GitHub radar
27B Model in 6 GB: Ternary-Bonsai from Prism ML
Prism ML released a 27-billion parameter model based on Qwen that occupies 5.95 GB and runs on a regular laptop.
Ternary-Bonsai-2-27B is a 27-billion parameter language model based on Qwen3.8-27B, released by Prism ML. It uses ternary weight representation, where each parameter stores one of three values, resulting in a true 1.72 bits per weight across all layers. The deployed file weighs 5.95 GB, compared to approximately 54 GB for the standard FP16 version. According to the authors, the model retains 98.2% of FP16 intelligence and scores 84.78 on average across 14 thinking-mode benchmarks, while a standard IQ2_XXS build of the same base model scores 72.59. On an Apple M5 Max laptop it runs at approximately 47 tokens per second. The context window is 262,144 tokens. The model files are in GGUF format, compatible with llama.cpp on CUDA, Metal, and CPU.
Why a vibe-coder should care
At 5.95 GB, the model runs on a regular laptop with 8 GB of RAM or more, which makes a 27B-class reasoning model available for local use without a cloud subscription. For tasks where privacy or offline access matters, this is a practical option. In my assessment, the quality is appropriate for local use: suitable for simple repetitive tasks or when there is no budget for a paid subscription, but not a replacement for frontier models in agent work.
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