Qwen 3.8 27B review: excellent model defaults to hilarious overthinking
marlene_zw · x · 2026-08-17
Simon Willison published an in-depth review of Alibaba's Qwen 3.8 27B model. Licensed under Apache 2.0, this vision-capable LLM shows impressive benchmark improvements over its predecessor and the closed-weight Qwen 3.7-Plus. Running a quantized version on both a MacBook Pro and an NVIDIA DGX Spark, Willison discovered that the model defaults to maximum reasoning depth. This causes it to engage in spectacular overthinking even for mundane tasks, rapidly consuming context windows and resources. While entertaining, this default configuration is inefficient for practical use.
Related event: Qwen 3.8 27B Review: Best Local Open Model with Near-Opus Performance(6 posts)→
More from Models
- User claims Claude silently routes biology queries to a weaker model, rails at Anthropic — jonkhler · 2026-08-17
- Claude appears to be down again, user jokes: 'just in time for bed' — IgorBrigadir · 2026-08-17
- Enterprises lack tools to keep up with rapid SOTA model releases — mattturck · 2026-08-17
- Karpathy: LLMs Want to Imitate, Not Succeed; Over 50% of AI Projects Misuse Agents — leslysandra · 2026-08-17
- Claude seems to be rolling out something new, netizens hope it's not watermarking — ctjlewis · 2026-08-17
- Qwen3.8-27B-AEON-ULTIMATE-UNCENSORED released with BF16 weights — AIandDesign · 2026-08-17