Qwen 3.8 27B Vs. Qwen 3.6 27B on oMLX

View original

Qwen

qwen

Heat trend

Collecting trend data

The percentage is based on available heat signal, not comment count or independent people.

AI summary

A comparison between Qwen 3.8 27B and Qwen 3.6 27B on oMLX reveals a significant quality improvement, with scores rising from 81.1 to 87.7 (+8%). However, this enhancement comes at a cost: speed decreased from 35 to 29 tok/s (-16%), and runtime increased fivefold from 8m51s to 44m39s. Output tokens also saw a substantial jump from 18K to 78K, indicating that while quality is noticeably better, it requires more tokens and time.

Noticeably better quality, but you're paying for it with tokens and time. Full benchmark results (all hardware, all quants): llm-bench.io Qwen3.6-27B Vs. Qwen3.8-27B