NVIDIA/Mixture of Experts

NVIDIANemotron 3 Super 120B-A12B

Hybrid Mamba-2 + MoE. 512 experts, 22 active. Native 1M context with 91.75% RULER accuracy.

chatcodingreasoningagentic
120B
Parameters (12B active)
977K
Context length
4
Benchmarks
14
Quantizations
Architecture
MoE
Released
2026-03-11
Layers
88
KV Heads
2
Head Dim
128
Family
nemotron

Quantization Options

QuantBitsVRAMQuality
IQ2_XXS2.3836.2 GBlow
IQ2_M2.9344.4 GBlow
Q2_K3.1647.9 GBlow
IQ3_XXS3.2549.2 GBlow
Q3_K_S3.6455.1 GBlow
Q3_K_M460.5 GBlow
IQ4_XS4.4667.4 GBmoderate
Q4_K_S4.6770.5 GBmoderate
Q4_K_M4.8973.8 GBgood
Q5_K_S5.5784.0 GBgood
Q5_K_M5.786.0 GBgood
Q6_K6.5698.9 GBexcellent
Q8_08.5128.0 GBlossless
FP1616240.5 GBlossless

Select your GPU above to see speed estimates and compatibility for each quantization.

Benchmarks (4)

MMLU-PRO83.7
GPQA Diamond79.2
IFEval72.6
HLE18.3

Run this model

Easiest way to get starteddocs →
curl -fsSL https://ollama.com/install.sh | sh
$ollama run nemotron-3-super:120b-a12b-q4_K_M

Tag may need adjustment — check ollama.com/library/nemotron-3-super for available tags.

Setup guide

GPUs that can run this model

At Q4_K_M quantization. Sorted by minimum VRAM.

NVIDIA H100 SXM5 80GB
80 GB VRAM • 3350 GB/s
NVIDIA
$25000
NVIDIA H100 PCIe 80GB
80 GB VRAM • 2000 GB/s
NVIDIA
$25000
NVIDIA A100 SXM 80GB
80 GB VRAM • 2039 GB/s
NVIDIA
$10000
NVIDIA A100 PCIe 80GB
80 GB VRAM • 1935 GB/s
NVIDIA
$10000
NVIDIA A100 SXM4 80 GB
80 GB VRAM • 2040 GB/s
NVIDIA
$15000
NVIDIA A100 PCIe 80 GB
80 GB VRAM • 1940 GB/s
NVIDIA
$10000
NVIDIA A100X
80 GB VRAM • 2040 GB/s
NVIDIA
NVIDIA H100 PCIe 80 GB
80 GB VRAM • 2040 GB/s
NVIDIA
$25000
NVIDIA H100 SXM5 80 GB
80 GB VRAM • 3360 GB/s
NVIDIA
$25000
NVIDIA H100 CNX
80 GB VRAM • 2040 GB/s
NVIDIA
$25000
NVIDIA A800 PCIe 80 GB
80 GB VRAM • 1940 GB/s
NVIDIA
NVIDIA A800 SXM4 80 GB
80 GB VRAM • 2040 GB/s
NVIDIA
NVIDIA H800 PCIe 80 GB
80 GB VRAM • 2040 GB/s
NVIDIA
NVIDIA H800 SXM5
80 GB VRAM • 3360 GB/s
NVIDIA
NVIDIA RTX 6000D
84 GB VRAM • 1570 GB/s
NVIDIA
$7500
NVIDIA B200
90 GB VRAM • 4100 GB/s
NVIDIA
$30000
NVIDIA H100 NVL 94 GB
94 GB VRAM • 3940 GB/s
NVIDIA
$30000
NVIDIA H100 SXM5 94 GB
94 GB VRAM • 3360 GB/s
NVIDIA
$25000
RTX Pro 6000
96 GB VRAM • 1792 GB/s
NVIDIA
$8565
NVIDIA H100 PCIe 96 GB
96 GB VRAM • 3360 GB/s
NVIDIA
$25000
NVIDIA H100 SXM5 96 GB
96 GB VRAM • 3360 GB/s
NVIDIA
$25000
Intel Data Center GPU Max 1350
96 GB VRAM • 2460 GB/s
INTEL
NVIDIA RTX PRO 6000 Blackwell Server
96 GB VRAM • 1790 GB/s
NVIDIA
$9999
NVIDIA RTX PRO 6000 Blackwell
96 GB VRAM • 1790 GB/s
NVIDIA
$9999
AMD Instinct MI300A
120 GB VRAM • 5300 GB/s
AMD
$12000
Apple M4 Max (128GB)
128 GB VRAM • 546 GB/s
APPLE
$3999
AMD Instinct MI250X
128 GB VRAM • 3277 GB/s
AMD
$10000
Apple M1 Ultra (128GB)
128 GB VRAM • 800 GB/s
APPLE
$4999
Apple M2 Ultra (128GB)
128 GB VRAM • 800 GB/s
APPLE
$3999
AMD Radeon Instinct MI250
128 GB VRAM • 3280 GB/s
AMD
$12000

Find the best GPU for Nemotron 3 Super 120B-A12B

Build Hardware for Nemotron 3 Super 120B-A12B

Nemotron 3 Super 120B-A12B120B Parameter Mixture of Experts LLM

Model Specifications

Parameters
120B (12B active)
Architecture
Mixture of Experts
Context Length
977K tokens
Capabilities
chat, coding, reasoning, agentic
Release Date
2026-03-11
Provider
NVIDIA
Family
nemotron

VRAM Requirements

QuantizationBPWVRAMQuality
IQ2_XXS2.3836.2 GB65%
IQ2_M2.9344.4 GB75%
Q2_K3.1647.9 GB78%
IQ3_XXS3.2549.2 GB82%
Q3_K_S3.6455.1 GB85%
Q3_K_M460.5 GB88%
IQ4_XS4.4667.4 GB92%
Q4_K_S4.6770.5 GB93%
Q4_K_M4.8973.8 GB94%
Q5_K_S5.5784.0 GB96%
Q5_K_M5.786.0 GB96%
Q6_K6.5698.9 GB97%
Q8_08.5128.0 GB100%
FP1616240.5 GB100%

Benchmark Scores

GPQA Diamond79.2
MMLU-PRO83.7
IFEval72.6
HLE18.3

How to Run Nemotron 3 Super 120B-A12B

Run Nemotron 3 Super 120B-A12B locally with Ollama (needs 73.8 GB VRAM at Q4_K_M):

ollama run nemotron-3-super:120b-a12b

Compatible GPUs (30)

GPUs that can run Nemotron 3 Super 120B-A12B at Q4_K_M quantization: