Your sovereign, private Nemotron 3 Ultra
Nemotron-3-Ultra-550B-A55B-NVFP4 is a frontier-scale LLM from NVIDIA (550B total / 55B active, hybrid LatentMoE architecture with interleaved Mamba-2 and MoE layers plus Multi-Token Prediction layers), designed for the most demanding agentic workloads: complex multi-step agents, long-context analysis, and high-accuracy reasoning over code, math and science. A reasoning model in the extra-large tier, served on ToothFairyAI's AU, EU and US endpoints — private by default, billed in US dollars per million tokens at the published rates below. No training on your data, no vendor lock-in.
How Nemotron 3 Ultra compares
Independent intelligence index for Nemotron 3 Ultra against every peer in the Extra-large tier (and the tier above where needed) — same source, same tests.
Throughput — tokens per second, median
Latency — end-to-end seconds for a 500-token answer
Capability indexes
Domain-weighted agentic performance for Nemotron 3 Ultra.
Capability index vs extra-large peers
The evaluations behind the index
Independent benchmark results for Nemotron 3 Ultra — Elo scores as published; pass rates shown as percentages.
Per-evaluation economics
| Evaluation | Score | Time / task | Output tokens / task |
|---|---|---|---|
| AA-Briefcase | 876.28 | 4.7 min | 47525 |
| GDPval-AA | 1013.25 | 2.7 min | 27244 |
| AutomationBench-AA | 3.0% | 0.6 min | 5609 |
| Terminal-Bench 4.0 | 0.5% | 5.1 min | 51610 |
| SciCode | 40.3% | 0.4 min | 3947 |
| Humanity's Last Exam | 28.4% | 3.7 min | 37164 |
| GDP.pdf | 5.0% | 0.3 min | 3105 |
| CritPt | 3.1% | 8.2 min | 82526 |
| AA-Omniscience | -40.0% | 0.1 min | 1196 |
| AA-LCR (long context) | 79.3% | 0.4 min | 4356 |
Run Nemotron 3 Ultra privately
One key, one endpoint, three jurisdictions.
- Size tierExtra-large
- Serving routeStandard
- Input rate$1.56/1M
- Output rate$4.68/1M
- Cached input rate$0.31/1M
- Context window262K tokens
- Max output16K tokens
- ReasoningYes — emits thinking traces
- Vision & video inputText only
- Tool callingSupported
- Data residencyAU, EU and US endpoints — processing stays inside the region you call