Seems less bright than
#1
by lobanov - opened
Just to report that the model seems less bright than qwen3.5-122b-a10b:Q4_K_XL by Unsloth with about 79GB footprint. The latter was able to build an app from a spec without any issues in one shot, while this model routinely got tripped by hanging commands launched in foreground or incorrect use of express JS API. I ran this model with oMLX using default settings.