What exactly is "reasoning off"?
Modern models write a long private chain of thought before answering. You never see it, but you are billed for every token. Our switch tells the model to skip it, on the models that support it. For most tasks — classification, extraction, summarisation, chat, code — the answer is the same at a fraction of the cost. For genuinely hard reasoning, leave it on.
Do I need a Chinese phone number or ID?
No. Registering directly with the upstream providers requires Chinese identity verification. We have already done that. You only need an email address.
Is the API really OpenAI-compatible?
Yes. The endpoint is /v1/chat/completions and works with the official openai Python and Node SDKs, LangChain, LlamaIndex and anything else that speaks the OpenAI protocol. Just change base_url.
Which models are available?
We currently serve the flagship models from DeepSeek, Zhipu GLM and Moonshot Kimi, and we add more regularly. The live list in your console is always authoritative.
How do I pay?
Card payments are being onboarded now. Until then we accept USDT and can issue credit manually — contact
[email protected] and we will set you up within a few hours.
What if a provider goes down?
We run multiple independent providers with automatic health checks. If one degrades, traffic moves to the others. Same key, same endpoint — nothing changes on your side.