Models and gateway names
The model field of /v1/chat/completions selects what runs the request. It accepts three kinds of value:
| Value | Example | Runs |
|---|---|---|
| Integrated model name | @openai/gpt-5-mini | The model directly, with no gateway configuration. |
| Gateway full ID | <GATEWAY_UUID> | The AI Gateway with that ID. |
| Gateway slug | support:50c3 | The AI Gateway whose slug matches, in the format name:final-id. |
Integrated model names start with @ and follow @creator/model-name. List them with GET /v1/models; the same list includes the gateways of your account.
Public API keys accept only a gateway’s full ID. Slugs and integrated model names are rejected; see Authentication.
An integrated model name can carry a provider routing suffix, such as @openai/gpt-6-luna#cheapest. See Inline routing.
Choose a model for a gateway #
Validate three points before putting a model into production:
- The model supports the input modalities you intend to send, such as image, audio, video, or file. Otherwise, enable multimodal pre-processing.
- The model supports function calling if the gateway uses tools,
QueryFunctionRAG, MCP, protocol functions, skills, or built-in tools. - The model accepts the parameters you configure. Some integrated models reject assistant prefill, temperature, stop sequences, or reasoning effort.
Replace a model #
A gateway lets you change the model behind a stable name, so client applications do not change when a model is retired. You still need to qualify the replacement with your own prompts and tools before switching. To tell a current model from an old one, see Deprecated models.
See pinning a model ID versus using an alias for a replacement checklist. For the GPT-4o and GPT-4.1 shutdown on January 14, 2027, see how to migrate to gpt-6-luna. For the GPT-5 and o3 shutdown on December 11, 2026, see how to migrate to GPT-5.6 Sol, Terra and Luna.