Qwen 3.8 Comparison: Smaller Q4 Model Outreasons Larger Q5

k-r-a-u-s-f-a-d-r · reddit · 2026-08-19

A detailed comparison between two versions of Qwen 3.8 27B (Q4KM vs Q5KM) on a self-referential logic puzzle. Despite being smaller, the Q4 model outperformed the Q5 model in reasoning steps, token efficiency, and error correction. The analysis highlights that the Q4 model used a more efficient range-bounding strategy, while the Q5 model, though rigorous, incurred significantly higher computational overhead.

Original post →

More from Models

Models channel →