NVIDIA: Nemotron 3 Ultra
nemotron-3-ultraOverview
An open frontier model for reasoning and orchestration: 550B parameters, 55B active, a hybrid Transformer-Mamba MoE architecture with LatentMoE and native speculative decoding via MTP. Up to 1M context, tuned for long agentic workflows.
Consumption multiplier
The model coefficient is a usage multiplier. Actual reference-token deductions are available in request logs.
Performance
Gateway metrics for the last 24 hours.
Loading metrics…
FAQ
Are there any rate limits?
The default limit is request rate per key. Each key can be restricted by models, spend and IP in your dashboard.
How is billing calculated?
Usage is deducted per request, accounting for model, output and cache coefficients. The final reference-token usage is available in request logs.
Is streaming supported?
Yes, stream: true behaves exactly like in the OpenAI API — tokens arrive as they are generated.