Nemotron 3 Ultra (550B A55B): Benchmarks, Pricing & Context Window
Nemotron 3 Ultra (550B A55B) is a language model from Nvidia, released in June 2026, with a 262K-token context window, and pricing from $0.500/M input, $0.100/M cached input, $2.20/M output.
Nemotron 3 Ultra is NVIDIA's frontier-scale open model with 550B total / 55B active parameters, built for agentic reasoning, long-context analysis, tool use, and high-stakes RAG. It uses a hybrid Latent Mixture-of-Experts (LatentMoE)
Nemotron 3 Ultra (550B A55B) benchmarks
Capability tiers
Standing within each category, adjusted for leaderboard depth.
Real tasks performance
High-confidence performance for Nemotron 3 Ultra (550B A55B) across real-world prompt categories. Only 95% intervals at most 4 points wide are shown.
Performance by conversation depth
How Nemotron 3 Ultra (550B A55B) holds up as conversations get longer.
Quality Tracker
Nemotron 3 Ultra (550B A55B) Performance Across Datasets
Scores sourced from the model's scorecard, paper, or official blog posts
llm-stats.com - Sun Sep 20 2026Nemotron 3 Ultra (550B A55B) pricing
Providers
Nemotron 3 Ultra (550B A55B) starts at $0.500 per million input tokens and $2.20 per million output tokens via DeepInfra. Reused prompt prefixes cost $0.100 per million cached input tokens.
| Provider | Input $/M | Cached input $/M | Output $/M | Context in / out | TTFT p95 s | Output p5 c/s | Modalities in / out |
|---|---|---|---|---|---|---|---|
 DeepInfra | $0.500 | $0.100 | $2.20 | 262.1K/262.1K | — | — | / |
Cached input is the discounted price for prompt tokens served from a provider cache. TTFT is time to first token. Output is characters per second; p5 is the sustained floor exceeded by 95% of observed requests.
Nemotron 3 Ultra (550B A55B) model size
Nemotron 3 Ultra (550B A55B) has 550 billion parameters and was trained on 20 trillion tokens. See how it compares to other models in the same parameter range.
Nemotron 3 Ultra (550B A55B) context window
Input and output token limits for Nemotron 3 Ultra (550B A55B), plus how it ranks on long-context understanding.
Try now
huggleMake it with
Nemotron 3 Ultra (550B A55B).
Nemotron 3 Ultra (550B A55B) latency
Nemotron 3 Ultra (550B A55B) time to first token, sustained output throughput, and failed-request rate from live model usage over the trailing 7 days.
Nemotron 3 Ultra (550B A55B) examples
Recent arena outputs from Nemotron 3 Ultra (550B A55B), picked from the highest-ranked matchups.
Nemotron 3 Ultra (550B A55B) license
Nemotron 3 Ultra (550B A55B) is released under the OpenMDW License v1.1 license, which permits commercial use, has 550.0B parameters, has a knowledge cutoff of September 2025.
- License
- OpenMDW License v1.1
- Commercial use allowed
- Parameters
- 550.0B
- Knowledge cutoff
- September 2025
Open Model, Data and Weights (OpenMDW) License Agreement, version 1.1 - permits commercial and non-commercial use.
Nemotron 3 Ultra (550B A55B) resources
Official sources for Nemotron 3 Ultra (550B A55B): provider documentation, paper or system card, source repository.
Nemotron 3 Ultra (550B A55B) vs other models
The most-compared alternatives to Nemotron 3 Ultra (550B A55B) are GPT-5 High, GPT-5.2 Pro, Claude Opus 4.5. Open any pair side-by-side for benchmarks, pricing, context, and latency.
Models like Nemotron 3 Ultra (550B A55B)
Models ranked just above and below Nemotron 3 Ultra (550B A55B) by LLM Stats score.
FAQ
Common questions about Nemotron 3 Ultra (550B A55B).