Summary & Key Takeaways ​

• Georgi Gerganov highly recommends Qwen3.6-27B as an effective local LLM for coding.
• He uses it daily for small, mundane programming tasks at ggml-org.
• The model runs efficiently on his M2 Ultra and RTX 5090 setups.
• Gerganov highlights its utility as a helpful tool for maintainers.
• He employs a lightweight agent harness with a custom system prompt for alignment.

Our Commentary ​

Georgi Gerganov's word carries weight in the local LLM scene. If he's saying Qwen3.6-27B is 'very capable' for coding, I'm paying attention. It's good to hear practical, real-world usage reports, not just benchmark numbers.