Qwen 3.8 27B is excellent, but it defaults to wildly overthinking things
Qwen 3.8 27B punches way above its weight class—but users are fighting its default reasoning mode to get speed.
Why it matters
A capable open-weight model shipping with a problematic default behavior is a frontier-lab trade-off story: capability vs. usability. Practitioners choosing between models need to know this quirk; enthusiasts tracking the open-weight race see Alibaba's Qwen competing on benchmarks while fighting its own design decisions.
The key facts
11 to knowQwen 3.8 27B released (open-weight)
Model exhibits 'wildly overthinking' as default behavior
Performance claimed as 'excellent' despite usability friction
Published by Simon Willison on simonwillison.net (influential independent AI analyst)
August 2026 release date
Open-weight model for local/on-prem deployment
Qwen 3.8 27B released
Model defaults to extended reasoning/overthinking behavior
Capability assessment by independent reviewer (Simon Willison)
Deployment consideration: reasoning-loop behavior affects practical use
Open-weight model (implication for adoption)
Go to the source
Simon Willisonsimonwillison.net