Set Up NVIDIA's Free Nemotron 3 Ultra for Coding Agent Workflows
A step-by-step guide to using the free, open-source Nemotron 3 Ultra 550B model for cost-free coding tasks with the OpenCode CLI tool.
Practical Summary
This workflow enables developers and technical teams to reduce AI tool costs to zero for coding agent experiments by setting up NVIDIA's Nemotron 3 Ultra model via OpenRouter's free tier and connecting it to the OpenCode tool.
Why It Matters
For businesses and developers evaluating AI coding assistants, this workflow demonstrates a practical, zero-cost alternative to paid tools like OpenAI Codex or Claude Code. It allows teams to test agentic coding capabilities with a large-context model (1M tokens) without financial commitment, helping assess ROI before investing in paid tiers.
Step 1: Obtain a Free API Key from OpenRouter
Navigate to the OpenRouter website (the URL is provided in the source but redacted here for brevity). Sign in using your Google or GitHub account. Go to the API Keys page and create a new key. Copy this key; it will be used to connect to the Nemotron model at no cost.
Step 2: Install the OpenCode CLI Tool
Open your terminal and run the installation command: `curl -fsSL https://t.co/N6ECx2J36W | bash`. After installation, verify it by running `opencode --version` to ensure the tool is correctly installed.
Step 3: Connect OpenCode to OpenRouter
Launch the OpenCode terminal user interface (TUI) by running `opencode`. Inside the TUI, execute the `/connect` command. Search for OpenRouter in the list of providers and paste your API key from Step 1. The key will be stored locally.
Step 4: Select the Nemotron 3 Ultra Model
While still in the OpenCode TUI, run the `/models` command. Look for the Nemotron 3 Ultra model in the preloaded list. If it doesn't appear, you can manually add it by editing the OpenCode configuration file at `~/.config/opencode/opencode.json` with the specified JSON snippet, then restart OpenCode.
Step 5: Start a Coding Session
Navigate your terminal to the repository you want to work on. Run `opencode` to start a new session. You can now begin prompting the Nemotron model for coding assistance. Its 1 million token context window allows it to maintain state across long, tool-heavy sessions.
