LodeHQ
AI news for people who shipFree, by email

Dev · simonwillison.net · Monday, 17 August 2026

Qwen 3.8 27B is great locally, if you shut off its runaway default thinking

Qwen 3.8 27B ships with reasoning effort defaulted to xhigh, so a trivial SVG prompt burned 22k tokens and 21 minutes on a laptop. Flip it off and the same output takes 2 minutes. A lively reminder that the default isn't the right one for consumer hardware.

Read the source: simonwillison.net ↗In the 17 August issue →