Qwen’s reasoning default turns simple work into long waits
A hands-on report on Qwen 3.8 27B found that its default extra-high reasoning setting used a long reasoning trace for simple prompts on local hardware. In one documented test, a visual-generation request took 21 minutes and used 22,276 reasoning tokens before the author reran it much faster with reasoning turned off.
Teams often treat a model’s default reasoning setting as a neutral quality choice. That can quietly turn routine drafting, sorting and extraction work into slow, costly runs, especially where a tool or model gateway exposes a reasoning-effort control. This is one practitioner’s local test, not a universal benchmark. Its useful lesson is operational: task complexity and the review standard should determine reasoning effort, rather than leaving every job on the most expensive setting.
Analysis
For one repeatable AI task, run a short three-mode comparison at low, medium and high reasoning effort. Record time, cost and the human-review result, then set a default that meets the quality bar instead of assuming the highest setting is best.
Source note
Pulse published by Collab365 Spaces, reviewed by Collab365 editorial team on . Cite as "Qwen’s reasoning default shows why AI tasks need effort budgets", Collab365 Spaces.