Qwen inference drops from 20 tps to 5 tps after latest LM Studio update

No_Cartographer3953 · reddit · 2026-11-21

A user reports that after the latest LM Studio (Strata) update, Qwen FN Q8 — previously running at 20 tps — is no longer listed in the launch model section, and running it via the existing download yields only 5 tps. It's unclear whether this is a regression or a quantization/backend change, and the user is asking the community if others see the same issue.

Original post →

More from Infra

Infra channel →