GitHub radar
Qwen3.8-27B: Multimodal AI for Local Run
Qwen (Alibaba) released Qwen3.8-27B — a 27B multimodal model that reads images and text, holds 262K tokens of context, and runs locally via Ollama.
Qwen3.8-27B is the new flagship of the open Qwen model series from Alibaba. It understands both text and images, supports up to 262,144 tokens of native context (expandable to 1 million), and excels at coding, agentic tasks, and research. Compared to the previous Qwen3.6-27B, it shows measurable gains: 61.7% vs 53.5% on the SWE-bench Pro coding benchmark. The model features adjustable reasoning depth — you can request a quick answer or trigger deep step-by-step thinking. Over 700 quantized GGUF versions are available for Ollama and llama.cpp, including official ones from unsloth.
Why a vibe-coder should care
If you already run local models and want to upgrade to something that handles images alongside text, this is the most capable open option at the 27B tier. The 4-bit quantized version fits in about 17 GB, which means a Mac with 32 GB RAM can run it. For coding help, long document analysis, or vision tasks — without sending data to any cloud service.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Install Qwen3.8-27B locally for me: https://huggingface.co/Qwen/Qwen3.8-27B — pick the right unsloth GGUF quantization for my Mac RAM, run it through Ollama, and show me how to call it
A Mac with 32 GB of RAM or more, or a GPU with 24 GB — the 4-bit version takes around 17 GB.
Open on Hugging Face▌ More finds