Simon Willison • 4/22/2026

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

This article discusses Qwen3.6-27B, a new open-weight dense model from Qwen that claims flagship-level agentic coding performance, outperforming the much larger Qwen3.5-397B-A17B model on coding benchmarks. The author provides a hands-on test using a quantized GGUF version (16.8GB) via llama-server, including setup commands and performance metrics. Examples include generating SVGs of a pelican on a bicycle and an opossum on an e-scooter, showing strong results for a local model. The article is relevant to IT/technology, focusing on AI model performance, local deployment, and coding benchmarks.

0 comments

#open source #llm #Local Inference