Ternary Bonsai 2 27B: 9x smaller than Qwen3.8 27B, keeps 98.2% of performance
TheZachMueller · x · 2026-09-18
- PrismML released Ternary Bonsai 2 27B, a ternary-quantized model based on Qwen3.8 27B that is 9x smaller than its full-precision counterpart at just 5.9 GB, while retaining 98.2% of aggregate benchmark performance.
- Two months after the first Bonsai 27B, the big change is quality: notable gains in agentic coding, multimodal reasoning, and long-horizon tool use.
- Available today under Apache 2.0, targeting local AI deployment.
More from Infra
- NVIDIA DGX Station with GB300 demos thousands of tokens per second fully local — TheZachMueller · 2026-09-18
- OpenAI exec: GPUs hit 7-40 IQ points per watt vs human's 5, a milestone we 'zoomed past' — GregCook2011 · 2026-09-18
- China refines 91% and makes 94% of sintered rare-earth magnets powering motors from EVs to GPU data centers — demian_ai · 2026-09-18
- Mystery trader drops $100M in premium on 2-week AI stock calls expiring Oct 2 — toptickcrypto · 2026-09-18
- First-ever PyTorch Day Japan lands in Tokyo on December 10, CFP open till Sept 27 — PyTorch · 2026-09-18
- LaurieWired's CppCon keynote covers new memory hierarchies and how to prepare — lauriewired · 2026-09-18