Use Any Chinese AI Model Without a Chinese Phone Number

📝 Transcript

Welcome back to the NovAI Podcast. In episode one we talked about why Chinese AI models are so much cheaper. In this episode we tackle the thing that actually stops most developers from using them: access. Specifically, how do you call DeepSeek, Qwen, Kimi or MiniMax from outside China when the official signup keeps asking for a Chinese phone number you do not have?

Let me walk you through the wall you hit when you try to go direct. You find the model you want, you go to the official platform, and immediately you are asked for a mainland Chinese mobile number to verify your account. Then, when you try to add credit, the payment options are Alipay or WeChat Pay, both of which realistically need a Chinese bank account or identity. And for some enterprise tiers you are asked for business registration documents. If you are a developer in the US, Europe, Africa, India or Southeast Asia, that is three separate hard stops, and most people just give up and go back to paying ten times more for a Western model.

There are a few workarounds people try. Some rent virtual Chinese phone numbers, which is fragile and often against terms of service. Some ask a friend in China to register for them, which creates account ownership and payment headaches. And some use the open source weights and self host, which is genuinely powerful if you have the GPUs and the ops appetite, but for most teams running inference infrastructure is a full time job they did not sign up for.

The clean answer, and the reason gateways like NovAI exist, is to put a single OpenAI compatible layer in front of all of these models. Instead of juggling five different accounts, five verification flows and five payment methods, you create one account, get one API key, and point your existing OpenAI SDK at a new base URL. That is the whole migration. In code, you literally change two things: the base URL and the API key. The model name becomes deepseek v four, or qwen three max, or kimi k three, and everything else, your streaming, your function calling, your JSON mode, keeps working, because the gateway speaks the same protocol you already use.

On the payment side, this is where it gets genuinely liberating for international developers. Instead of Alipay or WeChat, you top up with a normal credit card, with PayPal, or with USDT crypto if you prefer. You get one balance and one consolidated bill across every model and every provider, which makes cost tracking and budgeting dramatically simpler than reconciling five separate invoices.

A quick word on trust, because handing your traffic to a gateway deserves scrutiny. The things to check are simple: does it bill on real upstream token counts rather than an inflated markup, does it store your keys and conversations, and is latency acceptable for your region. A good gateway is transparent on all three, bills at cost plus a clearly stated margin or none at all, and does not retain your prompts. Do that diligence once and you have unlocked the entire Chinese model ecosystem without a single Chinese phone number.

So there you have it. The barrier was never the models, it was the onboarding. Route through one OpenAI compatible key and the whole ecosystem opens up in an afternoon. The full step by step written guide, with code examples for Python and Node, is linked in the show notes. Thanks for listening, and go build something for less.