GitHub radar
27B Model (6 GB) — Ternary-Bonsai by Prism ML
A 27B ternary model built on top of Qwen, running locally on your laptop, only takes 5.95 GB.
Ternary-Bonsai-2-27B is a 27B language model built on Qwen3.8-27B released by Prism ML. This model uses ternary weights (each weight can be represented by one of three values), achieving a true 1.72 bits per weight over all layers. With a size of 5.95 GB (vs. ~54 GB FP16), it claims to preserve 98.2% FP16 intelligence and achieves an avg score of 84.78 in 14 thinking-mode tests, comparing to 72.59 from a normal IQ2_XXS build of the same model. It can run at ~47 tok/s on an Apple M5 Max laptop. Context length: 262144 tokens. Files are in GGUF format, working with llama.cpp under CUDA/Metal/CPU.
Why a vibe-coder should care
With only 5.95 GB, this model can actually run on your everyday laptop, with 8GB RAM or above. Now you can use a 27B reasoning model locally without paying for cloud services! When data privacy or offline usage is required, this model is a great choice. In my opinion, this model is good enough for local usage, such as some trivial tasks or if you have no money to pay for any paid models, but this model is far away from state-of-the-art models for agents.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Install the Ternary-Bonsai-2-27B model from Prism ML: https://huggingface.co/prism-ml/Ternary-Bonsai-2-27B-gguf — download the GGUF file, run it via Ollama, and show me an example conversation.
A regular laptop with 8 GB of RAM — the model occupies about 6 GB; 16 GB is more comfortable.
Open on Hugging Face▌ More finds