Via nvidia.com
The GPU giant launched the Nemotron Coalition at GTC, pooling talent from Mistral AI, Black Forest Labs, and others to build frontier-class open models
Nvidia isn’t just selling the shovels in the AI gold rush anymore. It wants to dig, too.
The company announced the Nemotron Coalition at GTC on March 16, a collaborative effort with leading AI laboratories designed to produce open, frontier-level foundation models. The first output: a base model co-developed with Mistral AI on Nvidia’s DGX Cloud platform, which will be open-sourced and serve as the bedrock for the upcoming Nemotron 4 family.
What the coalition actually looks like #
The founding members read like a curated guest list for an AI power dinner. Mistral AI, the Paris-based startup that has become synonymous with efficient open models, is the lead co-developer. Black Forest Labs, the team behind the Flux image generation models, brings multimodal expertise. LangChain, the developer tooling company that has become infrastructure for thousands of AI applications, rounds out the initial roster.
The first model built through this partnership will run on DGX Cloud, Nvidia’s managed infrastructure service that gives partners access to clusters of its latest GPUs without needing to build and maintain their own data centers.
Jensen Huang, Nvidia’s CEO, framed the effort in characteristically grand terms. He said open models are essential for stimulating innovation and ensuring global participation in the AI revolution.
The road to Nemotron 4 #
The Nemotron 340B family arrived in June 2024, focused primarily on synthetic data generation. These models were tools for making other AI systems better, shared under an open model license that allowed broad commercial use.
Then came the Nemotron 3 family on December 15, 2025, with a sharper focus on efficiency for agentic AI workloads. The 120B Super model, the flagship of that generation, demonstrated coding and reasoning abilities competitive with some closed-source models.
Nemotron 4 is meant to close whatever remaining gap exists. The coalition structure is designed to accelerate that timeline by distributing the research burden across multiple organizations, each operating at the frontier of their respective specialties.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our