NVIDIA Nemotron 3: Nano, Super and Ultra, and Which One You Can Run

Run Nemotron 3 with Ollama

Start with Nano. Ollama's default nemotron-3-nano tag is the 30B model at Q4_K_M: a 24 GB file that fits a 24 GB graphics card with a short context, and is comfortable on a 32 GB card or a Mac with 32 GB or more of unified memory. The 4B Nano is the pick for laptops and 8 GB cards. Every tag below was checked against Ollama's registry on November 2, 2026.

# Nemotron 3 Nano 30B-A3B, 24 GB, 1M context
ollama run nemotron-3-nano:30b

# Nemotron 3 Nano 4B, 2.8 GB
ollama run nemotron-3-nano:4b

# Nemotron 3 Super 120B-A12B, 87 GB
ollama run nemotron-3-super

Nano thinks before it answers by default; the Ollama model page notes that reasoning can be switched off through the chat template at a small cost in accuracy on hard prompts. Sizes and context windows are from Ollama's nemotron-3-nano tag list and nemotron-3-super tag list. This site does not quote a tokens-per-second figure for Mixture-of-Experts models: speed depends on how many experts fire and where they sit in memory, so a bandwidth ceiling would mislead.

For tool-calling agents, look at the successor: Nemotron 3.5 Lightning is NVIDIA's later 30B-A3B model tuned for agent work, and it is on Ollama as nemotron-3.5-lightning.

Quick verdict

Nemotron 3 Ultra is the most capable open-weights model to come out of a US lab in this cycle, and the headline number — ~48 on the Artificial Analysis Intelligence Index — puts it well clear of the smaller open models (NVIDIA cites gpt-oss-120b at ~33 and its own Super at ~36 for context). The cleverness is in the architecture: a sparse 550B/55B-active MoE means it reasons at frontier quality while only paying the compute cost of a ~55B model per token, and the Mamba-Attention hybrid keeps the 1M-token context affordable.

The honest caveat is scale. "Open weights" does not mean "runs on your laptop" — at 550B parameters, Ultra is a server-class model. The tier most people will actually self-host is Nano (30B total, ~3.5B active, or the 4B) or the multimodal Nano Omni, with Super (120B / ~12B active) for mid-range boxes. If you want a frontier-class open model you can run on a workstation, look at DeepSeek V4 or GLM-5.2 alongside this. NVIDIA is the "leading US open weights" story here, not necessarily the absolute global leader.

Nemotron 3 Ultra specs at a glance

Sources: NVIDIA Newsroom "Nemotron 3 family" announcement, the NVIDIA Nemotron 3 Ultra Technical Report (for the 108-layer / 8,192-dim / 512-expert / top-22 details and the ~20T-token pretraining), and the Hugging Face model card. NVIDIA's December 2025 family press release rounds the lineup to "about 500B / up to 50B active"; the model card and marketed figure for Ultra is 550B / 55B active.

The architecture: why it's fast

Nemotron 3 Ultra is not a vanilla transformer. It stacks three ideas, each aimed at making a 550B model cheap to run:

The full Nemotron 3 family

Nemotron 3 ships in three sizes, plus a multimodal Nano variant. The smaller you go, the more realistically you can self-host it:

Nano Omni is the interesting one for builders: it's an open multimodal model that takes text, images, audio, video, documents, charts, and GUI screenshots as input on a single 30B-A3B hybrid MoE, and NVIDIA reports up to ~9× higher throughput than other open multimodal models. It collapses a typical vision + ASR + LLM stack into one model — handy for agents that need to "see and hear" without three separate services.

Sources: NVIDIA Newsroom and NVIDIA Technical/AI blogs (Nano Omni, Super, Ultra), the Hugging Face model cards, and Ollama's Nemotron 3 Nano readme (3.5B active of 30B; 128 experts plus 1 shared, 6 active per token). Parameter figures vary slightly between NVIDIA's rounded press copy and the Hugging Face repo names; we use the repo/marketed numbers where they differ. Not to be confused with the 2023 nemotron-3-8b models, an older and unrelated 8B release that shares the name.

Benchmarks & throughput

Most headline numbers here are NVIDIA's own claims or single-source third-party measurements, so treat them as vendor-reported until independent trackers settle. Where that's the case, we've flagged it.

Sources: NVIDIA Newsroom / Technical Blog, Artificial Analysis Intelligence Index, MarkTechPost architecture write-up, and DeepInfra's Nemotron 3 Ultra release post. NVIDIA reports the speedup figure; verify against independent benchmarks before relying on it.

Running Nemotron 3 locally — the hardware reality

This is the part the launch coverage glosses over. "Open weights" is real and valuable — you can download, fine-tune, and self-host them (Ultra under OpenMDW-1.1, Nano and Super under the NVIDIA Nemotron Open Model License) — but the tiers live in very different worlds:

File sizes are Ollama's; the hardware column is our fit estimate, not an NVIDIA spec — MoE models only activate a slice of their weights per token, but you still have to hold all of them in memory, so total-parameter count drives the VRAM you need. For most individuals and small teams, the practical Nemotron 3 you run at home is Nano or Nano Omni; Ultra is something you rent on a NIM/DeepInfra/OpenRouter endpoint and reserve self-hosting for when data residency demands it.

How to choose a tier

Where Nemotron 3 sits in the open-weight landscape

Nemotron 3 Ultra is the new high-water mark for a US open-weights release, but the open-model race is global. NVIDIA's own framing — "leading US open weights" — quietly acknowledges that the very top of the open-weight charts still includes Chinese labs. If you're shopping for an open model to actually deploy, it's worth lining Nemotron 3 up against the field rather than taking the launch-day index score at face value.

• →NVIDIA's previous open release, the dense Nemotron-70B, is a useful baseline for how far the MoE jump goes.
• →For a global picture of the open-weight leaders, see the best free local AI models.
• →If your workload is code, our best local AI models for programming guide ranks the open options head-to-head.
• →To pit Nemotron 3 against specific rivals, use the model comparisons hub.

Primary sources worth reading first-hand: the NVIDIA Newsroom announcement and the Nemotron model cards on Hugging Face.