A user tested the Qwen 3.6 27B (Q6_K GGUF) model as an autonomous coding agent using LM Studio and OpenCode on an RTX 5090. The setup leverages a 131k context window with Flash Attention and full GPU offload to develop a desktop ARPG in C# and Raylib. The agent is capable of reading code, editing files, and building the project autonomously.
Read original
reddit/r/LocalLLM