# Osilly/Vision-R1

[ICLR2026] This is the first paper to explore how  to effectively use R1-like RL for MLLMs and introduce Vision-R1, a reasoning MLLM that leverages cold-start initialization and RL training to incentivize reasoning capability.

Repository: https://github.com/Osilly/Vision-R1
Canonical: https://ross.abutalabs.com/products/vision-r1
Language: Python
License Family: other
Last push: 2026-03-20T11:58:18+00:00

## Health v2 (maintenance only)
Score: 53/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 73, release rhythm 35, longevity 41
- inputs: {"age_days": 576, "days_push": 166, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1571, forks 27 (observed 2026-08-28T04:05:05.390902+00:00)

## What it is
Vision-R1 is the official repository for a research paper on training reasoning-capable multimodal large language models using R1-like reinforcement learning with cold-start initialization. It provides training code, cold-start and RL datasets, and released model weights (7B to 72B) that improve mathematical visual reasoning benchmarks.

## Use cases
- train a multimodal LLM with reinforcement learning for visual reasoning
- reproduce Vision-R1 reasoning MLLM results
- download cold-start and RL datasets for multimodal reasoning training
- improve math reasoning in vision-language models
- fine-tune Qwen2.5-VL with R1-style RL
- research reasoning capability in MLLMs

## When to choose
- you need a reasoning-enhanced multimodal model for math/visual tasks
- you want to replicate or extend R1-style RL training for MLLMs
- you need released checkpoints and datasets for multimodal reasoning research

## When to avoid
- you need a production-ready inference service rather than research code
- you require a permissively licensed project (no license is specified)
- you only need general-purpose VLM inference without reasoning training

## Facets
- artifact type: learning-resource
- maturity: active
- function: machine-learning, llm-training, deep-learning
- domain: large-language-models, machine-learning, computer-vision, artificial-intelligence
- platform: python
- tags: multimodal, reinforcement-learning, reasoning, vision-language-model, research-paper, cold-start, rlhf, math-reasoning, gpu, linux

## Member repositories
- Osilly/Vision-R1 (main) score 53

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:05.390902+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:58:25.454861+00:00, confidence not recorded.
  - readme: https://github.com/Osilly/Vision-R1 (fetched 2026-08-28T04:05:05.390902+00:00, sha 25309a71d704)
- Data as of 2026-08-30T08:39:29.467469+00:00.