Report
PrismML releases 2B, 4B and 9B Bonsai 2 models
PrismML says the Qwen3.5-based ternary models are designed for local assistants on computers, phones and wearables.
TLDR
PrismML announced the three Bonsai 2 models on October 15, 2026, listing footprints of 0.43GB for 2B, 0.95GB for 4B and 2.0GB for 9B. It says they bring stronger reasoning, visual understanding and tool calling to compact devices. PrismML also says the 9B model uses roughly 40% fewer tokens on standard instruction-following and everyday math tasks. The models were released under Apache 2.0.
Combined views
1.3K
1 Source, first seen ago
