MAI-UI-8B-f32-GGUF

MAI-UI-8B from Tongyi-MAI (Alibaba's Tongyi Lab) is an 8B-parameter vision-language foundation model from the MAI-UI family (2B/8B/32B/235B-A22B variants), built on Qwen3-VL architecture and specialized as a GUI agent for mobile/Android environments, processing UI screenshots and natural language instructions to generate structured actions (clicks, swipes, text entry) plus native agent-user interactions and MCP tool calls via supervised fine-tuning and online RL with GRPO on self-evolving navigation data incorporating perturbed seed tasks, multi-view UI element reasoning, and device-cloud collaboration routing. It addresses key deployment challenges—native interaction, UI+tools operation, dynamic environments, and privacy-aware execution—establishing SOTA on AndroidWorld/MobileWorld benchmarks (outperforming Gemini 2.5 Pro, Seed1.8, UI-Tars-2) with 39.7% pure GUI success, 51.1% interaction tasks, and 37.5% MCP-augmented tasks through scaling parallel envs (32→512, +5.2 pts) and step budgets (15→50, +4.3 pts). Optimized for consumer GPUs (RTX 3090/A100) with device-cloud hybrid execution reducing API calls 40% while boosting on-device perf 33%, it enables realistic GUI automation across 20+ apps under permissive license for advancing human-computer interaction.

MAI-UI-8B [GGUF]

File Name	Quant Type	File Size	File Link
MAI-UI-8B.F32.gguf	F32	32.8 GB	Download
MAI-UI-8B.Q8_0.gguf	Q8_0	8.71 GB	Download
MAI-UI-8B.mmproj-f32.gguf	mmproj-f32	2.31 GB	Download
MAI-UI-8B.mmproj-q8_0.gguf	mmproj-q8_0	752 MB	Download

Quants Usage

(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)

Here is a handy graph by ikawrakow comparing some lower-quality quant types (lower is better):