Vero-600k
Vero is a fully open reinforcement learning (RL) recipe for training and evaluating multi-task visual reasoning with vision-language models. This repository contains the Vero-600K dataset, a curation of 600K reinforcement learning samples from 59 datasets across 6 diverse visual reasoning categories.
Highlights
- Scale: 600K curated RL samples from 59 datasets.
- Diversity: Covers 6 broad categories: STEM Reasoning, Chart & OCR, Spatial & Action, Knowledge & Recognition, Grounding & Counting, and Instruction Following.
- Task-Routed Rewards: Designed to handle heterogeneous answer formats across diverse tasks.
- Open Recipe: Fully open release of models, training code, evaluation suite, and dataset.
Dataset Structure
The dataset is organized into six broad task categories:
- STEM reasoning
- Chart and OCR
- Spatial reasoning and action
- Knowledge and recognition
- Grounding, counting, and visual search
- Captioning and instruction following
For detailed dataset format, curation details, and reward routing metadata, see the Data Guide.
Sample Usage
To download and format the Vero-600k dataset for RL training using the official setup script:
# Clone the repository
git clone https://github.com/zlab-princeton/vero.git
cd vero
# Run the formatting script
python scripts/download_and_format_vero_600k.py
This script exports images into vero-rl/data/images/ and generates the training/validation .verl.jsonl files required for the Vero RL pipeline.
Models
Vero models are trained from various open-weight bases including Qwen3-VL, Qwen2.5-VL, and MiMo-VL.
| Model | Base model | Params |
|---|---|---|
Vero-Qwen3I-8B | Qwen3-VL-8B-Instruct | 8B |
Vero-Qwen3T-8B | Qwen3-VL-8B-Thinking | 8B |
Vero-MiMo-7B | MiMo-VL-7B-SFT | 7B |
Vero-Qwen25-7B | Qwen2.5-VL-7B-Instruct | 7B |
Citation
If you use this dataset or the Vero recipe in your research, please cite:
@article{sarch2026vero,
title = {Vero: An Open RL Recipe for General Visual Reasoning},
author = {Sarch, Gabriel and Cai, Linrong and Wang, Qunzhong and Wu, Haoyang and Danqi Chen and Zhuang Liu},
year = {2026},
journal = {arXiv preprint arXiv:2604.04917},
}
License
This dataset is released under the Apache License 2.0. Users should also review the licenses and usage terms of the underlying base models and any upstream datasets included in this curation.