Hands-on with Qwen 3.8 27B: Default 'xhigh' reasoning causes hilarious over-thinking
Simon Willison · rss · 2026-08-17
Simon Willison tested Alibaba's open-source 27B vision model Qwen 3.8, uncovering severe over-thinking issues caused by its default settings.
Key Findings:
- Configuration Trap: The model defaults to xhigh reasoning effort, consuming excessive context and time even for simple tasks.
- Case Comparisons:
- "Pelican on a bicycle" SVG: Default mode took 21 minutes and 22k reasoning tokens; disabled reasoning took just over 2 minutes.
- "Draw a circle" prompt: The model generated a complex animated geometric study instead of a simple circle due to over-analysis.
- Vision Capabilities: Performed excellently in bounding box detection, accurately identifying objects in photos.
- Agent Potential: When building an image labeling tool, the high-reasoning mode, while verbose, successfully generated a fully functional HTML/JS tool in one shot; low reasoning required multiple iterations.
Recommendation: Users should set reasoning to low or off unless tackling complex tasks.
More from coding & agent
- WhatsApp Blocks New Device Linking? Dev Tries 4 Libraries, All Fail — Jason-Ping · 2026-08-17
- ComfyUI Mixed Mode for MiniMax H3: Multiple Generation Modes in One Timeline — Acceptable-Chest9695 · 2026-08-17
- MiniMax H3 Mixed Mode Deep Dive: Per-Segment Generation Modes — Acceptable-Chest9695 · 2026-08-17
- User observes ChatGPT quality degrades in new chats compared to long threads — DawniJones · 2026-08-17
- Red Hat: Building a production-grade operational layer for AI agents — blaizedsouza · 2026-08-17
- The Four Types of Agent Loops: Choosing the Right Structure for Your Task — blaizedsouza · 2026-08-17