Nvidia developing 1-trillion-parameter open-source AI model to rival top systems

Nvidia ↘ 0.02% is building a new family of open-source AI models called Nemotron 4, with the largest version expected to contain at least 1 trillion parameters, according to a report published Tuesday by The Information citing multiple employees working on the project.

The chipmaker has not set an official release date and has yet to complete final training, though employees believe the model could be ready as early as late fall, Reuters reported. The effort positions Nvidia as one of the few major U.S. companies releasing open-source frontier models at a time when rising AI costs and competitive pressure from low-cost Chinese alternatives are reshaping the industry landscape.reuters

A Strategic Bet on Open Models

"Nvidia is investing in Nemotron because we believe every company and every country needs accessible frontier open models to strengthen safety and security, accelerate innovation, and provide a foundation they can rely on from one generation to the next," Kari Briski, vice president of generative AI, said in an emailed statement.firstpost+1

The move comes as cheaper Chinese models have narrowed the performance gap with leading systems from Anthropic and OpenAI, and a series of recently disclosed hacks involving autonomous AI agents has heightened interest in open models that lack restrictions on cybersecurity use. Nvidia last month joined a coalition focused on AI safety and co-signed an open letter with Microsoft ↘ 0.44% and other firms defending open-weight models to prevent innovation from drifting overseas.reuters+2

The Nemotron Coalition, announced at GTC in March, includes Mistral AI, Black Forest Labs, and LangChain as founding members collaborating on frontier open models built on Nvidia's DGX Cloud infrastructure.cryptobriefing

Nemotron 3.5 Lightning and NeMo Switchyard

Alongside the Nemotron 4 development, Nvidia on Tuesday released two new products. Nemotron 3.5 Lightning is a 30-billion-parameter mixture-of-experts model with 3 billion active parameters, designed for always-on agentic tasks including code review, tool use, security alert monitoring, and billing queries. The company claims it delivers up to 4x faster token generation than similar-sized models, translating to roughly 30% faster agentic task completion.seekingalpha+2

Nvidia also launched NeMo Switchyard, an open-source model-routing library that automatically directs AI tasks to the most suitable available model, allowing enterprises to run multi-model setups without rebuilding applications.firstpost+1