We introduce 'Thinking with Video', a new paradigm leveraging video generation for multimodal reasoning. Our VideoThinkBench shows that Sora-2 surpasses GPT5 by 10% on eyeballing puzzles and reaches 69% accuracy on MMMU.
[CVPR 2026] Machine Mental Imagery: Empower Multimodal Reasoning with Latent Visual Tokens
Official code for NeurIPS 2025 paper "GRIT: Teaching MLLMs to Think with Images"
[CVPR 2026] OpenMMReasoner: Pushing the Frontiers for Multimodal Reasoning with an Open and General Recipe
Official repository for "Visual Generation Unlocks Human-Like Reasoning through Multimodal World Models", https://arxiv.org/abs/2601.19834
Open-source financial multimodal LLM for document understanding, chart/table reasoning, numerical reasoning, and financial analysis, with SFT, dual-branch RL, MOPD distillation, and public benchmark evaluation.
Latest Advances on (RL based) Multimodal Reasoning and Generation in Multimodal LLMs
[CVPR2026] Official codebase for the paper "Reasoning Within the Mind: Dynamic Multimodal Interleaving in Latent Space"
A Comprehensive Survey on Evaluating Reasoning Capabilities in Multimodal Large Language Models.
[ICLR 2026 Oral & ICML 2026] Generative Universal Verifier as Multimodal Meta-Reasoner
ThinkJEPA: Empowering Latent World Models with Large Vision-Language Reasoning Model
Rui Qian, Xin Yin, Dejing Dou†: Reasoning to Attend: Try to Understand How <SEG> Token Works (CVPR 2025)
[NeurIPS 2025] Official implementation for the paper "SeePhys: Does Seeing Help Thinking? -- Benchmarking Vision-Based Physics Reasoning"
[CVPR 2026] AdaptVision: Efficient Vision-Language Models via Adaptive Visual Acquisition
WeThink: Toward General-purpose Vision-Language Reasoning via Reinforcement Learning
[ICLR 2026] SceneCOT: Eliciting Grounded Chain-of-Thought Reasoning in 3D Scenes
HoloGeo: Evidence-driven VLM for robust image geo-localization and landmark-bias mitigation.
[ICLR 2026]🌴 ARES is an open-source framework for adaptive multimodal reasoning, featuring a two-stage pipeline—Adaptive Cold-Start and Entropy-Shaped Policy Optimization—to balance reasoning depth and efficiency.
Project COLON-X (Shaping the neXt frontier in intelligent COLONoscopy)
Repository to train multimodal latent reasoning tokens for Qwen 2.5 VL.
To associate your repository with the multimodal-reasoning topic, visit your repo's landing page and select "manage topics."