NVIDIA logo

Nemotron 3 Nano

NVIDIAOpen WeightsPending Human Review

Nemotron 3 Nano is a 31.6B parameter hybrid large language model developed by NVIDIA, released on December 15, 2025. It employs a novel Hybrid Mamba-Transformer Mixture-of-Experts (MoE) architecture, activating only ~3.2B parameters per token for efficient inference. Optimized for agentic AI, reasoning, and coding, it supports a 1 million token context window and features a 'Reasoning ON/OFF' toggle with a configurable thinking budget. NVIDIA has released the model with open weights, training data, and training recipes under the NVIDIA Open Model License.

2025-12-15
31.6B (Total), ~3.2B (Active)
Hybrid Mamba-Transformer Mixture-of-Experts (MoE)
NVIDIA Open Model License

Specifications

Parameters
31.6B (Total), ~3.2B (Active)
Architecture
Hybrid Mamba-Transformer Mixture-of-Experts (MoE)
License
NVIDIA Open Model License
Context Window
1,000,000 tokens
Max Output
128,000 tokens
Training Data Cutoff
Nov 2025
Type
text
Modalities
text

Benchmark Scores

MMLU78.56

Massive Multitask Language Understanding (MMLU) tests knowledge across 57 subjects including mathema...

MMLU-Pro is an enhanced benchmark with over 12,000 challenging questions across 14 domains including...

GSM8K92.34

Grade School Math 8K (GSM8K) consists of 8.5K high-quality grade school math word problems....

MATH82.88

A dataset of 12,500 challenging competition mathematics problems requiring multi-step reasoning....

A sample of 500 diverse problems from the MATH benchmark, spanning topics like probability, algebra,...

Evaluates code generation capabilities by asking models to complete Python functions based on docstr...

ARC91.89

AI2 Reasoning Challenge (ARC) tests reasoning through grade-school science questions....

Tests common sense natural language inference through completion of scenarios....

A balanced collection of culturally sensitive and culturally agnostic MMLU tasks designed for effici...

Multilingual Grade School Math (MGSM) extends GSM8K to 10 languages....

American Invitational Mathematics Examination (AIME) 2025 problems....

Graduate-level Problems in Quantitative Analysis (GPQA) evaluates advanced reasoning on graduate-lev...

A challenging benchmark of novel problems designed to test the limits of AI capabilities....

Software Engineering Benchmark (SWE-bench) evaluates models on real-world software engineering tasks...

Tool Augmented Understanding Benchmark (TAU-bench) evaluates models on their ability to use tools....

Evaluates models on their ability to use terminal commands to solve system administration tasks....

The first comprehensive evaluation of LLMs' function calling capabilities, testing various forms inc...

A multi-domain challenge set created by Scale AI to test models across diverse tasks....

Advanced Specifications

Model Family
Nemotron 3
Finetuned From
NVIDIA-Nemotron-3-Nano-30B-A3B-Base-BF16
API Access
Available
Chat Interface
Available
Multilingual Support
Yes
Variants
NVIDIA-Nemotron-3-Nano-30B-A3B-BF16NVIDIA-Nemotron-3-Nano-30B-A3B-Base-BF16NVIDIA-Nemotron-3-Nano-30B-A3B-FP8GGUF (community)GPTQ (community)
Hardware Support
CUDANVIDIA RTXNVIDIA H100NVIDIA B200

Capabilities & Limitations

Capabilities
agentic AIreasoningcodemathfunction callingtool useRAG
Known Limitations
Reduced accuracy on complex tasks if reasoning traces are disabled
Notable Use Cases
multi-agent systemscoding assistantdocument QAlong-context reasoning
Function Calling Support
Yes
Tool Use Support
Yes

Related Models