Research map / Vision-language and multimodal / Multimodal LLMs
R1-Onevision: Advancing Generalized Multimodal Reasoning through Cross-Modal Formalization
Yi Yang, Xiaoxuan He, Hongkun Pan, Xiyan Jiang, Yan Deng, Xingtao Yang, Haoyu Lu, Dacheng Yin, Fengyun Rao, Minfeng Zhu, Bo Zhang, Wei Chen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add SFT on R1-Onevision data (ablation) | Qwen2.5-VL-7B · MathVision | accuracy: 25.4 -> 26.3 | Improved |
| add SFT on R1-Onevision data (ablation) | Qwen2.5-VL-7B · MathVerse (ALL) | accuracy: 43.6 -> 43.4 | No clear effect |
| add SFT on R1-Onevision data (ablation) | Qwen2.5-VL-7B · MathVerse (Vision Only) | accuracy: 38.2 -> 39.7 | Improved |
| add RL after SFT (ablation) | R1-Onevision-7B · MathVision | accuracy: 26.3 -> 29.9 | Improved |
| add RL after SFT (ablation) | R1-Onevision-7B · MathVerse (ALL) | accuracy: 43.4 -> 46.4 | Improved |
| add RL after SFT (ablation) | R1-Onevision-7B · MathVerse (Vision Only) | accuracy: 39.7 -> 40.0 | Improved |
| add SFT stage before RL (ablation) | R1-Onevision-7B · MathVision | accuracy: 28.0 -> 29.9 | Improved |
| apply SFT+RL pipeline to 3B base | Qwen2.5-VL-3B · MathVision | accuracy: 21.7 -> 23.7 | Improved |
| apply SFT+RL pipeline to 3B base | Qwen2.5-VL-3B · MathVerse (ALL) | accuracy: 34.7 -> 38.6 | Improved |
| apply SFT+RL pipeline to 3B base | Qwen2.5-VL-3B · MathVerse (Vision Only) | accuracy: 31.2 -> 35.5 | Improved |
| scale base model from 7B to 72B | Qwen2.5-VL · R1-Onevision-Bench (avg) | accuracy: 32.1 -> 52.0 | Improved |
| apply R1-Onevision training to 7B base | Qwen2.5-VL-7B · R1-Onevision-Bench (avg) | accuracy: 32.1 -> 36.2 | Improved |
Later papers that compared with this method
12 later accepted papers report R1-Onevision-7B as a baseline in their results tables; 9 of them report a better number on most of the results they share with it.
- Thinking in Uncertainty: Mitigating Hallucinations in MLRMs with Latent Entropy-Aware Decoding (CVPR 2026 · better on 8 of 8 shared results)
- Understanding and Mitigating Hallucinations in Multimodal Chain-of-Thought Models (CVPR 2026 · better on 3 of 3 shared results)
- Diagnosing and Mitigating Modality Interference in Multimodal Large Language Models (NeurIPS 2026 · better on 2 of 2 shared results)
- Does Seeing More Mean Knowing More? Mono-Anchored Advantage Normalization for Multi-Source Visual Reasoning (NeurIPS 2026 · better on 2 of 2 shared results)
- Vision-aligned Latent Reasoning for Multi-modal Large Language Model (ICML 2026 · better on 1 of 1 shared result)
- More Thought, Less Accuracy? On the Dual Nature of Reasoning in Vision-Language Models (ICLR 2026 · better on 1 of 1 shared result)
- Perceive, Interact, Reason: Building Tool-Augmented Visual Agents for Spatial Reasoning (NeurIPS 2026 · better on 1 of 1 shared result)
- GThinker: Towards General Multimodal Reasoning via Cue-Guided Rethinking (CVPR 2026 · better on 1 of 1 shared result)
4 more not listed.
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Vision-R1: Incentivizing Reasoning Capability in Multimodal Large Language Models (ICLR 2026 · similarity 0.93)
- OneThinker: All-in-one Reasoning Model for Image and Video (CVPR 2026 · similarity 0.93)
- ROVER: Benchmarking Reciprocal Cross-Modal Reasoning for Omnimodal Generation (ICLR 2026 · similarity 0.92)
- RPTS: Tree-Structured Reasoning Process Scoring for Faithful Multimodal Evaluation (AAAI 2026 · similarity 0.92)
- Perception-R1: Advancing Multimodal Reasoning Capabilities of MLLMs via Visual Perception Reward (ICLR 2026 · similarity 0.91)
- MMFineReason: Closing the Multimodal Reasoning Gap via Open Data-Centric Methods (NeurIPS 2026 · similarity 0.91)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
