PrismML Releases Bonsai 2 Ternary Models (2B, 4B, 9B) Based on Qwen3.5

Key Info

PrismML released Bonsai 2 in 2B, 4B, and 9B sizes based on Qwen3.5, using ternary weights for compact on-device deployment.

Highlights

• Model sizes and footprints: 2B at 0.43GB, 4B at 0.95GB, 9B at 2.0GB
• Focus on local assistant use: instruction following, avoiding repetitive loops, mobile tool calling
• The 9B model uses roughly 40% fewer tokens on standard instruction-following and everyday math tasks
• Successors to the first-generation Bonsai 1.7B and 4B models