README.md

---
title: README
emoji: 🔥
colorFrom: blue
colorTo: red
sdk: static
pinned: false
---
![main-figure](main-figure.png)
**Orsta** is a family of high-performance Vision-Language Models (7B–32B) trained on 8 diverse tasks including detection, grounding, math, and visual puzzles. These models are optimized for both **visual perception** and **reasoning**. They are trained using the [V-Triune](https://github.com/MiniMax-AI/One-RL-to-See-Them-All) framework, a unified RL system that streamlines multi-task learning across vision-language domains. Orsta delivers strong performance, with improvements ranging from **+2.1** to an impressive **+14.1** across its various 7B and 32B model variants, with performance benefits extending to a wide range of downstream tasks.

Explore our models, tasks, and results in the [technical report](https://github.com/MiniMax-AI/One-RL-to-See-Them-All/blob/main/MiniMax-One-RL-to-See-Them-All-v250523.pdf).