PrismML's ternary Bonsai 2 fits a 9B vision-language model into ~2 GB
lmoroney · x · 2026-10-17
PrismML released Ternary Bonsai 2 2B, 4B, and 9B models (Apache 2.0), based on Qwen3.5, joining the existing 27B to target local agents on phones, PCs, and wearables.
- Ternary weights (−1/0/+1 with small group-wise scales) shrink footprints to 0.43GB, 0.95GB, and 2GB—over 9× smaller than full-precision counterparts.
- Development focused on practical agent behaviors: avoiding repetitive loops, instruction following, tool calling on mobile, and controlling verbosity.
- The family aims to enable on-device agents that reason, see, and use tools within tight memory and power budgets.
Related event: PrismML open-sources ternary Bonsai 2 models, 9B at just 2GB(10 posts)→
More from Infra
- E2B founder: most agent sandbox providers are just Firecracker wrappers on AWS — zeeg · 2026-10-17
- Data movement, not compute, is AI's real cost: near-memory investment map in 3 layers — FinanceYF5 · 2026-10-17
- Hash Rate Ep. 189: Jev models on Bittensor subnet SN92, decentralized training vs traditional LLMs — markjeffrey · 2026-10-17
- Broadcom: each gigawatt of AI compute could yield $30B in annual revenue for customers — Beth_Kindig · 2026-10-17
- Anthropic ships OpenTelemetry support for Claude Platform, joins GenAI traces standard group — dsp_ · 2026-10-17
- Texas grid queue hits 474GW; white paper argues AI data centers should become grid resources — espricewright · 2026-10-17