Text and embeddings
Chat Completions, Responses and embeddings with provider-specific capabilities.
Use POST /v1/chat/completions for role-based messages, or POST /v1/responses for input. Both support JSON and streaming when the selected model lists that endpoint. POST /v1/embeddings accepts the selected embedding model and input.
{
"model": "MODEL_ID_FROM_CATALOG",
"messages": [{"role":"user","content":"Write a short welcome message."}],
"max_completion_tokens": 256
}Send a stable Idempotency-Key and Authorization: Bearer .... Use the model catalog's required fields, allowed values and limits. Tool calling, structured output, vision inputs and reasoning parameters are model/provider capabilities; endpoint compatibility does not guarantee every parameter on every route.
Provider-specific parameters pass through subject to reviewed constraints. Private routing and billing controls are reserved. Keep tool execution and validation on your application server.
See streaming, pricing and the playground. Feeef task APIs are separate from OpenAI Responses background mode.