← Back to feed

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

L4 · DeveloperModels & ReleasesSimon Willison's Blog· 8/16/2026

Developer-focused analysis of a new model's performance characteristics and practical runtime considerations for local deployment.

AI Summary

Qwen 3.8 27B is a new open-source vision-capable LLM that defaults to an 'xhigh' reasoning effort setting, causing extensive overthinking even for simple tasks.

Excerpt

Heees my review of Qwen 3.8 27B - I can't remember the last time I've had this much fun playing with a local model that runs on my own computers simonwillison.net/2026/Aug/16/...

Read Original
0 upvotes · 0 downvotes · 1 min read

Related Articles

L4 · DeveloperModels & ReleasesSimon Willison's Blog
Claude Haiku 5.5

Anthropic released Claude Haiku 5.5 with competitive pricing ($0.10/$0.50 per million tokens) matching GPT-6 Luna, plus new API credits for subscribers.

L3 · BuilderModels & ReleasesHacker News
Claude Haiku 5.5

Anthropic releases Claude Haiku 5.5, a significantly cheaper and faster small model optimized for high-volume tasks like summarization, classification, and customer support.

L4 · DeveloperModels & ReleasesHugging Face Blog
Multimodal open d1 decision models for the edge

LiquidAI releases d1-3B and d1-omni-600M, open multimodal decision models optimized for edge devices with 16-50ms inference times on NVIDIA Jetson hardware.

L3 · BuilderModels & Releases@AnthropicAI
Anthropic (@AnthropicAI): Claude Haiku 5.5, our latest model: https://t.co/a81H0avciK

Anthropic released Claude Haiku 5.5, a small model that's 75% cheaper and more capable than its predecessor.

L4 · DeveloperModels & ReleasesAI Business
Reflection Beam aims U.S. open-weight AI model at Chinese rivals

Nvidia-backed startup Reflection Beam launches an open-weight AI model targeting Chinese competitors in the U.S. market.