# Trying hybrid inference (GPU+CPU) with 200-400B models

> Source: <https://forum.level1techs.com/t/trying-hybrid-inference-gpu-cpu-with-200-400b-models/247625#post_11>
> Published: 2026-09-15 00:52:37+00:00

Some updates, been running on 4x V100 so far its been beating in TG the 5090 for Qwen 3.8 27B

I have been using it more for Qwen 3.8 Flash Next, beats x2 DGX Spark.

https://x.com/jkyamog/status/2096954517984346459?s=20