Use Public Inference

Open a Public Model

1. Open Model Inference → Public Inference.
2. Each card shows the model ID, access type (framework and specification for local, vendor for cloud), and status.
3. Click a card for details. Cloud models show the Inference API URL, provider, and access mode Cloud. Local deployments show the replica range and resource specification.
4. Cloud models do not link to a platform repository.

Playground Testing

When the service is Running, the detail page provides a conversation test:

1. Enter a prompt and adjust Temperature, Top-P, or Max Tokens as needed. OpenAI-compatible cloud channels usually accept Max Tokens only; other sampling fields are fixed upstream.
2. Use the Thinking mode switch:
   • Local deployment: when off, the request sends chat_template_kwargs.enable_thinking=false.
   • Cloud API: the request sends reasoning_effort as high (on) or low (off). Some cloud models (for example Kimi K3) cannot fully disable thinking; low only reduces thinking effort.
3. Replies that include reasoning show the thinking block separately.

API Calls

Switch to the API tab on the detail page for the current service URL and examples.

Local deployment

The URL is the shared instance endpoint. Usage matches the dedicated instance API. Private services require an access token.

Cloud API

Cloud models go through the console gateway. The request model must be the platform model ID registered on the channel (it may differ from the upstream ID):

curl https://<console-host>/aigateway/v1/chat/completions \
  -H "Authorization: Bearer <access-token>" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "<platform-model-id>",
    "messages": [{"role": "user", "content": "Hello"}],
    "reasoning_effort": "high"
  }'

Some deployments use /platform-model/aigateway/v1/chat/completions. Use the Inference API URL shown on the detail page.

Related Documentation

• Public Inference Overview
• Manage Public Inference
• Use Dedicated Inference Instance