
  34. 2024Context-Guided Spatio-Temporal Video Groundingcited by 45
  35. 2023VidChapters-7M: Video Chapters at Scalecited by 39
  36. 2025ARC-Hunyuan-Video-7B: Structured Video Comprehension of Real-World Shortscited by 36
  37. 2024Hermes 3 Technical Reportcited by 33
  38. 2024LOGO: A Long-Form Video Dataset for Group Action Quality Assessmentcited by 30

50 most cited of 58 All 58 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 29; 2026: 59.

88 papers.

Sort:

By area:

  1. 2025LongVT: Incentivizing "Thinking with Long Videos" via Native Tool Callingcited by 58
  2. 2025OneThinker: All-in-one Reasoning Model for Image and Videocited by 52
  3. 2025Open-o3-Video: Grounded Video Reasoning with Explicit Spatio-Temporal Evidencecited by 43
  4. 2025Thinking with Video: Video Generation as a Promising Multimodal Reasoning Paradigmcited by 41
  5. 2025MM-BrowseComp: A Comprehensive Benchmark for Multimodal Browsing Agentscited by 39
  6. 2025LOVE-R1: Advancing Long Video Understanding with an Adaptive Zoom-in Mechanism via Multi-Step Reasoningcited by 27
  7. 2025Active Video Perception: Iterative Evidence Seeking for Agentic Long Video Understandingcited by 26
  8. 2025Video-STAR: Reinforcing Open-Vocabulary Action Recognition with Toolscited by 23
  9. 2025ReWatch-R1: Boosting Complex Video Reasoning in Large Vision-Language Models through Agentic Data Synthesiscited by 22
  10. 2026Embed-RL: Reinforcement Learning for Reasoning-Driven Multimodal Embeddingscited by 22
  11. 2025AdaTooler-V: Adaptive Tool-Use for Images and Videoscited by 20
  12. 2026Video-o3: Native Interleaved Clue Seeking for Long Video Multi-Hop Reasoningcited by 20
38 more
  1. 2026VideoAuto-R1: Video Auto Reasoning via Thinking Once, Answering Twicecited by 19
  2. 2025SpaceVista: All-Scale Visual Spatial Reasoning from mm to kmcited by 18
  3. 2025DiffThinker: Towards Generative Multimodal Reasoning with Diffusion Modelscited by 18
  4. 2026PyVision-RL: Forging Open Agentic Vision Models via RLcited by 14
  5. 2025ViPER: Empowering the Self-Evolution of Visual Perception Abilities in Vision-Language Modelcited by 11
  6. 2025Thinking With Bounding Boxes: Enhancing Spatio-Temporal Video Grounding via Reinforcement Fine-Tuningcited by 11
  7. 2025Video-CoM: Interactive Video Reasoning via Chain of Manipulationscited by 9
  8. 2026VISD: Enhancing Video Reasoning via Structured Self-Distillationcited by 9
  9. 2025Reinforcement Learning for Large Model: A Surveycited by 8
  10. 2025Video-LMM Post-Training: A Deep Dive into Video Reasoning with Large Multimodal Modelscited by 8
  11. 2025VG-Refiner: Towards Tool-Refined Referring Grounded Reasoning via Agentic Reinforcement Learningcited by 8
  12. 2026Agentic Environment Engineering for Large Language Models: A Survey of Environment Modeling, Synthesis, Evaluation, and Applicationcited by 8
  13. 2026VideoThinker: Building Agentic VideoLLMs with LLM-Guided Tool Reasoningcited by 7
  14. 2026Think While Watching: Online Streaming Segment-Level Memory for Multi-Turn Video Reasoning in Multimodal Large Language Modelscited by 7
  15. 2025Agentic Video Intelligence: A Flexible Framework for Advanced Video Exploration and Understandingcited by 6
  16. 2025VideoChat-M1: Collaborative Policy Planning for Video Understanding via Multi-Agent Reinforcement Learningcited by 6
  17. 2026Thinking with Deltas: Incentivizing Reinforcement Learning via Differential Visual Reasoning Policycited by 6
  18. 2026MedSAM-Agent: Empowering Interactive Medical Image Segmentation with Multi-turn Agentic Reinforcement Learningcited by 6
  19. 2026VideoTemp-o3: Harmonizing Temporal Grounding and Video Understanding in Agentic Thinking-with-Videoscited by 6
  20. 2026EndoCoT: Scaling Endogenous Chain-of-Thought Reasoning in Diffusion Modelscited by 6
  21. 2026VideoZeroBench: Probing the Limits of Video MLLMs with Spatio-Temporal Evidence Verificationcited by 6
  22. 2025TimeScope: Towards Task-Oriented Temporal Grounding In Long Videoscited by 5
  23. 2025DeepSport: A Multimodal Large Language Model for Comprehensive Sports Video Reasoning via Agentic Reinforcement Learningcited by 5
  24. 2025VideoPerceiver: Enhancing Fine-Grained Temporal Perception in Video Multimodal Large Language Modelscited by 5
  25. 2026GraphThinker: Reinforcing Temporally Grounded Video Reasoning with Event Graph Thinkingcited by 5
  26. 2026ParaVT: Taming the Tool Prior Paradox for Parallel Tool Use in Agentic Video Reinforcement Learningcited by 5
  27. 2026LatentOmni: Rethinking Omni-Modal Understanding via Unified Audio-Visual Latent Reasoningcited by 5
  28. 2026VideoSearcher: Empowering Video Deep Research with Multi-Tool Agentic Reasoning via Reinforcement Learningcited by 5
  29. 2025Select Less, Reason More: Prioritizing Evidence Purity for Video Reasoningcited by 4
  30. 2025ViSS-R1: Self-Supervised Reinforcement Video Reasoningcited by 4
  31. 2026StreamingClaw Technical Reportcited by 4
  32. 2025Delayed Bidirectional Alignment via Disentangled Audio Semantics for Audio-Visual Segmentationcited by 3
  33. 2026VisionCoach: Reinforcing Grounded Video Reasoning via Visual-Perception Promptingcited by 3
  34. 2026TIR-Agent: Training an Explorative and Efficient Agent for Image Restorationcited by 3
  35. 2026Omni-MMSI: Toward Identity-attributed Social Interaction Understandingcited by 3
  36. 2026Watch, Remember, Reason: Human-View Video Understanding with MLLMscited by 3
  37. 2026Native Active Perception as Reasoning for Omni-Modal Understandingcited by 3
  38. 2026Latent Visual Cache for Video Reasoningcited by 3

50 most cited of 88 All 88 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2025Video-Thinker: Sparking "Thinking with Videos" via Reinforcement Learning
  2. 2025Video-R2: Reinforcing Consistent and Grounded Reasoning in Multimodal Language Models
  3. 2025Exploring the Effect of Reinforcement Learning on Video Understanding: Insights from SEED-Bench-R1
  4. 2025FrameThinker: Learning to Think with Long Videos via Multi-Turn Frame Spotlighting
  5. 2026VideoTIR: Accurate Understanding for Long Videos with Efficient Tool-Integrated Reasoning
  6. 2025Reinforcing Video Reasoning with Focused Thinking
  7. 2026AdaThinkV: Adaptive Thinking for Token-Efficient Video Reasoning
  8. 2026A Skill-augmented Agentic Framework and Benchmark for Multi-Video Understanding