Running Nemotron 3 Ultra in Your Terminal — A Free Setup Guide

550B parameters (55B active) · 1M token context · Free API · Mac & Windows · ~5 minutes

Clearing Up a Common Misconception First

You won't be downloading a 550-billion-parameter model onto your laptop — that would realistically require several high-end datacenter GPUs. What you're actually doing is much simpler: NVIDIA hosts Nemotron 3 Ultra on its own servers and gives you free API access to it. On your machine, you just install a lightweight terminal app called OpenCode, which acts as the coding agent that talks to the model.

In short: OpenCode runs locally, Nemotron runs remotely on NVIDIA's infrastructure, and your API key is what links the two together.

Step 1 — Install OpenCode

On macOS, open Terminal and run:

curl -fsSL https://opencode.ai/install | bash

This finishes in well under a minute — you'll see a progress indicator and confirmation once it's done.

On Windows, there's no single curl-style installer, so pick one of three routes (easiest first):

• winget (built into Windows 11, and available on updated Windows 10) — from PowerShell: winget install SST.opencode
• npm — install Node.js LTS first, then reopen PowerShell: npm install -g opencode-ai@latest
• Scoop, if it's already part of your setup: scoop bucket add extrasscoop install extras/opencode

One Windows-specific tip: use Windows Terminal rather than the legacy Command Prompt. OpenCode's interface relies on proper color rendering, and PowerShell running inside Windows Terminal is the combination that behaves correctly.

Step 2 — Confirm the Installation Worked

On Mac, the installer commonly reports that no config file was found for zsh — that just means your shell doesn't yet know where OpenCode's binary lives. Fix it with:

echo 'export PATH=$HOME/.opencode/bin:$PATH' >> ~/.zshrc
source ~/.zshrc

Then check:

opencode --version

A version number confirms success. If you get "command not found," the PATH change didn't take — close and reopen Terminal and try again.

On Windows, close PowerShell and open a fresh window first — new commands only register in a new session, and this alone resolves most installation hiccups. Then run:

opencode --version

If PowerShell still doesn't recognize the command, try once more in a brand-new window, or as a last resort reinstall via the npm method above. And if you hit a message about script execution being disabled, that's PowerShell's default security policy — run this once and reopen PowerShell:

Set-ExecutionPolicy -Scope CurrentUser RemoteSigned

Step 3 — Grab a Free NVIDIA API Key

This step is identical on both platforms, and it's where most people stumble, so it's worth doing carefully.

1. Head to the Nemotron 3 Ultra model page or go directly to API key settings.
2. Sign in (a personal Gmail account works fine).
3. Navigate to Settings → API Keys → Generate API Key.
4. Copy the key immediately — it starts with nvapi- and is only ever shown once.

A few gotchas to watch for:

• "No API Key Permissions" usually means you're signed into a company or SSO-managed organization where an admin controls key creation. Use the Switch Org option to select your personal account, or sign up fresh with a personal email if nothing else works.
• Can't find where to enter the OTP? The verification code goes into the same browser tab you started signing in from — not the terminal. NVIDIA sometimes opens the login flow in a separate window, so check behind your main browser. Avoid refreshing or navigating back, since that invalidates the code.
• Don't use the "Get API Key" shortcut on the model page — that generates a temporary key that expires within hours. Generate a proper long-lived one from the dedicated API Keys page instead.

Step 4 — Connect NVIDIA to OpenCode

From your project folder, launch OpenCode:

cd ~/Desktop/my-project
opencode

(On Windows: cd $HOME\Desktop\my-project then opencode. A shortcut: right-click an empty area in File Explorer and choose "Open in Terminal" to skip typing the path.)

Once inside OpenCode, run:

/connect

Select NVIDIA from the provider list and paste in your nvapi- key. It's stored locally, so you'll only need to do this once.

Step 5 — Select the Model

Still inside OpenCode:

/models

Type "nemotron" to filter the list, then select Nemotron 3 Ultra 550B A55B. Or set it directly:

/model nvidia/nemotron-3-ultra-550b-a55b

From here you can just start giving it instructions — for example, asking it to read through your project folder and explain what it does. It can read files, write code, execute commands, and iterate on its own errors.

Step 6 — Prefer Python? Use the API Directly

If you'd rather integrate the model into your own scripts, the API is OpenAI-compatible.

Save your key as an environment variable:

echo 'export NVIDIA_API_KEY="nvapi-your-key-here"' >> ~/.zshrc
source ~/.zshrc
pip install openai

(On Windows: setx NVIDIA_API_KEY "nvapi-your-key-here" — note that setx only takes effect in new terminal sessions.)

Then in Python:

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://integrate.api.nvidia.com/v1",
    api_key=os.environ["NVIDIA_API_KEY"],
)

stream = client.chat.completions.create(
    model="nvidia/nemotron-3-ultra-550b-a55b",
    messages=[{"role": "user", "content": "Write a Python script that renames every file in a folder to lowercase."}],
    temperature=1,
    top_p=0.95,
    max_tokens=16384,
    extra_body={"chat_template_kwargs": {"enable_thinking": True}},
    stream=True,
)

for chunk in stream:
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="")

Troubleshooting Cheat Sheet

What You're Actually Running

• Parameters: 550B total, 55B active per token
• Context window: 1M tokens
• Architecture: Hybrid Mamba-Transformer Mixture-of-Experts
• Cost: Free, via NVIDIA's API

It's built for sustained agentic work — planning, coding, tool use, and debugging across large codebases. It's described as the strongest open model to come out of a US lab, though not necessarily the top-scoring open model overall.

Source guide: AI with Unnati — Nemotron 3 Ultra Setup Guide