tinyanalysis

Leaderboard / NVIDIA

Nemotron 3 Nano 30B-A3B

NVIDIA · Nemotron 3 · released 2025-12-15 · data checked 2026-11-01

Nemotron 3 Nano 30B-A3B is a 31.6B-parameter Mixture-of-Experts model (3.5B active per token) from NVIDIA (United States), released 2025-12-15 under the NVIDIA Open Model license. It scores 62.6 on the Small LM Index, #16 of 34 medium models. It needs about 20.3 GB of memory at 4-bit with an 8K-token context and costs $0.10 per million tokens through the cheapest API provider.

Small LM Index
62.6
#16 of 34 scored medium models
Likely 58.9–66.0 · from 6 of 6 core benchmarks
Parameters
31.6B
3.5B active · MoE
Price per 1M tokens
$0.10
$0.060 in / $0.24 out · blended 3:1
Context window
1M
tokens

Benchmarks

Each bar shows this model's score. The tick marks the best medium model on that benchmark. Sources: 8 reported by NVIDIA. How we label sources.

MMLU-Pro Lab-reported Knowledge78.3
Class best: Qwen3.6 27B 86.2
GPQA Diamond Lab-reported Knowledge73.0
Class best: Nanbeige4.3 33B 89.5
Humanity's Last Exam Lab-reported Reasoning10.6
Class best: Qwen3.8 27B 30.8
AIME Lab-reported Math · AIME 2025, no tools89.1
Class best: Ternary Bonsai 2 27B 95.8
LiveCodeBench Lab-reported Coding · LiveCodeBench v668.3
Class best: Qwen3.8 27B 90.3
SWE-bench Verified Lab-reported Coding · SWE-Bench (OpenHands)38.8
Class best: Qwen3.6 27B 77.2
τ²-Bench Lab-reported Agentic49.0
Class best: MiniCPM-V 4.7 35B-A3B 84.7
IFBench Lab-reported Instruction following71.5
Class best: Sage-A1.5 81.6

Specifications

Lab
NVIDIA (US · United States)
Architecture
Mixture-of-Experts
Total parameters
31.6B
Active parameters
3.5B
Size class
Medium
Context window
1M tokens
Input modalities
text
Reasoning
Hybrid (thinking on/off)
License
NVIDIA Open Model
Openness
Open weights
Memory to run
~20.3 GB at 4-bit, 8K context (more sizes)
Hugging Face
1.7M downloads (30 days) · 840 likes
API providers
1 on OpenRouter
API price
$0.060 in / $0.24 out per 1M (OpenRouter, 2026-11-03)
Weights on Hugging Face ↗Try on OpenRouter ↗

Compare with

Ranked in

Sources

Something wrong here? Report it.

Run it locally

Uses ggml-org/NVIDIA-Nemotron-3-Nano-30B-A3B-GGUF (Q4_K_M). Swap the tag for another size: BF16, Q4_K_M, Q8_0.

Ollama
ollama run hf.co/ggml-org/NVIDIA-Nemotron-3-Nano-30B-A3B-GGUF:Q4_K_M
llama.cpp
llama-server -hf ggml-org/NVIDIA-Nemotron-3-Nano-30B-A3B-GGUF:Q4_K_M
Then open http://localhost:8080 for a chat window.
MLX (Mac)
pip install -U mlx-lm
mlx_lm.chat --model mlx-community/NVIDIA-Nemotron-3-Nano-30B-A3B-OptiQ-4bit
vLLM (GPU server)
vllm serve nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16
Full-precision weights, for a server GPU.

Open in LM Studio

Memory needed: weights, context cache and runtime buffers.

Quantization4K context32K context256K context
16-bit F1664 GB65 GB66 GB
8-bit Q8_035 GB35 GB36 GB
4-bit Q4_K_M20 GB20 GB22 GB

Check it against your GPU, Mac or phone →