Llama-Nemotron: Efficient Reasoning Models

Llama-Nemotron is an open family of heterogeneous reasoning models developed from Llama 3 using neural architecture search, knowledge distillation, and continued pretraining, followed by supervised fine-tuning and large-scale reinforcement learning. The family includes three sizes—Nano (8B), Super (49B), and Ultra (253B)—that deliver competitive reasoning performance against state-of-the-art models like DeepSeek-R1, while offering superior inference throughput and memory efficiency. A key innovation is the first open-source dynamic reasoning toggle, allowing users to switch between standard chat and reasoning modes during inference. The models are released under an open license suitable for enterprise use, along with resources to support open research.