Q38FN is absolutely a step change IMO. Literally better than Opus 4.6 for the work I’m using it for, but I can run it forever for free on my GB10 box. Wild.
replyAnd it’s an experimental likely undertrained model. Wait til Qwen 4 Flash…
I assume it was put out to get inference engines ready to do qwen4 models. For AMD halo, the Halgogen engine is the fastest i dound so fare
reply