Descriptions:
Creator Nate Herk walks through a step-by-step method for running free and low-cost AI models inside Claude Code by swapping the default Anthropic backend for OpenRouter — a routing layer that aggregates 500+ models under a single API key. The key configuration involves editing Claude Code’s settings file to inject an environment variable pointing to OpenRouter’s endpoint and replacing the Anthropic API key with an OpenRouter key. The technique works in VS Code or terminal but not in Claude Code’s desktop app, which hard-codes the Anthropic backend.
The video highlights a newly released model called Stealth OX Alpha — from an anonymous provider — offered at $0 per million tokens on OpenRouter. Herk reports burning 61.6 million tokens in a single day of testing at a total cost of roughly $0.13, and demonstrates the setup producing a functional, on-brand e-commerce landing page with working product pages and a shopping cart. The tradeoff is significant latency: a task Herk estimates would take minutes with standard Claude required six hours with the free model, partly due to upstream idle timeouts that forced him to instruct the agent to work in smaller chunks.
Herk is candid about the anonymity risk: Stealth’s provider origin is unknown, and he cautions against passing sensitive business data through it. For experimentation, prototyping, or non-sensitive workloads, the approach offers a practical way to use the Claude Code agent harness at near-zero cost while the underlying model handles coding, file search, and verification tasks.
📺 Source: Nate Herk | AI Automation · Published August 21, 2026
🏷️ Format: Tutorial Demo
