Player not loading? Watch on YouTube
This tutorial configures Claude Code Router to send Claude Code requests to two external inference providers. The router runs locally on port 3456 by default, but the demonstrated models run at the providers. Model inference is hosted by OpenRouter and Baseten.
The installation command is npm install -g @musistudio/claude-code-router. The presenter opens configuration with ccr ui, adds a provider API key and model, then saves and restarts the router. Launching with ccr code sets the environment variables for routing. Claude Code may still display Sonnet 4.6 even when another model handles requests.
The comparison uses Qwen 3.6 Plus through OpenRouter, with model ID qwen/qwen3.6-plus:free, and Nvidia Nemotron 3 Super through Baseten, with nvidia/Nemotron-120B-A12B. Both receive a request to turn LinkedIn profile text into a personal homepage. The presenter reports 2 minutes 17 seconds for Qwen and 5 seconds for Nemotron, and prefers Qwen's page design.
These are results from one task, not a general benchmark of either coding assistant. Qwen used a skill, which the presenter says may explain part of the design difference. The presenter shows $0.00 in OpenRouter logs for the Qwen run and warns that the free route has strict rate limits that can interrupt work.
