Story perspectives

Nvidia Unveils Nemotron 3 AI Models, Boosts Performance

12/15/2025

Summary

Story summary

Nvidia has launched the Nemotron 3 family of open-source AI models, with Nano (30B), Super (100B), and Ultra (500B).
The Nemotron 3 Nano is available now and offers cost reductions and improved accuracy, delivering up to 4x higher token throughput than its predecessor.
Early adopters include Accenture and Oracle.
Nvidia also introduces a NeMo Gym for testing models in simulated environments, aiming to foster innovation and collaboration in AI development.

History

Historical Context

OpenAI launched GPT-3 in May 2020, a revolutionary language model with 175 billion parameters, significantly surpassing its predecessor, GPT-2, which had 1.5 billion parameters. This advancement attracted a $1 billion investment from Microsoft, highlighting the escalating competition in AI technology. The release of Nvidia's Nemotron 3 family further illustrates the rapid evolution and growing significance of AI across various industries.

1 of 2

Story summary

Nvidia has launched the Nemotron 3 family of open-source AI models, with Nano (30B), Super (100B), and Ultra (500B).
The Nemotron 3 Nano is available now and offers cost reductions and improved accuracy, delivering up to 4x higher token throughput than its predecessor.
Early adopters include Accenture and Oracle.
Nvidia also introduces a NeMo Gym for testing models in simulated environments, aiming to foster innovation and collaboration in AI development.

1 / 2