Completions

Note

For schema-validated JSON output without OpenAI's json_schema mode, define a tool whose parameters are the JSON Schema you want and set tool_choice to that function ({"type": "function", "function": {"name": "..."}}). The model is then required to call the function with arguments matching the schema — read the JSON from tool_calls[0].function.arguments instead of content.

Note

Reasoning-tier models use a fixed internal temperature and may reject other values. If you get an error with these models, try omitting temperature from your request or removing any client-side default.

Note

Not every model you see in MeinGPT chat is automatically usable via the API. Which models are enabled for your organization is set by a workspace admin (see Workspace Configuration).

Call GET /models/v1 to retrieve the exact list of models enabled for your API key, and use the value from the id field verbatim as model in your request.

If you send a model ID that no longer exists (for example because the model was retired), the API rejects the request with 400 and the message Validation error: Invalid input at 'model'. Retired model IDs are not redirected to a successor model for API calls. Only stored references in assistants, workflows and chats are migrated automatically. Replace the ID with a current ID from GET /models/v1. Notices about retired models are usually emailed to workspace admins; share them with the teams that run API integrations.

If you send a model value that isn't enabled for your organization, the API responds with 403 Forbidden and an error message of this form:

{
  "status": "error",
  "message": "Model 'your-model-id' is not enabled for this organization."
}

This message is distinct from the general 403 error "completions API not enabled for organization" (which affects the entire completions API, not a single model). Free-tier models (currently GPT-5 Nano, API id gpt-5-nano) are exempt from this restriction and are always available.

Another 403 error occurs when the usage tier assigned to your account has reached its separate "API usage" limit or has €0.00 set there. This check runs before the model-enablement check, so it affects every paid model alike - switching models does not help. A workspace admin can assign you a usage tier with an API budget.