NVIDIA
Nemotron 3 Ultra 550B
NVIDIA's 550B MoE flagship in the Nemotron 3 family.
Sales-assisted Early Access
How you can run it
On Token Factory
- Serverless
- Available
- Dedicated endpoint
- Available
- Modalities
- Text
- Context window
- 262K tokens
- Quantization
- NVFP4 Quantized
The model
- Developed by
- NVIDIA
- Family
- Nemotron
- Parameters
- 550B
- Licence
- OpenMDW License Agreement, version 1.1
Per-token rate for this modelDedicated ratesUpstream model card
The API is OpenAI-compatible. See the documentation
Before you request access
- What happens after I request access?
- We scope the workload with you, confirm what we can serve, set up your account and provision access.
- What does a dedicated endpoint commit me to?
- Capacity is allocated in whole 8-GPU nodes and carries a minimum term. We confirm sizing and term before anything is reserved.
- How are model changes and deprecations handled?
- Catalog changes are communicated before they take effect. Ask us for the current policy on this model.
What the call covers
Workload fit for Nemotron 3 Ultra 550B, which deployment mode suits it, what we can commit to serving and on what capacity, commercial terms, and onboarding.