NVIDIA's open-weight model with efficient MoE architecture; competitive on coding and reasoning benchmarks.