It’s objectively worse in my local tests compared to Mistral. Again their model doesn’t include MT-bench benchmark because it’s really really bad at answering a follow up question(s). (this is also a problem in Ultra). It’s reasoning is also pretty bad compared to mistral.