Loading
90 days of the enterprise, pro and team plans free, no card needed. See the plans
NVIDIA
nvidia/nemotron-3-ultra-550b-a55bCatalogue: Indexed in the reference catalogue; nothing on this deployment answers as this model yet.
Published by the vendor; the platform bills at the target's declared price.
NVIDIA nvidia/nemotron-3-ultra-550b-a55b
NVIDIA Nemotron 3 Ultra is an open frontier-reasoning and orchestration model from NVIDIA, with 55B active parameters out of 550B total (MoE). Built on a hybrid Transformer-Mamba mixture-of-experts architecture, it...
Nemotron 3 Ultra earns a request when it needs step-by-step reasoning, tool calls, very long inputs; a short, general request would go to MiniMax M3.1 Flash (free) instead - free per million input tokens against $0.60 here; and the hardest requests, where the quality floor rules it out, to Claude Opus 5.5 (batch). Your rules decide the floor; the trace shows the comparison for every request.
Lower input price and an intelligence index within 5 points.
Nothing on this deployment answers as this model yet. Connect your own provider under Model providers, or add a target to your catalogue - the router picks it up without retraining.
# targets.yaml - declare a target that runs Nemotron 3 Ultra
targets:
- id: nemotron-3-ultra-550b-a55b
kind: llm
name: Nemotron 3 Ultra
metadata:
model: nvidia/nemotron-3-ultra-550b-a55b # links the target to this catalogue entry
capabilities:
domains: [general]
supports_tools: true
context_window: 262144
cost:
usd_per_1k_tokens: 0.001500Input list price per million tokens; Nemotron 3 Ultra is the most expensive of 12 listed.
Photo: RDNE Stock project
Cached input reads $0.12 per 1M · weights on Hugging Face as nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16
A higher intelligence index, best first, with the price it comes at.
Other models by the same vendor in the catalogue.
The catalogue prints input and output prices and what three typical monthly workloads cost on each.
