With FlowModels, you name the model yourself in every request, in the model field. The value is always the full identifier made up of provider and model, for example:
| Identifier | Provider | Model |
|---|---|---|
openai/gpt-5.6-luna |
OpenAI | gpt-5.6-luna |
anthropic/claude-sonnet-5-5 |
Anthropic | claude-sonnet-5-5 |
xai/grok-4.7 |
xAI | grok-4.7 |
groq/openai/gpt-oss-120b |
Groq | openai/gpt-oss-120b |
The split is at the first slash. Everything before it is the provider, everything after it is the model, which can itself contain slashes, as with Groq.
The identifiers above are examples. Your current model list is always what counts.
Where to find the identifiers
- In the dashboard: Under Models, the Model ID column shows exactly the value you send in the
modelfield. - Via API:
GET /v1/modelslists all the models your team can use. You can fetch a single model withGET /v1/models/groq/openai/gpt-oss-120b, and the slashes in the identifier simply stay in place.
More on prices and capabilities in FlowModels models and prices and What a model can do.
What doesn't work
- Short names without a provider, such as
gpt-5.6-luna, lead to404 model_not_found. The identifier always needs the provider in front. - Models that aren't in your list can't be called. The list contains only enabled models with a complete price.
- Models for the wrong endpoint: Chat models belong at
/v1/chat/completionsor/v1/responses, embedding models at/v1/embeddingsand transcription models at/v1/audio/transcriptions. Otherwise you get400 unsupported_for_model. - Switching to a different model: If the chosen model is currently unreachable or retired, FlowModels doesn't fall back to a different model. You get an error and decide for yourself, see Understanding FlowModels error messages.
For responses with a Deprecation or Sunset header or 410 model_deprecated, you'll find the details in FlowModels models and prices.