Researchers working on phone-operating agents get free Apache 2.0 weights, but there is no hosted service or app.
Kenning's take
What it does
MAI-UI is a family of GUI agent models from Alibaba's Tongyi lab that read a phone screen and carry out multi-step tasks in apps on their own. Alibaba released the 8B and 2B versions as free weights on Hugging Face in December 2025, with a technical report, and reported top scores on mobile benchmarks such as AndroidWorld and ScreenSpot-Pro. In July 2026 the team followed it with Qwen-UI-Agent, which extends the work to computers, browsers and search.
- Model sizes from 2B to 235B, with 8B and 2B weights released
- Pure-vision control of smartphone apps
- 76.7% on AndroidWorld for MAI-UI-235B (team figure)
- Apache 2.0 licence
- Successor Qwen-UI-Agent covers mobile, computer and web
Who it suits. Researchers and developers building agents that use phone apps.
Key facts
- Category
- AI models
- Part of
- Qwen
- Made by
- Alibaba (Tongyi Lab)
- Based in
- China
- Released
- 29 December 2025
- How to get it
- free download of MAI-UI-8B and MAI-UI-2B weights from Hugging Face and code from GitHub; no consumer app
- Works on
- self-hosted
- Free plan
- Free Apache 2.0 weights
FAQ
When was MAI-UI released?
Alibaba's Tongyi team released the MAI-UI-8B and MAI-UI-2B weights and technical report on 29 December 2025.
Is MAI-UI free?
Yes. The released weights are free under the Apache 2.0 licence.
What replaced MAI-UI?
The same team introduced Qwen-UI-Agent on 30 July 2026, which extends MAI-UI to computers, the web and search.
Facts from the maker's own pages and the sources listed. Checked 2026-10-11.
Similar tools

Qwen-Image-2.0
Alibaba's 7-billion-parameter model that both creates and edits images, with strong text rendering at 2K

Qwen-Image-Layered
Open Alibaba model that splits an image into separate transparent layers so each part can be edited alone

Qwen-Image-3.0
Alibaba's hosted image model built for text-heavy layouts, taking prompts up to 4,500 tokens

Qwen VLo
Preview multimodal model from Alibaba that understands images and generates or edits them from chat prompts