MintAct: A Unified Visual Agent for Digital Environments (2B/4B/8B)

Forum topic · 小凯 · 2026-09-22

Summary

MintAct is a family of vision-language models (2B, 4B, and 8B parameters) that unifies three core agent capabilities—UI grounding, multi-step navigation across mobile, desktop, and web environments, and visual tool use—within a single model family. Through careful design of environments, data, and training recipes, MintAct matches the performance of per-domain specialist models across all these capabilities. The authors built a scalable environment and reinforcement learning infrastructure: hundreds of concurrent instances run across heterogeneous per-domain backends, serving both trajectory data collection and online RL. An asynchronous training framework gives explicit control over the cross-domain training distribution and remains stable under noisy environment feedback and off-policy drift. Experiments show state-of-the-art results among models of comparable size on multiple benchmarks, including OSWorld-Verified at 48.9. The paper is authored by Mingfei Gao, Rui Tian, Haiming Gang, Bohan Zhai, and colleagues, posted to arXiv (2609.22083) in September 2026.

Overview

Research area: Computer Vision (CV) Authors: Mingfei Gao, Rui Tian, Haiming Gang, Bohan Zhai, Le Zhang, Yuanzheng Gong, Di Feng, Ege Özsoy, Kaixin Ma, Vishwesh Kirthivasan, Oğuzhan Fatih Kar, Roman Bachmann, Anders Boesen Lindbo Larsen, Afshin Dehghan Release date: 2026-09-18 arXiv: 2609.22083

Abstract (translated)

We present MintAct, a family of vision-language models that unifies UI grounding, multi-step navigation across mobile, desktop, and web, and visual tool use, trained at 2B, 4B, and 8B scales. Through careful design of our environments, data, and training recipes, MintAct models match the performance of per-domain specialists across all of these capabilities. To enable this, we develop a scalable environment and reinforcement learning (RL) infrastructure. On the environment side, we host hundreds of concurrent instances across heterogeneous per-domain backends, serving both trajectory data collection and online RL. To enable efficient and scalable RL training, an asynchronous framework keeps explicit control over the cross-domain training distribution and remains stable under noisy environment feedback and off-policy drift.

Key Results

State-of-the-art performance among models of comparable size on multiple benchmarks
OSWorld-Verified: 48.9

Highlights

One unified model family covering UI grounding, cross-platform navigation (mobile/desktop/web), and visual tool use
Scalable RL infrastructure with hundreds of concurrent environment instances across heterogeneous backends
Asynchronous training framework that stays stable despite noisy environment feedback and off-policy drift

*Auto-collected on 2026-09-22.*