nemotron-3-ultra-550b-a55b
NVIDIA/🇺🇸 US/chat
NVIDIA Nemotron 3 Ultra is an open frontier reasoning and orchestration model with 55B active parameters out of 550B total (hybrid Transformer-Mamba MoE). Text input/output with up to 1M context, built for long-running agentic workflows: agent orchestration, coding agents, deep research, and complex enterprise tasks. Strong at multi-step reasoning and planning with high-throughput inference. Part of the NVIDIA Nemotron family.MoreLess
Which id to call
nvidia/nemotron-3-ultra-550b-a55bThis exact deployment on NVIDIA, with no routing and no failover. Send it as the model field.
Input /1M
free
NVIDIA
Output /1M
free
no charge
Context
1.0M
66K output
Added
Jun 2026
chat
Capabilities 2/8
Provider rates
What nemotron-3-ultra-550b-a55b costs
Provider prices per 1M tokens, updated September 8, 2026.moreless
These are the upstream provider rates. Pay as you go adds 5%, or 0% if you bring your own keys, and there is no per-request fee. Prompt caching and routing change what you pay against these rates, not the rates themselves.
Input /1M
free
Output /1M
free
Cache write /1M
-
Cache read /1M
-
What a workload costs
- 100K input + 10K output
- free
- 1M input + 100K output
- free
- 10M input + 1M output
- free
At the rates above. This endpoint does not offer prompt caching, so repeated context is billed as new input every time.
model=
Which id to call
One base url, https://router.requesty.ai/v1, and one key for every id here.
nvidia/nemotron-3-ultra-550b-a55bThis exact deployment on NVIDIA, with no routing and no failover. Send it as the model field.
OpenAI compatible
Call it in three lines
Change the base url, use your Requesty key, set the model to any id on the left. Existing OpenAI SDK code needs no other edit.
123456789101112131415from openai import OpenAI client = OpenAI( api_key="YOUR_REQUESTY_API_KEY", base_url="https://router.requesty.ai/v1", ) response = client.chat.completions.create( model="nvidia/nemotron-3-ultra-550b-a55b", messages=[ {"role": "user", "content": "Explain quantum computing in one paragraph."}, ], ) print(response.choices[0].message.content)
Live from production
NVIDIA on nemotron-3-ultra-550b-a55b, measured
What this provider was measured doing on nemotron-3-ultra-550b-a55b across Requesty traffic.moreless
Whole-window figures, because that is the grain published per provider. They cover NVIDIA serving this model in every region it serves it from, so a region-pinned deployment shares them with its siblings. A figure is absent where no qualifying sample exists.
First token
3.95s
median
p95 wait
31.54s
slowest 5%
Output speed
53/s
median tokens
Cache hit
69.9%
of input tokens
Compare these against the other 1 provider serving nemotron-3-ultra-550b-a55b.
Reference
Specs and data terms
What the catalog reports for this deployment, and what the provider does with the traffic.
Released 2026-06-04
Benchmark scores
Benchmarks measure the model, not this endpoint, so they are the same wherever these weights are served.
Artificial Analysis Coding Index: a composite of coding evaluations including LiveCodeBench, SciCode and Terminal-Bench.
Graduate-level physics, chemistry & biology questions designed to resist Googling.
Artificial Analysis Intelligence Index: a composite of multiple evaluations measuring overall model capability.
Scores from official model cards, Artificial Analysis and public leaderboards. They measure specific skills and do not capture every aspect of model quality, so test on your own workload.
Same provider
More from NVIDIA
Newest first, on the same provider and the same key.
Reference
nemotron-3-ultra-550b-a55b questions
How much does nemotron-3-ultra-550b-a55b cost?
nemotron-3-ultra-550b-a55b is priced at free per million input tokens and free per million output tokens when accessed via Requesty. Those are the upstream provider rates: pay as you go adds 5% on top, or 0% if you bring your own provider keys.
What is the context window of nemotron-3-ultra-550b-a55b?
nemotron-3-ultra-550b-a55b has a context window of 1.0M tokens, with a maximum output of 66K tokens per response. That's roughly 1,398 words of input you can fit in a single prompt.
How does nemotron-3-ultra-550b-a55b perform on benchmarks?
nemotron-3-ultra-550b-a55b scores 86.7% on GPQA Diamond, 83.3% on τ²-Bench, 49.3% on Coding Index. See the full benchmark chart above for results across MMLU Pro, GPQA Diamond, SWE-Bench Verified, HumanEval, MATH, AIME, MMMU, and LiveBench.
What can nemotron-3-ultra-550b-a55b do?
nemotron-3-ultra-550b-a55b supports tool calling, extended reasoning. You can call it through any OpenAI-compatible client by pointing base_url to Requesty.
How do I use nemotron-3-ultra-550b-a55b with the OpenAI SDK?
Install the OpenAI SDK, set base_url to "https://router.requesty.ai/v1", set your API key to your Requesty key, and set the model to "nvidia/nemotron-3-ultra-550b-a55b". The Quickstart above shows Python, JavaScript and cURL snippets.
Can I run nemotron-3-ultra-550b-a55b through Requesty?
Yes. nemotron-3-ultra-550b-a55b runs through Requesty's OpenAI-compatible API, served from NVIDIA. You do not host the model yourself: point base_url at Requesty, set the model to "nvidia/nemotron-3-ultra-550b-a55b", and requests are routed to the upstream provider with automatic failover. The same key gives you 600+ other models too.
Call nemotron-3-ultra-550b-a55b through one endpoint
One key for this endpoint and 600+ other models. No markup on provider prices, automatic failover, caching built in.