There are ~30 world-class AI models you can't use. Here's how to access them.

In 2026, there are roughly 30 Chinese AI models that match or beat Western models on specific benchmarks.

DeepSeek V4 Flash is the most-called model on OpenRouter. Qwen3.7 Max beats GPT-4o on Chinese reasoning. GLM-5.2 tops agentic coding. Kimi K3 handles 1M tokens natively.

Try signing up for any of them from outside China.

You'll need:

A Chinese phone number
Alipay or WeChat Pay
To navigate a UI that's in Chinese
To figure out which of their 30 model variants you actually need
Most developers give up at step one.

AICraft fixes this. It's an API gateway that connects to 130+ models — both Chinese and Western. One key, one endpoint.

from openai import OpenAI

client = OpenAI(
api_key="sk-...",
base_url="https://api.aicraftapi.net/v1"
)

Chinese models — no phone number needed

client.chat.completions.create(model="deepseek/deepseek-v4-flash", ...)
client.chat.completions.create(model="qwen/qwen3.7-max", ...)
client.chat.completions.create(model="z-ai/glm-5.2", ...)

Western models — same key

client.chat.completions.create(model="openai/gpt-5.6-sol", ...)
client.chat.completions.create(model="anthropic/claude-opus-4-8", ...)
You also get:

One bill instead of 6+
Automatic failover across providers
A router that picks the best model for each task
This is not for everyone. If you only use GPT and deploy in US-East, call OpenAI directly.

It's for developers who want access to the full landscape — not just the Western half.

Alpha is open. Join here:

👉 aicraftapi.net/waitlist