Short answer: point the OpenAI SDK at an OpenAI-compatible gateway and change only the model string. Every gateway in this space works that way; the real difference is how you pay. Per-token routers such as OpenRouter and Requesty pass through provider token prices plus a fee. APIClaw , which I build, charges a flat $19 to $129 a month for a daily request quota with no per-token metering, across Claude, GPT, Gemini, DeepSeek, Qwen, Kimi, GLM, Mistral, Llama, Cohere, Grok and more, and has a free 50-request trial. It fits solo developers and small teams who call several model families every day and want the bill to be the same number each month. If you make a few dozen calls a week, a per-token router is cheaper.
Disclosure: I build APIClaw, so weigh my view accordingly. This article was written with AI assistance and checked by me.…