Simon Willison • 4/22/2026

Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model

This article discusses the release of Qwen3.6-27B, a 27-billion parameter dense model that claims flagship-level agentic coding performance, surpassing the much larger Qwen3.5-397B-A17B MoE model across coding benchmarks. The author details running a quantized 16.8GB version locally using llama-server, including setup commands and performance metrics. Examples of generated SVG outputs (pelican on bicycle, opossum on e-scooter) demonstrate the model's capabilities. The article is relevant to IT/technology as it covers a new AI model for coding, local deployment, and performance benchmarking.

0 comments

#Qwen #Open Weights #Gguf