Testing Qwen 3.8 27B minimal thinking in a specific harness

rosie254 · reddit · 2026-08-15

A user observed that Qwen 3.8 27B barely performed reasoning tasks in a specific harness. The issue is attributed to potential system prompt design and tool overload (8k tokens), tested at high reasoning effort.

Original post →

More from Models

Models channel →