Nemotron 3 by NVIDIA: status and timeline

Nemotron 3

Open-weight LLMs in three sizes using hybrid mixture-of-experts, activating a fraction of parameters per token. Designed for customization and multi-agent deployments.

DeployStage 4 of 5

Nano (30B, 3B active) released Dec 2025; Super (~100B, 10B active) and Ultra (~500B, 50B active) in H1 2026

Updated 15 Dec 2025Checked 10 Oct0 updates this week

Milestones

Next · Super and Ultra models release

Nemotron 3 Nano released on Hugging Face15 Dec 2025Complete.
Super and Ultra models releaseTarget Q2 2026Not yet reached.

Most important updates

NVIDIA releases Nemotron 3 Nano with open models and three trillion training tokens15 Dec 2025

Upcoming

Q2 2026Super and Ultra models release (next)

Current obstacles

Multi-agent coordinationDeploying multiple open models as cooperative agents requires new frameworks for orchestration, context-sharing and handoff.

Physics limits

Mixture-of-Experts routingRouting decisions and load balancing across experts adds complexity. Communication overhead can offset compute savings at very large scales.

How it works

3 parts

Mixture-of-Experts

Sparse activation

Only a fraction of parameters activate per token. Routes different inputs to specialized sub-networks, reducing compute while handling diverse tasks.

Open datasets

Training data released for fine-tuning

NVIDIA released its training and safety datasets so enterprises can customize Nemotron on proprietary data without sharing with NVIDIA.

Multi-agent focus

Reasoning and collaboration

Designed for scenarios where multiple models work together, coordinating through text. Smaller active parameter count keeps latency low.

Spec sheet

Spec	Nemotron 3
Nemotron 3 Nano size	30B (3B active)^R (reported)
Nemotron 3 Super size	~100B (10B active)^R (reported)
Nemotron 3 Ultra size	~500B (50B active)^R (reported)
Training datasets released	3 trillion tokens^R (reported)

^R reported by the company

Update log

1 update

Mon 15 Dec 2025

Major: NVIDIA releases Nemotron 3 Nano with open models and three trillion training tokensPressSoftware

About NVIDIA

The team behind Nemotron 3

NVIDIA

AI GPUs, rack systems, CUDA, networking

NVIDIA designs the GPUs, networking and software behind most AI training and inference. Its CUDA software made GPUs the default for AI. In October 2025 it became the first company worth $5 trillion.

Vera Rubin in production in 2026; cloud partners get it from the second half of the year.
Licensed Groq's inference technology and hired its leaders for about $20B (Dec 2025).
Plans to invest up to $100B in OpenAI, tied to at least 10 GW of NVIDIA systems (Sep 2025).

nvidia.com (opens nvidia.com)Company profile →

Founded: 199333 yrs
Headquarters: United States
Status: Public
Listed: NVDA (opens google.com)NASDAQ
Valuation: $5.4Tmarket capSep 2026
Staff: ~42,000Jan 2026
Works in: ComputingAI accelerators & CPUsAIRobotics
Coverage: 10 programs · 30 updateslatest 3d agochecked 10 Oct
Customers: OpenAICoreWeave
Partners: IntelGroq
Suppliers: TSMCSK hynix
People: Jensen Huang (opens en.wikipedia.org)Co-founder, President and CEOColette Kress (opens en.wikipedia.org)CFOBill Dally (opens en.wikipedia.org)Chief Scientist

Spec

Nemotron 3

Nemotron 3 Nano size

30B (3B active)^R (reported)

Nemotron 3 Super size

~100B (10B active)^R (reported)

Nemotron 3 Ultra size

~500B (50B active)^R (reported)

Training datasets released

3 trillion tokens^R (reported)