ChinaGenAI API access

Connect your software

One API connection. Your choice of tools.

Use the same free API in apps, small tools and scripts that support a custom OpenAI-compatible Chat Completions endpoint. The service chooses the upstream model, from any supported provider.

Free API key issuance is paused

The free API pool is paused while upstream access and its cost limits are verified. Guides and local tools remain available. The settings below explain the connection; they do not supply a working key while issuance is paused.

Copy these three settings into your app

API base URL
https://chinagenai.com/v1
API key
Your ChinaGenAI site key — check key access
Model label
free — automatic routing
  1. In your app, choose a custom OpenAI-compatible provider or endpoint.
  2. Enter the base URL, your site key and free. If the app asks for the complete Chat Completions URL, use https://chinagenai.com/v1/chat/completions.
  3. Send one short test. Confirm it succeeds before connecting a larger workflow.

The model field is a client label, not a promise of a particular model. A prefilled model name is also routed through the same free service. Brand names in our guides help people find information; they do not lock this API to Qwen, a Chinese provider or an overseas provider.

Try a standard request

curl 'https://chinagenai.com/v1/chat/completions' \
  -H 'Authorization: Bearer YOUR_API_KEY' \
  -H 'Content-Type: application/json' \
  -d '{
  "model": "free",
  "messages": [
    {
      "role": "user",
      "content": "Reply with one short sentence."
    }
  ],
  "max_tokens": 256,
  "stream": false
}'

Replace YOUR_API_KEY locally with your site key. The request builder can prepare the command without sending it. We have not claimed a successful live free-pool call in this release.

Compatibility and current limits

Client featureCurrent gateway behavior
Model discoveryGET /v1/models; the free route appears when the pool is available.
Text chatPOST /v1/chat/completions; system, developer, user, assistant and tool messages.
Streaming clientsstream: true returns SSE-compatible chunks after the upstream response completes. It does not provide immediate token-by-token delivery.
Function tools and JSON modeForwarded in the standard chat format; actual model support must pass the upstream activation checks. The gateway does not execute your tools.
Browser toolsBearer-key requests and CORS preflight are supported on /v1/. Account and key-management endpoints remain separate.
Other OpenAI productsResponses API, embeddings, images, audio, files and fine-tuning are not exposed by this chat endpoint. Apps that require those APIs need their own supported connection.

When enabled: 20 attempts per account per UTC day, 2 active requests across the site, 48-hour keys, up to 6 messages and 6,000 combined input characters including tool schemas. Requested output is capped at 256 tokens. Long requests may truncate; check finish_reason. Failed attempts count, and the site budget can pause the pool. No ad click or referral is required.

If your app cannot connect

  • 401: replace an expired or revoked site key. A provider’s key cannot be used as a site key.
  • 400: check the request format, input size and supported endpoint. Some apps append /v1 automatically; avoid a duplicated path.
  • 429: wait for an active slot or the next UTC day. Replacing a key does not reset allowance.
  • 503: check service availability; paused issuance is not a software configuration error.
  • 502: the upstream failed or timed out. There is no automatic paid fallback.