The auto model

Send "model": "auto" and DigitSpace picks the model as well as the provider. It considers only models that can do what the request needs, at or above a quality level, and then ranks them with the project's routing strategy.

Quality levels

DigitSpace keeps a list of models for auto, each placed at one of three levels: basic, standard or advanced. The quality you ask for sets the lowest level allowed.

Set the project's quality under Routing → Auto model quality in the dashboard. A quality in the request overrides it for that request.

Requests that use tools or reasoning always get a standard model or better, even on auto:economy, because small models handle them poorly.

What auto checks

Before ranking, auto drops models that can't serve the request:

• images in the messages need a model with vision;
• tools need tool support;
• a json_schema response format needs structured output;
• reasoning_effort needs reasoning;
• the estimated input plus max_tokens must fit the context window.

If no model at the level qualifies, auto widens to the rest of its list, and then to every model.

Seeing what was chosen

The response's model field says auto. To see the model and provider that answered, read the x-digitspace-provider header, or open the request in Activity, which shows the model, the quality level used and why.