Descriptions:
Bart Slodyczka demonstrates how to run Claude Code entirely for free by routing it through OpenRouter instead of Anthropic’s paid API. The setup requires no local model installation, making it viable on low-RAM machines, Raspberry Pis, or cloud VPS instances where downloading models like the 14GB-VRAM-minimum GPT-OSS would be impractical.
The tutorial walks through creating a local `.claude` folder and `settings.json` file with four required variables: the OpenRouter API key, the Anthropic base URL redirected to OpenRouter’s endpoint, an auth token, and an explicitly blank Anthropic API key. Slodyczka shows how to scope this configuration either locally to a single project or globally across all Claude Code sessions. Free models tested include the Qwen 3 Coder 480B parameter model and OpenRouter’s general free model router, which aggregates providers including Step 3.5 Flash — a model ranked in the top 20% for coding by Artificial Analysis and already heavily used by OpenClaw deployments.
A practical limitation is documented firsthand: individual free models like Qwen 3 can hit upstream rate limits during peak demand, making the general free router a more reliable default. The video also flags an unconfirmed $10 historical credit requirement on OpenRouter accounts. The end result is a fully functional Claude Code environment backed by frontier-scale models at zero marginal cost per token.
📺 Source: Bart Slodyczka · Published March 08, 2026
🏷️ Format: Tutorial Demo
