PrismML has released two rounds of Ternary Bonsai 2 models: a 27B version built on Qwen3.8 27B at a 5.9GB footprint in September 2026, and 2B, 4B, and 9B versions built on Qwen3.5 base models for phones in October 2026.
PrismML's work on Ternary Bonsai 2 spans two releases separated by about a month. The first, in September 2026, was a 27B model built on Alibaba's Qwen3.8 27B. It ships at 5.9GB, one-ninth the size of the base model, and keeps 98.2% of the base model's benchmark scores. It is licensed under Apache 2.0 and published on Hugging Face. PrismML says it runs on Mac, iPhone, and iPad. The company lists gains over the earlier Bonsai 27B in reasoning, coding, vision, and long-horizon agent performance.
The second release, on October 15, 2026, scales the same quantization approach down to phone-sized weights. Ternary Bonsai 2 2B, 4B, and 9B are ternary versions of the Qwen3.5 base models, using the values -1, 0, and 1. Memory use is 0.43GB, 0.95GB, and 2GB respectively. The smallest model runs at 120 tokens per second on the iPhone 18 Pro Max. The Locally AI app supports the models.
What the two stories establish is a lineup split by target hardware: a large model sized for desktop and laptop use, and smaller models sized for phones, all using ternary weights. The stories do not give sales figures, a release date for any additional model, or benchmark scores for the phone-sized versions.
Written with AI from the cited Yomimono stories below, each itself sourced. See our editorial policy. Every fact links the story it came from.
Structured graph also available as JSON at /public/entities/prismml.
CC BY 4.0.
PrismML released Ternary Bonsai 2 2B, 4B, and 9B on October 15, 2026. The models are ternary versions of the Qwen3.5 base models, using the values -1, 0, and 1, with memory use of 0.43GB, 0.95GB, and 2GB. Ternary Bonsai 2 2B runs at 120 tokens per second on the iPhone 18 Pro Max, and the Locally AI app supports them.
PrismML announced Ternary Bonsai 2 27B, a compressed model built on Alibaba's Qwen3.8 27B. It ships at 5.9GB, one-ninth the size of the base model, and retains 98.2% of Qwen3.8 27B's benchmark scores. PrismML lists gains in reasoning, coding, vision, and long-horizon agent performance over the earlier Bonsai 27B. The model is released under the Apache 2.0 license and published on Hugging Face. PrismML says it runs on Mac, iPhone, and iPad.