Llama 3.3 Nemotron Super 49B v1

Llama 3.3 Nemotron Super 49B v1 is NVIDIA's post-trained Super variant built from Llama 3.3 70B Instruct. Neural Architecture Search reduces serving overhead without making the checkpoint small. Reasoning and direct responses remain distinct operating modes.

The model has 49 billion parameters and 131,072 context tokens for longer agent tasks. The open weights use NVIDIA's Open Model License. The exact v1 identity remains distinct. The later Llama 3.3 Nemotron Super 49B v1.5 is a revised checkpoint, not a simple rename.