I'm trying command-a-reasoning on their hf space and it seems pretty good. Better than GLM4.5-air for sure.