Run Bonsai 27B on Your Phone: 1-Bit Local AI
Summary
Run the first 27B-class model on a phone: MLX, llama.cpp, tool calls, and the memory math.
On July 14, 2026, a small lab called PrismML shipped something that sounded impossible a year ago: a 27-billion-parameter model that runs on a phone. It is called Bonsai 27B, and the 1-bit build weighs just 3.9 GB — small enough to fit inside the per-app memory budget of an iPhone 17 Pro.
The trick is not a new architecture. Bonsai 27B is a low-bit repackaging of Qwen3.6-27B. Same weights, same layers — but each parameter is squeezed down to roughly one bit instead of sixteen. The result is a multimodal model with a 262K-token context that you can run fully offline, on a laptop or a handset, with no API bill and no data leaving the device.
Keep reading — it's free
Enter your email to keep reading — plus the best of AI & tech, daily. Free, forever.
Already a member? Sign in
Comments
Be the first to comment