AISimon Willisonabout 2 hours ago
Qwen 3.8 27B excels but overthinking by default
11 min read
Qwen 3.8 27B, an Apache 2 licensed vision-capable LLM from Alibaba, defaults to extra-high reasoning effort, causing excessive token use and slow outputs. Running locally on an M5 Max MacBook Pro and NVIDIA DGX Spark, it produced detailed results but required adjusting context limits.
Level
Hype check
Tap to vote and see what everyone thinks.
