I gave Qwen 3.6 a real coding task on my RTX 5080, and now I barely open Claude

A local model finally finished the job without me arguing with it.