Chat Completions#
Chat Completions sends an ordered message history to a live Provider and returns the OpenAI-compatible choices response. Use it for applications already built around /v1/chat/completions, including streaming, tool calling, and structured output where the selected route supports them.
Find a compatible route#
Use Inference Models or the authenticated /v1/models
endpoint to find a live chat-completions route. Check supports_tools and
supports_structured_output in the Model Catalog.
A model name alone does not guarantee either capability across every Provider.
Call with HTTP#
curl https://inference.tensorplay.cn/v1/chat/completions \ -H "Authorization: Bearer $MEGA_TOKEN" \ -H "Content-Type: application/json" \ --data '{ "model": "mega/gpt-5.4-mini", "messages": [ {"role": "system", "content": "Be concise."}, {"role": "user", "content": "What is routed inference?"} ], "temperature": 0.2, "max_tokens": 128 }'MEGA uses the selected Provider route for the public model ID. Compatible request fields are handled according to that route's advertised capabilities.
Stream output#
Set stream to true:
curl -N https://inference.tensorplay.cn/v1/chat/completions \ -H "Authorization: Bearer $MEGA_TOKEN" \ -H "Content-Type: application/json" \ --data '{ "model": "mega/gpt-5.4-mini:fastest", "messages": [{"role": "user", "content": "Count from one to five."}], "stream": true }'The Provider's Server-Sent Events are preserved. Automatic failover is possible only before the first stream byte reaches the client. If the stream breaks later, decide at the application layer whether replay is safe.
Use tools#
Only choose a route whose catalog record has supports_tools: true. Send the standard OpenAI tool schema and inspect tool_calls in the assistant response:
{ "model": "mega/gpt-5.4-mini", "messages": [{"role": "user", "content": "What is the weather in Shanghai?"}], "tools": [{ "type": "function", "function": { "name": "get_weather", "description": "Read current weather for a city", "parameters": { "type": "object", "properties": {"city": {"type": "string"}}, "required": ["city"], "additionalProperties": false } } }]}MEGA routes and relays tool calls but does not execute them. Validate arguments, authorize the action, execute it in your application, then append the tool result to the next message history.
Request structured output#
For a route with supports_structured_output: true, send the Provider-compatible response_format. Prefer JSON Schema where the selected model advertises it:
{ "response_format": { "type": "json_schema", "json_schema": { "name": "answer", "strict": true, "schema": { "type": "object", "properties": {"summary": {"type": "string"}}, "required": ["summary"], "additionalProperties": false } } }}Capability validation is route-specific. If an explicit Provider rejects a supported-looking field, refresh the live catalog and capture the Inference-Id for Troubleshooting.