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
- Six-month prediction: hiring for generic agent harnesses built from scratch will cool — yi111 · 2026-10-16
- Omarchy ISO wouldn't boot on NVIDIA, so he had Claude patch it — indragie · 2026-10-16
- Dev uses Grok to clean up 3D printer firmware, catches it sneaking in nonsense code — yacineMTB · 2026-10-16
- A rude AI agent took 89% of a shared resource in 102-run multi-agent personality experiment — Brave-History-6502 · 2026-10-16
- Guardrails can make risky delegation work — but it takes a full-time team, says iandanforth — iandanforth · 2026-10-16
- What can you truly delegate? Only tasks with unambiguous, verifiable results — iandanforth · 2026-10-16