Nvidia introduces the Nemotron-Nano-9B-v2, a hybrid Mamba-Transformer model designed for efficient reasoning in text-only applications. The model permits users to toggle reasoning traces via control tokens, balancing accuracy and latency, particularly benefitting customer support applications. Evaluation results show the model achieving 72.1% on AIME25 and surpassing competitors like Qwen3-8B. Released under a permissive Nvidia Open Model License, it enables immediate commercial use without additional fees, promoting enterprise-friendly deployment.
Nvidia introduces the Nemotron-Nano-9B-v2, a hybrid Mamba-Transformer model designed for efficient reasoning in text-only applications. The model permits users to toggle reasoning traces via control tokens, balancing accuracy and latency, particularly benefitting customer support applications. Evaluation results show the model achieving 72.1% on AIME25 and surpassing competitors like Qwen3-8B. Released under a permissive Nvidia Open Model License, it enables immediate commercial use without additional fees, promoting enterprise-friendly deployment.