Alibaba Tongyi MAI releases MAI Gym, a 9B open GUI agent hitting SOTA under 10B
AdinaYakup · x · 2026-10-28
Alibaba's Tongyi MAI team released MAI Gym, a 9B Apache 2.0 model that operates both phones and computers. Sim-to-real RL lifted task success from 28.8% to 48.1%, with up to 20K parallel training environments. It's reported as SOTA among open GUI agents under 10B. The accompanying diagram shows the full online RL pipeline: rollout scheduling, trajectory buffering, a training backend, an environment manager supporting MobileWorld, OSWorld, WebArena and more, with deployment via Kubernetes, Docker, or existing endpoints.
Related event: Alibaba Open-Sources 9B GUI Agent Model for Phones and PCs(2 posts)→
More from coding & agent
- How LangChain built its design agent: taste comes from context, not the model — LangChain · 2026-10-28
- LangChain open-sources Content Studio, the design agent behind 3,000+ brand assets — LangChain · 2026-10-28
- Multimodal GraphRAG models missing and contradictory evidence for storm damage assessment — JeremyCMorgan · 2026-10-28
- Vibe coding in real life: 20 hours of personal-project work done in 2 — gleech · 2026-10-28
- LangChain open-sources Content Studio, a Slack design agent, and shares its harness design — BraceSproul · 2026-10-28
- "If AI can do it, humans shouldn't": team lets AI agent review PRs — Yuchenj_UW · 2026-10-28