PrismML›PrismML› offers Bonsai

Bonsai is the flagship model family from PrismML, engineered to run large language models on consumer devices such as phones. The family applies 1-bit and Ternary quantization to cut memory and compute requirements while preserving model quality, and spans parameter sizes from 1.7B through 27B.

The 27B variant, based on Qwen3.6-27B and announced in July 2026, was reported as the first 27-billion-parameter model to run on a phone. The family also includes an image-generation model, Bonsai Image, and is paired with Bonsai Studio, an iOS application, and Hermes, PrismML's agentic inference stack.