LaunchDevelopersNovember 12, 2026

ComfyUI doubles local LLM token rates with Qwen3.x rewrite

Read original source →blog.comfy.org

ComfyUI v0.45.0 rewrote its text-generation hot path for Qwen3.x models, hitting up to 2x token rates with no quality change. Qwen3.5 9B on an RTX 5090 went from 106 to 219 tokens/sec; Qwen3.8 27B image description rose from 55 to 122.

1 source

More stories today

Open the live feed