Indeed. I 've been using Sage A1.5 35b-A3b to have more speed because of strix halo limited memory bandwidth. Looking forward to testing Inkling-2-mini once it has lamma.cpp support. Genuinely excited from a model outside Qwen3.6-35B-A3B ecosystem competing in that space.
P.S. Qwen3.8-Flash-Next is also a very good model for Srix-Halo/128Gb Framework Desktop but I liked to have more RAM for other things as well even when I was running it in Q4. Usable though.