Nemotron 3 Super (120B A12B)

Name: Nemotron 3 Super (120B A12B)
Author: NVIDIA

NVIDIA

Nemotron-3 Super 120B-A12B is a Mixture-of-Experts model by NVIDIA with 120 billion total parameters and 12 billion active parameters. It balances high performance with computational efficiency, delivering strong results in reasoning, coding, and knowledge tasks while requiring only a fraction of the total parameters per forward pass.

Key Specifications

Parameters

120.0B

Context

Release Date

March 1, 2026

Average Score

62.1%

Timeline

Key dates in the model's history

Announcement

March 1, 2026

Last Update

March 21, 2026

Today

October 5, 2026

Technical Specifications

Parameters

120.0B

Training Tokens

Knowledge Cutoff

Family

Capabilities

MultimodalZeroEval

Benchmark Results

Model performance metrics across various tests and benchmarks

Programming

Programming skills tests

SWE-Bench Verified

Codex • Self-reported

53.7%

Reasoning

Logical reasoning and analysis

GPQA

With tools • Self-reported

83.0%

Other Tests

Specialized benchmarks

HMMT 2025

With tools • Self-reported

95.0%

RULER

100 @ 1M • Self-reported

92.0%

AIME 2025

Without tools • Self-reported

90.0%

WMT24++

en→xx • Self-reported

87.0%

MMLU-Pro

— • Self-reported

84.0%

AA-LCR

Self-reported by the model provider • Self-reported

58.3%

Arena-Hard v2

Self-reported by the model provider • Self-reported

73.9%

Bird-SQL (dev)

Self-reported by the model provider • Self-reported

41.8%

BrowseComp

With Search • Self-reported

31.3%

Humanity's Last Exam

With tools • Self-reported

22.8%

IFBench

prompt • Self-reported

72.6%

LiveCodeBench

v5 2024-07 to 2024-12 • Self-reported

81.2%

MMLU-ProX

Average over languages • Self-reported

79.4%

Multi-Challenge

Self-reported by the model provider • Self-reported

55.2%

SciCode

Subtask • Self-reported

42.0%

SWE-bench Multilingual

OpenHands • Self-reported

45.8%

Tau2 Airline

Self-reported by the model provider • Self-reported

56.3%

Tau2 Retail

Self-reported by the model provider • Self-reported

62.8%

Tau2 Telecom

Self-reported by the model provider • Self-reported

64.4%

Terminal-Bench

Hard subset • Self-reported

25.8%

Terminal-Bench 2.0

Self-reported by the model provider • Self-reported

31.0%

License & Metadata

License

nvidia-open-model

Announcement Date

March 1, 2026

Last Updated

March 21, 2026

Articles about Nemotron 3 Super (120B A12B)

NVIDIA GTC 2026: A Trillion Dollars, a Talking Snowman, and the End of Inference as We Know It

Jensen Huang unveiled Vera Rubin GPUs, Groq 3 LPUs, DLSS 5, and a live Disney robot. Here's everything that matters from GTC 2026.

March 23, 2026

9 min

Similar Models

All Models

Llama 3.1 Nemotron Ultra 253B v1

NVIDIA

253.0B

Best score:0.8 (GPQA)

Released:Apr 2025

Nemotron 3 Ultra (550B A55B)

NVIDIA

550.0B

Best score:0.9 (GPQA)

Released:Jun 2026

Llama 3.1 Nemotron 70B Instruct

NVIDIA

70.0B

Best score:0.8 (MMLU)

Released:Oct 2024

Nemotron 3.5 Lightning (30B A3B)

NVIDIA

30.0B

Best score:0.8 (GPQA)

Released:Aug 2026

Price:$0.08/1M tokens

Llama-3.3 Nemotron Super 49B v1

NVIDIA

49.9B

Best score:0.7 (GPQA)

Released:Mar 2025

Jamba 1.5 Large

AI21 Labs

398.0B

Best score:0.9 (ARC)

Released:Aug 2024

Price:$2.00/1M tokens

Command R+

Cohere

104.0B

Best score:0.8 (MMLU)

Released:Aug 2024

Price:$0.25/1M tokens

LongCat-Flash-Chat

Meituan

560.0B

Best score:0.9 (MMLU)

Released:Aug 2025

Recommendations are based on similarity of characteristics: developer organization, multimodality, parameter size, and benchmark performance. Choose a model to compare or go to the full catalog to browse all available AI models.