{"slug": "prismml-releases-ternary-ai-models-as-small-as-0-43gb-for-local-agents", "title": "PrismML releases ternary AI models as small as 0.43GB for local agents", "summary": "PrismML released the Bonsai 2 family of ternary-weight AI models in 2B, 4B and 9B sizes on October 15th, with the smallest file at 0.43GB, under a free Apache 2.0 license. The models, based on Qwen3.5 and announced by Caltech professor and PrismML co-founder and CEO Babak Hassibi, add vision and tool calling for local AI agents on computers, phones and wearables. PrismML's headline performance figures are company-reported, so developers should test the models on their target hardware.", "body_md": "# PrismML releases ternary AI models as small as 0.43GB for local agents\n\n**The free Apache 2.0 Bonsai 2 models use ternary weights and add vision and tool calling, alongside performance figures reported by PrismML.**\n\n        By [RuntimeWire Staff](https://runtimewire.com/author/runtimewire-staff)\n        · Published \n\nPrimary source: [PR Newswire](https://www.prnewswire.com/news-releases/prismml-expands-bonsai-2-family-with-2b-4b-and-9b-models-for-on-device-ai-302910593.html)\n\n## Why it matters\n\nSmaller model files can make local inference more practical on devices with tight memory and power limits. PrismML's release adds vision and tool use to that pitch, while its headline performance figures are company-reported, so developers should test the models on their target hardware.\n\n[Babak Hassibi](https://x.com/BabakHassibi?ref=runtimewire), a [Caltech professor](https://www.babak.caltech.edu/bio.html?ref=runtimewire) and co-founder and CEO of [PrismML](https://prismml.com/?ref=runtimewire), released 2B, 4B and 9B models in the Bonsai 2 family on October 15th. Based on [Qwen3.5](https://runtimewire.com/models/huggingface/comfy-org-qwen3.5-c0fb5b4307e653b2), the models are intended to bring reasoning, visual understanding and tool use to local AI agents on computers, phones and wearables.…", "url": "https://wpnews.pro/news/prismml-releases-ternary-ai-models-as-small-as-0-43gb-for-local-agents", "canonical_source": "https://runtimewire.com/article/prismml-bonsai-2-small-models-on-device-ai", "published_at": "2026-10-15 20:40:04+00:00", "updated_at": "2026-10-15 23:58:14.077480+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-agents", "ai-products", "ai-tools"], "entities": ["PrismML", "Bonsai 2", "Babak Hassibi", "Caltech", "Qwen3.5"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/prismml-releases-ternary-ai-models-as-small-as-0-43gb-for-local-agents", "markdown": "https://wpnews.pro/news/prismml-releases-ternary-ai-models-as-small-as-0-43gb-for-local-agents.md", "text": "https://wpnews.pro/news/prismml-releases-ternary-ai-models-as-small-as-0-43gb-for-local-agents.txt", "jsonld": "https://wpnews.pro/news/prismml-releases-ternary-ai-models-as-small-as-0-43gb-for-local-agents.jsonld"}}