Research map / Large language models / Reasoning and chain of thought
R1-VL: Learning to Reason with Multimodal Large Language Models via Step-wise Group Relative Policy Optimization
Jingyi Zhang, Jiaxing Huang, Huanjin Yao, Shunyu Liu, Xikun Zhang, Shijian Lu, Dacheng Tao
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 policy warm-up SFT (ablation) | Qwen2-VL-7B · MathVista | accuracy: 58.2 -> 61.2 | Improved |
| add StepRAR on top of warm-up (ablation) | StepGRPO (Qwen2-VL-7B) · MathVista | accuracy: 61.2 -> 62.4 | Improved |
| add StepRVR on top of warm-up (ablation) | StepGRPO (Qwen2-VL-7B) · MathVista | accuracy: 61.2 -> 61.9 | Improved |
| combine StepRVR with StepRAR (ablation) | StepGRPO (Qwen2-VL-7B) · MathVista | accuracy: 62.4 -> 63.5 | Improved |
| add outcome-level reward RL on top of warm-up (ablation) | StepGRPO (Qwen2-VL-7B) · MathVista | accuracy: 61.7 -> 62.3 | Improved |
| use step-wise reward instead of outcome-level reward (ablation) | StepGRPO (Qwen2-VL-7B) · MathVista | accuracy: 62.3 -> 63.5 | Improved |
| increase rollouts per question M from 2 to 4 (ablation) | StepGRPO R1-VL-7B · MathVista | accuracy: 62.5 -> 63.5 | Improved |
| increase rollouts per question M from 4 to 6 (ablation) | StepGRPO R1-VL-7B · MathVista | accuracy: 63.5 -> 63.7 | No clear effect |
| increase rollouts per question M from 4 to 5 (ablation) | StepGRPO R1-VL-7B · MathVista | accuracy: 63.5 -> 63.2 | No clear effect |
| train with StepGRPO instead of the base Qwen2-VL-2B | Qwen2-VL-2B · benchmark average | score: 37.5 -> 41.6 | Improved |
| train with StepGRPO instead of the base Qwen2-VL-7B | Qwen2-VL-7B · benchmark average | score: 48.7 -> 52.1 | Improved |
| train with StepGRPO on Qwen2.5-VL-7B | Qwen2.5-VL-7B · benchmark average | score: 55.5 -> 58.4 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Mulberry-7B — Mulberry: Empowering MLLM with o1-like Reasoning and Reflection via Collective Monte Carlo Tree Search (NeurIPS 2025) · better on 1 of 1 shared result
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
- SophiaVL-R1: Reinforcing MLLMs Reasoning with Thinking Reward (ICLR 2026 · similarity 0.91)
- RL Squeezes, SFT Expands: A Comparative Study of Reasoning LLMs (ICLR 2026 · similarity 0.90)
- Reinforcement Learning with Verifiable Rewards Implicitly Incentivizes Correct Reasoning in Base LLMs (ICLR 2026 · similarity 0.90)
- Reshaping Reasoning in LLMs: A Theoretical Analysis of RL Training Dynamics through Pattern Selection (ICLR 2026 · similarity 0.90)
- R-4B: Incentivizing General-Purpose Auto-Thinking in MLLMs via Bi-Mode Annealing and Reinforce Learning (CVPR 2026 · similarity 0.90)
- Accelerating RL for LLM Reasoning with Optimal Advantage Regression (NeurIPS 2025 · similarity 0.90)
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.
