NVIDIA is building open-source AI model with 1 trillion parameters
What's the story
NVIDIA is said to be working on the next generation of its AI models, The Information reported. The new family of models, dubbed Nemotron 4, is aimed at competing with top open-source models around the world. The flagship model in this series is expected to have at least one trillion parameters.
Development progress
Model's training still being finalized
While NVIDIA has not officially announced a release date for Nemotron 4, employees working on the project believe it could be ready by late fall. The company is still finalizing the model's training.
This development comes as NVIDIA continues to prioritize AI safety and open AI technologies, having recently partnered with other companies to create tools for improving these areas.
Industry impact
NVIDIA's push for open-source AI
NVIDIA is one of the few major US companies to release open-source models.
The move has gained traction this year as AI costs rise and affordable Chinese models get closer to the performance of top systems from leading American labs such as Anthropic and OpenAI.
Last month, NVIDIA signed an open letter with tech giants like Microsoft, supporting open-weight models for broader innovation access.
New releases
Recent advancements in Nemotron series
Separately, NVIDIA recently launched the Nemotron 3.5 Lightning model. The new addition is aimed at tasks such as code review, tool use, security alert monitoring, and answering billing questions.
The company also launched NeMo Switchyard, an open-source model-routing library that automatically determines the best-suited AI models for specific tasks.