Aug 11 - Nvidia is working on a new family of open-source AI models called Nemotron 4, with the stated objective of matching top open-weight models worldwide, according to people involved in the effort. Multiple employees cited for the project said the largest Nemotron 4 instance is expected to contain at least 1 trillion parameters.
Those working on the effort say Nvidia has not finalized a release schedule and training remains incomplete. Employees who spoke about timelines indicated the model could be ready as early as late fall, though the company has not announced a date. Nvidia did not immediately respond to requests for comment on the development.
The move places Nvidia among a small group of major U.S. technology firms releasing open-source models - a space that has gained increased attention this year as costs for proprietary AI systems have risen and lower-cost models from Chinese developers have closed some capability gaps with leading American labs.
Concerns about cybersecurity have heightened scrutiny of open models following disclosures of hacks involving autonomous AI agents. Open-weight models typically lack the built-in use restrictions found in some proprietary systems, and the security implications of broader public access continue to be a focus for industry participants.
In parallel with Nemotron 4, Nvidia recently introduced Nemotron 3.5 Lightning, described as an addition to its portfolio tailored for tasks including code review, tool integration, monitoring security alerts, and answering billing-related queries. The company also released NeMo Switchyard, an open-source model-routing library intended to automatically direct tasks to the most appropriate models.
Separately, Nvidia has formed a coalition with other firms to develop and share tools addressing AI safety and cybersecurity challenges. The company also signed an open letter with other large technology companies endorsing open-weight models, arguing that keeping model weights accessible will prevent innovation from shifting entirely overseas.
Late last year Nvidia rolled out the third generation of its open-source model family as offerings from Chinese AI labs continued to expand. The Nemotron 4 effort represents the next step in that trajectory, though important milestones - including final training completion and an official release timetable - remain unresolved.
Clear summary
Nvidia is developing Nemotron 4, an open-source model family expected to include a largest model with at least 1 trillion parameters. Final training is incomplete and no release date has been announced, though some employees expect readiness by late fall. Nvidia concurrently launched Nemotron 3.5 Lightning and NeMo Switchyard and has joined industry safety and open-weight advocacy efforts.