Bonsai Distillation Explained: From Qwen3.6-27B to a Phone-Friendly 3.9 GB Model
August 04, 2026A step-by-step look at how Prism ML distilled a 27B parameter model into a 3.9 GB phone-capable file while keeping 89.5% of full-precision reasoning.
Posts tagged "on-device ai"
Can a 3.9 GB Model Really Code? Testing Bonsai 27B 1-Bit on a Laptop
August 04, 2026I ran the 1-bit Bonsai 27B model locally on a laptop. Here's what happened when a 3.9 GB GGUF tried to write code, solve math problems, and keep up with full-precision models.
How Bonsai 27B Packs 27B-Class Reasoning into 1-Bit Weights
August 04, 2026A technical walkthrough of how Prism ML distills Qwen3.6-27B into binary and ternary weights, keeps reasoning alive, and runs it on a phone.
Is Kimi Distilled from Claude? Here Is What We Know
July 30, 2026Is Kimi distilled from Claude? We investigate the evidence behind Kimi distillation — output similarity, Anthropic's terms of service, community consensus, and what it means for developers using Kimi distilled models.
Cursor's Secret Architecture: Kimi-k, Claude Distillation, and the New AI Stack
July 20, 2026A technical dive into how Moonshot AI's Kimi-k models and the art of distillation are powering the near-instant coding experience in Cursor.