nvidia · Open weights
Nvidia Nemotron 3 Ultra 550B Thinking
Nvidia's Nemotron 3 Ultra 550B A55B model from the Nemotron 3 family. It uses a hybrid Mamba-Transformer MoE architecture. Provider-specific context limits vary, with the longest current route supporting up to 1M context. Thinking enabled.
- Reasoning
- Tools
- Parallel tools
- Input
- Text
- Output
- Text
- Discovery tier
- Open weights
Pricing
USD per 1M tokens- Input
- $0.50
- Output
- $2.50
Reference prices as published by the model provider. API example
Specifications
- Context window
- 1M
- Max output
- 65.5K
- Average output (7 days)
- 654 tokens
- Longer than 58% of 468 models
Side by side
Nvidia Nemotron 3 Ultra 550B Thinking compared
Specifications, prices, speed and benchmark scores next to each other. Benchmarks below show every compared model too.
| Nvidia Nemotron 3 Ultra 550B ThinkingnvidiaThis model | Nvidia Nemotron 3.5 Lightning ThinkingnvidiaRemove | |
|---|---|---|
| Provider | nvidia | nvidia |
| Discovery tier | Open weights | Open weights |
| Added | Jun 4, 2026 | Aug 12, 2026 |
| Limits | ||
| Context window | 1M | 1M |
| Max output | 65.54K | 65.54K |
| Average output | 654 tokens | 6.1K tokens |
| Pricing, USD per 1M tokens | ||
| Input | $0.50 | $0.05Best |
| Output | $2.50 | $0.20Best |
| Cache read | — | $0.01 |
| Performance | ||
| Throughput | 79.0 tok/s | — |
| Time to first token | 2.1 s | — |
| Request success (24h) | 99.7% | — |
| Benchmark indexes | ||
| Intelligence Index | 22.9Best | 12.9 |
| Coding Index | 49.3Best | 26.8 |
| Capabilities | ||
| Reasoning | Yes | Yes |
| Tools | Yes | Yes |
| Parallel tools | Yes | Yes |
| Reasoning effort | — | none, high |
Public evaluations
Benchmarks
Independent evaluation results, with how each score ranks among the models in the Noumen catalog.
- Nvidia Nemotron 3 Ultra 550B Thinking
- Nvidia Nemotron 3.5 Lightning Thinking
Intelligence Index
- Nvidia Nemotron 3 Ultra 550B Thinking22.9
- Nvidia Nemotron 3.5 Lightning Thinking12.9
Nvidia Nemotron 3 Ultra 550B Thinking: Better than 56% of 326 scored models
Indexes
Coding Index
Composite of the coding evaluations
- Nvidia Nemotron 3 Ultra 550B Thinking: 49.3Top
- Nvidia Nemotron 3.5 Lightning Thinking: 26.8
Agentic work
AutomationBench-AA
Workflow automation, penalized for guardrail violations
- Nvidia Nemotron 3 Ultra 550B Thinking: 3.0%Top
- Nvidia Nemotron 3.5 Lightning Thinking: 0.8%
AutomationBench-AA tasks completed
Workflows finished end to end without a guardrail violation
- Nvidia Nemotron 3 Ultra 550B Thinking: 0.3%Top
- Nvidia Nemotron 3.5 Lightning Thinking: 0.0%
AA-Briefcasev1.1
Agentic knowledge work, rated head to head
- Nvidia Nemotron 3 Ultra 550B Thinking: 876 EloTop
- Nvidia Nemotron 3.5 Lightning Thinking: 506 Elo
GDPval-AAv2.1
Economically valuable occupational tasks, rated head to head
- Nvidia Nemotron 3 Ultra 550B Thinking: 1,013 EloTop
- Nvidia Nemotron 3.5 Lightning Thinking: 638 Elo
Document reasoning
GDP.pdf
Professional reasoning over PDF documents, all-pass rate
- Nvidia Nemotron 3 Ultra 550B Thinking: 5.0%Top
- Nvidia Nemotron 3.5 Lightning Thinking: 3.0%
AA-LCRv1.1
Reasoning across long documents
- Nvidia Nemotron 3 Ultra 550B Thinking: 79.3%Top
- Nvidia Nemotron 3.5 Lightning Thinking: 60.3%
MLCR-AA
Long-context reasoning over medical records
- Nvidia Nemotron 3 Ultra 550B Thinking: 5.6%Top
- Nvidia Nemotron 3.5 Lightning Thinking: 2.2%
Reasoning and knowledge
Humanity’s Last Exam
Expert-level questions across academic fields
- Nvidia Nemotron 3 Ultra 550B Thinking: 28.4%Top
- Nvidia Nemotron 3.5 Lightning Thinking: 10.6%
GPQA Diamond
Graduate-level science questions
- Nvidia Nemotron 3 Ultra 550B Thinking: 86.7%Top
- Nvidia Nemotron 3.5 Lightning Thinking: 74.3%
IFBench
Following precise output instructions
- Nvidia Nemotron 3 Ultra 550B Thinking: 81.4%
- Nvidia Nemotron 3.5 Lightning Thinking: —
Coding
Terminal-Benchv4.0
Practical software engineering in a terminal
- Nvidia Nemotron 3 Ultra 550B Thinking: 0.5%Top
- Nvidia Nemotron 3.5 Lightning Thinking: 0.5%Top
SciCode
Scientific computing in Python
- Nvidia Nemotron 3 Ultra 550B Thinking: 40.3%Top
- Nvidia Nemotron 3.5 Lightning Thinking: 32.1%
Legacy evaluations
Terminal-Bench Hard
Earlier terminal task suite
- Nvidia Nemotron 3 Ultra 550B Thinking: 36.4%
- Nvidia Nemotron 3.5 Lightning Thinking: —
τ²-Bench Telecom
Tool use in a customer-support simulation
- Nvidia Nemotron 3 Ultra 550B Thinking: 83.3%
- Nvidia Nemotron 3.5 Lightning Thinking: —
AA-LCR (unversioned)
Long-context reasoning, earlier grading
- Nvidia Nemotron 3 Ultra 550B Thinking: 79.3%Top
- Nvidia Nemotron 3.5 Lightning Thinking: 60.3%
Scores from Artificial Analysis, updated Oct 26, 2026. Artificial Analysis
Live routing data
Performance and reliability
Recent averages across requests routed for this model. Your own speed depends on prompt length, reasoning effort and load.
- Throughput
- 79.0 tok/s
- Faster than 52% of 254 models
- Time to first token
- 2.1 s
- Quicker than 75% of 254 models
- Prompt cache hits
- 18.9%
- Request success (24h)
- 99.7%
- Reliable
Cache hits are the share of input tokens recently served from a prompt cache across all requests. They are not a forecast of your own hit rate.
- Healthy (98% or more)
- Degraded (90–98%)
- Down (under 90%)
- No requests
Developers
Use it through the API
Noumen’s API is OpenAI-compatible: any Chat Completions client works with a base URL and a key.
curl https://www.noumen.xyz/api/v1/chat/completions \
-H "Authorization: Bearer $NOUMEN_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "nvidia/nemotron-3-ultra-550b-a55b:thinking",
"messages": [{ "role": "user", "content": "Explain this stack trace." }]
}'import os
from openai import OpenAI
client = OpenAI(base_url="https://www.noumen.xyz/api/v1", api_key=os.environ["NOUMEN_API_KEY"])
reply = client.chat.completions.create(
model="nvidia/nemotron-3-ultra-550b-a55b:thinking",
messages=[{"role": "user", "content": "Explain this stack trace."}],
)
print(reply.choices[0].message.content)nvidia
Related models
Other nvidia models, starting with the same family.