It is probably the same base model as Llama 3.0.

They mention postraining improvements.