New models out. Anyone trying them in the GB10?
I am trying Laguna on a single DGX Spark now and getting about 20-30 tps for reasoning. So far, I tested some reasoning over a document that I made it call some tools on it, and it worked as good as Deepseek V4 Flash over two Sparks while using only one Spark. On the other hand, it is failing for single prompt single shot tasks like testing out HTML style small game or simulation generation.