genuinenly very impressive model
as a regular gpt user ( the online user) not local gpt-oss, this bad boy is on par with the online model for my use case (coding related questions/medium horizon reasoning/logic questions), has very strong reasoning capabilities.and it reaches gpt levels in terms of formating so thats good already. I honestly dont know how nvidia achieved this with a 30b total 3b active model. Very nice architecture, feels really good. reminds me of gpt-o3, gpt-oss 20b is definetly 2nd or even 3rd place to this, which is very impressive as thats a very solid model too. The super/ultra models will shake the industry, and im not even exxagarating, if you have the hardware you'll be able to effectively replace online models with them
this is the best model under 30B I've tried to pair with openclaw. Just works perfect. gpt-oss-20b, glm 4.7 falsh, qwen 3 coder none of them works as good as this model.