Not yet, but I'm sure they will release some benchmarks soon. As for it not being an MoE model, there's still a ton of value in having a small non-MoE model for many usecases, and improvements that get discovered to train the small model can potentially improve the next version of the MoE model down the line.