NVIDIA is consistently forging ahead in the realm of artificial intelligence. Following the launch of the Nemotron 3 series of open-source large models at the close of last year, NVIDIA has officially announced the arrival of the next-generation Nemotron 3 Super today. Tailored explicitly for AI agents, this model is equipped with a staggering 120 billion parameters, with 12 billion of them being actively utilized. Nemotron 3 Super has undergone in-depth optimization for the Blackwell architecture, resulting in a threefold enhancement in inference speed. When compared to its predecessor, the Nemotron Super model, it achieves up to five times higher throughput and a twofold increase in accuracy. This model is geared towards tackling the expensive challenges presented by long-term thinking and context explosion, thereby expediting the automation of intricate tasks and the realization of multi-step decision-making scenarios. Presently, Nemotron 3 Super is compatible with Google Cloud and Oracle, with support for Amazon AWS and Microsoft Azure slated to be activated in the near future.