AnalysisAI ModelsAugust 16, 2026

Qwen 3.8 27B excels but defaults to overthinking

Qwen 3.8 27B, an Apache 2 licensed 27B vision-capable LLM from Alibaba, shows benchmark boosts over Qwen 3.6 27B and Qwen 3.7-Plus. Its default reasoning effort is 'xhigh', causing excessive thinking; Simon Willison found it took 21 minutes to generate an SVG.

1 source

More stories today

Open the live feed
Qwen 3.8 27B excels but defaults to overthinking — AIBriefs