NVIDIA Nemotron 3 Ultra - Box Dev Docs

According to NVIDIA, Nemotron 3 Ultra is an open, text-only reasoning model designed for complex multi-step agents, long-context analysis, and high-accuracy reasoning over code, math, and science.

Model details

Item	Value	Description
Model name	Nemotron 3 Ultra	The name of the model.
Model category	Standard	The category of the model: Standard, Premium, or Ultra.
API model name	`azure__nemotron_3_ultra_nvfp4`	The name of the model that is used in the . You must provide this exact name for the API to work.
Compliance	N/A	Government compliance frameworks and authorizations applicable to this model.
Box Agent compatibility	Yes	Specifies whether the model is compatible with Box Agent.
Hosting layer	Microsoft Azure	The trusted organization that securely hosts the LLM.
Model provider	NVIDIA	The organization that provides this model.
Release date	June 4th, 2026	The release date for the model.
Knowledge cutoff date	September 2025 (pre-training), May 2026 (post-training)	The date after which the model does not receive information updates.
Input context window	1m tokens	The number of tokens supported by the input context window.
Maximum output tokens	Not specified	The number of tokens that can be generated by the model in a single request.
Empirical throughput	Not specified	The number of tokens the model can generate per second.
Open source	Yes	Specifies whether the model’s code is available for public use.
IP infringement protection	No	Box provides no intellectual property infringement indemnity for this model. Consider any potential IP issues that might arise from using the model’s outputs.

Additional documentation

For additional information, see official NVIDIA Nemotron 3 Ultra documentation.

Last modified on November 4, 2026