27B reasoning with ternary weights.
Ternary Bonsai 2 27B is now on NanoGPT, included in the subscription.
PrismML reports 98.2% of Qwen3.8-27B's aggregate benchmark performance. With vision, tool calling and a 262K context window.
https://nano-gpt.com/conversation?model=prism-ml%2Fternary-bonsai-2-27b