Qwen 3.8 27B is excellent, but it defaults to wildly overthinking things

Alibaba's Qwen research lab released a new language model called Qwen 3.8, designed to run on consumer laptops. Simon Willison's Weblog reports the twenty-seven billion parameter model delivers impressive performance—it can generate detailed graphics, identify objects in photographs, and even assist with coding. But it has a notable quirk: the default reasoning setting is extremely high, causing the model to overthink simple requests, sometimes spending twenty minutes to produce what it could generate in two minutes. Willison recommends switching to a lower reasoning setting for more practical results.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton