NVIDIA
Nemotron 3 Super
Release Date 2026-03-11(Knowledge cutoff 2026-02-01)262K IN · 1M OUTOpen ModelAPI available
Nemotron 3 Super is NVIDIA's open hybrid Mamba-Transformer MoE model with 120 billion total parameters, activating just 12 billion for maximum compute efficiency. Its hybrid architecture integrates Mamba layers for sequence efficiency with Transformer layers for precision reasoning, delivering over 5× throughput compared to its predecessor. With a native 1M-token context window and NVFP4 precision optimized for Blackwell GPUs, it scores 85.6% on PinchBench — the best among open models — making it well-suited for complex multi-agent applications, software development, and agentic reasoning.
API cost
Market price (median)
$0.0825IN$0.425OUT
Price range2 providers
$0.08~$0.085IN$0.4~$0.45OUTIN
OUT
9/2310/8
SourceOpenRouterLast updated
Real-world usage
No.3511 places0.4% share · 264B tokens/wk
460B8/24
321B8/31
352B9/7
330B9/14
380B9/28
264B10/5
SourceOpenRouterLast updated
Output Speed
Very slowper 2,000 tokens
19tok/s
First Output 1.1stotal: 106s
SourceOpenRouterLast updated
User Comparisons
Writing Style
Answer length
Shorter side
top 63%
Vocabulary diversity
Plainer side
top 74%
Sentence rhythm
Varied side
top 24%
Language mixing
High
top 21%
List usage
Often
31% of lines
Heading usage
Occasionally
10% of lines
Reasoning share
Lighter side
10% of total
Lies (Hallucination)
Honest answers
50%
Correct 0Honest don't-know 2Lies 2
Lies without doubt
2 questions
Response Speed
instability ±294%
First response 12.0sTotal 28.4s
33 tok/s
18.4ms per token
User Reviews
- •
The output speed is super fast and the content is accurate too. What's especially unique is that it's learning about itself.
SourceAIB Inc.Last updated
Expert Performance Evaluation
Overall performance
All overall performance AI rankings12.8Reasoning effort: Thinking1361
Automation (agents)
All automation AI rankings10.3%Reasoning effort: Thinking67.8%
Documents and office work
All documents and office work AI rankings65.7%Reasoning effort: Thinking1350
Research and fact-checking
All research and fact-checking AI rankings139031.3%22.8%
Math and science
All math and science AI rankings20.8%Reasoning effort: Thinking3.1%137380.0%13681399
Specialized fields
137813931351136687.0%64.1%63.2%65.1%93.0%83.0%90.9%59.3%41.7%85.0%66.0%96.7%20.5%17.3%39.0%8.5%77.0%66.0%57.0%80.0%53.0%82.3%50.3%60.0%75.0%50.4%54.4%82.5%61.1%73.6%
See each evaluation’s detail page for its source and last updateData sources & removal requests
Related Info
Public
NVIDIA🇺🇸 Santa Clara1993$5.8T~36,000 (total)