Georgi Gerganov, the creator of llama.cpp, recently confirmed on Hacker News that Qwen3.6-27B serves as a highly capable local model for coding tasks. He has utilised the model almost daily over the last month and a half, running it on either his M2 Ultra or his RTX 5090 workstation. Gerganov applies the tool to mundane maintenance duties at ggml-org, noting it is a helpful resource for a maintainer despite the lack of impressive breakthroughs. He currently employs a lightweight harness consisting of the pi agent with all features stripped down and a short system prompt to align the output with his specific coding style.
This endorsement matters because it validates the viability of running sophisticated coding agents locally without relying on cloud infrastructure. The statement suggests that models like Qwen3.6-27B can effectively handle practical development workflows on standard consumer hardware, reducing the need for extensive human review of pull requests. It reinforces the trend towards privacy-focused and offline-capable AI tools, demonstrating that high-quality assistance is accessible directly from personal devices.
* Qwen3.6-27B is confirmed as a practical local model for daily coding tasks on M2 Ultra or RTX 5090 hardware.
* Georgi Gerganov uses a stripped-down pi agent with a custom system prompt to align local model output with his style.
* Local deployment allows for effective AI-assisted programming without the latency or privacy concerns of cloud-based services.
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.
