Archive · 12 Aug 2026
Nvidia developing trillion-parameter Nemotron 4 open-source AI models
Nvidia is a company that makes very powerful computer chips.
It also makes AI models, which are like robot brains that can read and answer questions.
Nvidia is building a new family of these models called Nemotron 4.
The biggest Nemotron 4 model is expected to have at least one trillion parts called parameters.
Parameters are like tiny dials that help the model learn.
Nvidia wants its models to be some of the best open AI models that anyone can use.
This matters because cheaper models from China are getting very close to the best American ones.
Nvidia is sharing its models openly and working with other companies to make AI safe and secure.
It also released a smaller helper model called Nemotron 3.5 Lightning and a tool called NeMo Switchyard that picks the best model for each job.
Nvidia is developing Nemotron 4, a new open-source AI model family whose flagship is expected to feature at least 1 trillion parameters, according to The Information.
The report, citing several employees, surfaced on Tuesday; Nvidia pointed to previous remarks about working on Nemotron 4 but did not confirm the details.
No release date has been announced and final training is not complete, but employees reportedly believe the model could be ready as early as late fall.
Nvidia last month joined a coalition to develop and share AI safety and cybersecurity tools and backed an open letter, alongside Microsoft, supporting open-weight AI models.
Nvidia also introduced Nemotron 3.5 Lightning for tasks such as code review and security-alert monitoring, plus the open-source NeMo Switchyard model-routing library.
- Who
- Nvidia, the chipmaking giant, with several employees working on the project cited by The Information.
- What
- Development of Nemotron 4, an open-source AI model family whose flagship is expected to have at least 1 trillion parameters, along with the release of Nemotron 3.5 Lightning and the NeMo Switchyard library.
- When
- Reported on Tuesday; no release date announced, though employees believe the model could be ready as early as late fall.
- Where
- Not specified in the articles.
- Why
- To rival top open-source models globally as AI costs soar and cheap Chinese models near the capabilities of leading US labs such as Anthropic and OpenAI.
Open-weight advocates
Security and safety skeptics
Open-weight AI models and cybersecurity risk
Open-weight advocates
Nvidia says open models strengthen safety and security, accelerate innovation, and should remain broadly accessible; it backed an open letter with Microsoft supporting open-weight AI models.
Security and safety skeptics
A spate of recently disclosed hacks involving autonomous AI agents has drawn attention to open models because they do not have curbs on cybersecurity use.
- Company
- Nvidia
- Model family
- Nemotron 4
- Expected flagship size
- At least 1 trillion parameters
- Release date
- Not announced; possibly ready by late fall, per employee reports
- Source of report
- The Information, citing several Nvidia employees
- Other release
- Nemotron 3.5 Lightning
- New open-source tool
- NeMo Switchyard model-routing library
- Open-model advocacy
- Backed open letter with Microsoft supporting open-weight AI models
