Independent community resource — Nemotron3AI.com is not affiliated with, endorsed by, or sponsored by NVIDIA Corporation.

Guide · 7 min

Three official routes to Ultra.

Choose between checkpoints, a hosted API, and an Ollama cloud workflow based on control, setup effort and data-handling requirements.

Official checkpoints

NVIDIA’s Ultra research page links to post-trained NVFP4 and BF16 checkpoints, the base model, GenRM, training datasets, recipes and technical report. This route offers maximum control but demands substantial infrastructure.

View NVIDIA Ultra resources

OpenRouter free endpoint

The listed free endpoint provides API access with rate limits. OpenRouter currently warns users not to submit confidential or personal information and links to NVIDIA’s trial terms and privacy practices. Review those notices at the time of use.

Open the endpoint page

Ollama cloud model

Ollama provides CLI, cURL, Python and JavaScript examples for its cloud-hosted Ultra model. Its displayed context and service characteristics may differ from the underlying model’s maximum capabilities.

Open the Ollama library page