NVIDIA Develops Nemotron-4 Model, Largest Version Expected to Exceed 1 Trillion Parameters

NVIDIA is developing a new series of AI models called Nemotron-4, with the largest variant expected to have at least 1 trillion parameters. Multiple employees involved in the project revealed that the model could be ready as early as late this fall, though final training has not yet been completed and no official release date has been set. Last month, NVIDIA formed an alliance with several companies to jointly develop and share AI safety and cybersecurity tools, and signed an open letter alongside tech firms like Microsoft endorsing open-weight models. Additionally, NVIDIA has already released the Nemotron-3.5 Lightning model, designed for tasks such as code review, tool calling, security alert monitoring, and answering billing-related queries, and launched NeMo Switchyard—an open-source model routing library that automatically routes tasks to the most suitable model based on requirements.