GitHub radar
Qwen3.8-27B with Vision Support — Local GGUF
The ISTA-DASLab team has published a GGUF version of Qwen3.8-27B (Alibaba) with integrated vision projector, allowing the model to reason over images, screenshots and diagrams.
Qwen3.8-27B is a 27B parameter model developed by Alibaba. It is one of the best open models for reasoning, math, and coding. The ISTA-DASLab (Austrian Institute of Technology) team has published a GGUF version of this model using non-uniform quantization. This means that each tensor is quantized individually based on its sensitivity to maximize the model quality while keeping the model size small. This version of the model comes with the integrated vision projector, meaning it can handle images, screenshots, diagrams, and photos. The model size is between 8.4 GB and 11.8 GB depending on the quantization. This should fit on any laptop with at least 16 GB RAM.
Why a vibe-coder should care
Enjoy Alibaba’s best open model locally including image processing capabilities without having to use the cloud or paying a monthly fee. Perfect for processing screenshots, UI designs and diagrams in your local agent flow.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Download this GGUF build of Qwen3.8-27B with vision support: https://huggingface.co/ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF — pick the right file for my setup, run it via LM Studio or llama.cpp, and show me how to send it an image.
A regular laptop with 16 GB of RAM or more — the model takes 8 to 12 GB depending on the file you choose.
Open on Hugging Face▌ More finds