FitMyLLM
nemotron/Mixture of Experts

nemotronNemotron Cascade 2 30B-A3B

We're excited to introduce Nemotron-Cascade-2-30B-A3B, an open 30B MoE model with 3B activated parameters that delivers strong reasoning and agentic capabilities. It is post-trained from the Nemotron-3-Nano-30B-A3B-Base.

chatreasoningcodingmathtool_use
32B
Parameters (3B active)
250K
Context length
13
Benchmarks
14
Quantizations
Architecture
MoE
Released
2026-03-19
Layers
52
KV Heads
2
Head Dim
128
Family
nemotron

Quantization Options

Context length:
QuantBitsVRAM @ 16KQuality
IQ3_XXS3.25
13.6 GB
13.5 + 0.1 KV
low
IQ3_XS3.5
14.6 GB
14.5 + 0.1 KV
low
Q3_K_S3.64
15.1 GB
15.0 + 0.1 KV
low
IQ3_M3.76
15.6 GB
15.5 + 0.1 KV
low
Q3_K_M4
16.6 GB
16.5 + 0.1 KV
low
Q3_K_L4.3
17.8 GB
17.7 + 0.1 KV
moderate
IQ4_XS4.46
18.4 GB
18.3 + 0.1 KV
moderate
Q4_K_S4.67
19.3 GB
19.2 + 0.1 KV
moderate
Q4_K_M4.89
20.1 GB
20.0 + 0.1 KV
good
Q5_K_S5.57
22.9 GB
22.8 + 0.1 KV
good
Q5_K_M5.7
23.4 GB
23.3 + 0.1 KV
good
Q6_K6.56
26.8 GB
26.7 + 0.1 KV
excellent
Q8_08.5
34.6 GB
34.5 + 0.1 KV
lossless
FP1616
64.6 GB
64.5 + 0.1 KV
lossless

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

Deploying for a team or in production? Size GPUs, cost & scaling in Enterprise →
Ready to run this?Rent by the hour

Rent a GPU and run Nemotron Cascade 2 30B-A3B now

Spin up an A100 / H100 / 4090 in ~60s. Pay by the second. Cancel anytime.

Community Ratings

Loading ratings...

Benchmarks (13)

Run this model

Easiest way to get started·Beginners
Docs ↗
curl -fsSL https://ollama.com/install.sh | sh
$ollama run nemotron-cascade-2

Ollama's default build of this model. For Q4_K_M specifically, pick a tag from ollama.com/library/nemotron-cascade-2/tags.

Setup guide
>_

Auto-setup with fitmyllm CLI

Detects your GPU, recommends the best model, downloads it, and starts chatting — zero config. Benchmarks your speed and contributes anonymous data to improve predictions.

pip install fitmyllmthen run fitmyllmLearn more
Auto-detect GPULive tok/s in chatSpeed benchmarks9 inference engines

GPUs that can run this model

At Q4_K_M quantization. Sorted by minimum VRAM.

Apple M4 Pro (24GB)
24 GB VRAM • 273 GB/s
APPLE
$1399
NVIDIA L4 24GB
24 GB VRAM • 300 GB/s
NVIDIA
$2500
Apple M2 (24GB)
24 GB VRAM • 100 GB/s
APPLE
$999
Apple M3 (24GB)
24 GB VRAM • 100 GB/s
APPLE
$999
Apple M4 (24GB)
24 GB VRAM • 120 GB/s
APPLE
$699
NVIDIA Tesla M40 24 GB
24 GB VRAM • 288 GB/s
NVIDIA
NVIDIA Tesla P10
24 GB VRAM • 694 GB/s
NVIDIA
NVIDIA Tesla P40
24 GB VRAM • 347 GB/s
NVIDIA
NVIDIA RTX A5000
24 GB VRAM • 768 GB/s
NVIDIA
$2500
NVIDIA L40 CNX
24 GB VRAM • 864 GB/s
NVIDIA
$5000
NVIDIA L40G
24 GB VRAM • 864 GB/s
NVIDIA
$5000

Find the best GPU for Nemotron Cascade 2 30B-A3B

Build Hardware for Nemotron Cascade 2 30B-A3B
Spec sheet

Nemotron Cascade 2 30B-A3B — 32B MoE.

Specifications
Parameters
32B (3B active)
Architecture
Mixture of Experts
Context Length
250K tokens
Capabilities
chat, reasoning, coding, math, tool_use
Release Date
2026-03-19
Family
nemotron
VRAM requirements
QuantBPWVRAMQuality
IQ3_XXS3.2513.5 GB82%
IQ3_XS3.514.5 GB84%
Q3_K_S3.6415.0 GB85%
IQ3_M3.7615.5 GB86%
Q3_K_M416.5 GB88%
Q3_K_L4.317.7 GB90%
IQ4_XS4.4618.3 GB92%
Q4_K_S4.6719.2 GB93%
Q4_K_M4.8920.0 GB94%
Q5_K_S5.5722.8 GB96%
Q5_K_M5.723.3 GB96%
Q6_K6.5626.7 GB97%
Q8_08.534.5 GB100%
FP161664.5 GB100%
Benchmark scores
MMLU-PRO79.8
LiveCodeBench87.2
SWE-bench50.2
AIME92.4
GPQA Diamond76.1
HLE11.4
AA Intelligence28.4
AA Coding25.8
IFBench80.4
Terminal-Bench Hard21.2
τ²-Bench54.1
SciCode34.8
AA Long Context34.0
Run command

Run Nemotron Cascade 2 30B-A3B locally with Ollama — needs 20.0 GB VRAM at Q4_K_M:

$ollama run nemotron-cascade-2
Compatible GPUs
30 @ Q4_K_M