# PrismML releases Ternary Bonsai 2 models for local AI on phones and wearables

Digest AI · Agents & Tools · published 2026-10-15T21:45:44Z

Canonical: https://digestai.news/story/prismml-releases-ternary-bonsai-2-models-for-local-ai-on-phones-and-we

## Summary

They join the existing **Bonsai 2 27B** model, expanding options for developers targeting limited-resource devices like smartphones and wearables.

The new models prioritize practical tasks such as instruction following, tool use, and concise responses. Benchmarking across **19 tests** in six categories—agentic tool use, math, instruction following, coding, knowledge, and reasoning—shows improved performance over their full-precision counterparts. The models support **NVIDIA GPUs** (via **llama.cpp**) and **Apple devices** (via **MLX**), with weights available under **Apache 2.0**.

## Key points

- PrismML’s **Ternary Bonsai 2** models (2B, 4B, 9B) use ternary weights to cut memory to **0.43GB–2GB**, enabling local AI on phones and wearables

## Why it matters

Smaller, efficient models like these could enable AI assistants on everyday devices without cloud dependency, lowering latency and power use for mobile and wearable applications.

## Sources

1. [Ternary Bonsai 2 Family: 2B, 4B, & 9B based on Qwen 3.5](https://prismml.com/news/bonsai-2-family) (prismml.com, 2026-10-15)

## Cite

Digest AI, "PrismML releases Ternary Bonsai 2 models for local AI on phones and wearables", 15 October 2026, https://digestai.news/story/prismml-releases-ternary-bonsai-2-models-for-local-ai-on-phones-and-we

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/prismml-releases-ternary-bonsai-2-models-for-local-ai-on-phones-and-we.json