Back to Research

Research

Nemotron 3 Nano: The Small Model That Doesn't Act Small

30 billion total parameters. 3 billion active at any given moment. A hybrid Mamba-Transformer-MoE architecture that delivers 3.3x higher throughput than comparable models. Here's what the benchmarks actually say, where it leads, and where it falls short.

March 2026 | Dry Ground AI Research

30B

Total Parameters

31.6B with embeddings

3B

Active Parameters

Per token during inference

1M

Context Window

Native, not extended

3.3x

Throughput

Faster than Qwen3-30B-A3B

~30GB

VRAM (FP8)

Fits on a single A100

Hybrid

Architecture

Mamba-2 + Transformer + MoE

Architecture

Three systems in one backbone

Nemotron 3 Nano is not a standard transformer. It is a hybrid that combines three different architectural approaches into a single 52-layer backbone. Each layer type handles a different class of problem, and the combination delivers capabilities that none of them achieve alone.

The result: a model that processes 1M tokens natively, runs on a single GPU, and generates output at 126 tokens per second on an H200.

Mamba-2 Layers

23 of 52Sequence processing

State-space models that scale linearly with sequence length. When you are processing a 256K-token context window, this is the difference between "fits on one GPU" and "needs a cluster."

Transformer Attention Layers

6 of 52Precise recall

Pure Mamba can struggle with finding one specific fact buried in 100K tokens of noise. These attention layers provide high-fidelity retrieval when it matters most.

MoE Layers

23 of 52Efficient knowledge routing

Each token activates only 6 experts plus 1 shared expert out of 128 total. 30B parameters worth of knowledge, 3B of compute per token. The efficiency gains compound fast at scale.

Strengths

Where Nemotron 3 Nano leads

Against open models under 30B MoE, Nemotron 3 Nano takes the top spot on long-context retrieval, math reasoning with tools, and instruction following.

BenchmarkScoreWhy it matters
AIME 2025 (with tools)99.2Math reasoning with tool access. Near-perfect.
RULER @ 256K tokens92.9%Long-context retrieval accuracy. Best in class.
RULER @ 1M tokens86.3%Still retrieving accurately at 1M tokens. Most models fall apart here.
Arena Hard v267.7Head-to-head instruction following.
IFBench71.5Complex multi-constraint instruction following.

Comparison

Head-to-head: Nemotron 3 Nano vs Qwen3.5-35B-A3B

Two models, nearly identical on paper. Roughly 30B total parameters, roughly 3B active per token, both targeting single-GPU deployment, both open-weight. But the architectures are fundamentally different, and so are the results.

The SWE-bench gap is the one that stings. Fixing 39% of real GitHub issues versus 69% is not a minor difference. For automated code repair workflows, that is a different capability tier entirely. But benchmarks measure a model in isolation. In production, you rarely use one model alone.

BenchmarkNemotron 3 NanoQwen3.5-35B-A3BGap
MMLU-Pro78.385.3-7.0
GPQA Diamond73.084.2-11.2
SWE-bench Verified38.869.2-30.4
AIME 2025 (with tools)99.2--
RULER @ 1M tokens86.3%77.5%+8.8
Throughput (single H200)126 tok/s~38 tok/s3.3x

Performance

The speed advantage is real

On identical hardware (single H200, 8K input / 16K output), Nemotron 3 Nano delivers 3.3x higher throughput than Qwen3-30B-A3B. That is not marketing. That is architectural. Mamba-2's linear complexity versus attention's quadratic cost shows up every single inference call.

For agentic workloads with dozens of sequential LLM calls per task, this compounds. A 3.3x speed advantage on every call in a 20-step agent chain means your workflow finishes in a third of the time. Or you serve 3x the concurrent users on the same hardware.

NVIDIA designed this model for exactly this use case. Multi-agent systems generate up to 15x the tokens of standard chat interactions. Every token-per-second matters.

3.3x

Throughput advantage

vs Qwen3-30B-A3B

126

Tokens per second

Single H200 GPU

15x

Token multiplier

Multi-agent vs standard chat

Strategy

The Super + Nano deployment pattern

Nemotron 3 Nano is not meant to do everything. It is meant to do most things fast and cheap, while a bigger model handles the hard stuff. This tiered approach is how production AI systems actually work. No one runs GPT-5 for every API call. The economics do not support it. The latency does not support it. And most tasks do not require it.

Nano

30B total / 3B active

Routing, simple tasks, high-volume work

Fast, cheap, good enough for 80% of agent actions.

Super

120B total / 12B active

Complex reasoning, multi-step planning

Slower, more expensive, but significantly more capable for hard tasks.

Proprietary

Varies

Expert-level work

When you need frontier-class performance on a specific task.

Feature

Token budget: think less when you can

Nemotron 3 Nano supports configurable thinking budgets. You can toggle reasoning on or off and set a maximum token budget for the model's chain-of-thought process.

Not every request needs deep reasoning. A routing decision in a multi-agent system does not need 2,000 tokens of internal deliberation. A classification task does not need chain-of-thought. Controlling the reasoning budget reduces latency and cost on tasks that do not benefit from extended thinking.

This is a practical feature that most benchmarks do not capture, but operators notice immediately.

Deployment

Self-hosted deployment profile

NVIDIA provides official FP8 weights and TensorRT-LLM optimization out of the box. Framework support includes vLLM, SGLang, and TensorRT-LLM. The Mamba-2 architecture is well-supported across NVIDIA's stack, though non-NVIDIA inference frameworks are still catching up.

ConfigurationVRAM RequiredCompatible Hardware
BF16 (full precision)~60GBA100 80GB, H100, H200
FP8 (NVIDIA official)~30GBA100 40GB+, A40, RTX 6000
Q4 (quantized)~18GBRTX 4090, RTX 5090

Assessment

The straight assessment

Right choice when:

  • Throughput is your primary constraint
  • You need reliable long-context retrieval (256K-1M tokens)
  • You are building multi-agent systems with sequential calls
  • You want configurable reasoning depth per request
  • Deploying on NVIDIA hardware with TensorRT-LLM

Wrong choice when:

  • Raw coding capability matters most (SWE-bench 38.8 vs 69.2)
  • You need multimodal input (text only, no vision)
  • Broad knowledge retrieval is critical (MMLU-Pro trails by 7 pts)
  • Apache 2.0 licensing is a hard requirement

Takeaway

The bigger picture

The 3B-active-parameter class is becoming one of the most interesting spaces in open-weight AI. Models like Nemotron 3 Nano and Qwen3.5-35B-A3B prove that you do not need 70B+ active parameters to build production-grade AI systems. The MoE architecture has changed the math entirely.

For enterprises evaluating self-hosted options, the question is not "which model is best." It is "which model is best for each layer of my system." The answer is almost certainly a combination.

NVIDIA is betting that speed and efficiency win the infrastructure layer while quality wins the reasoning layer. Based on these benchmarks, that bet looks well-placed.

Want to talk self-hosted AI?

We run our own GPU infrastructure, deploy our own models, and build systems that work in production. If self-hosted inference is on your radar, let's talk.

Book a Consultation

We use cookies to improve your experience. Cookie Policy · Privacy Policy