AiReasoningSupport

What one model can do with extended thinking. Providers describe each model through this shape so the UI offers only the choices that change the request, and the request builders clamp to the same table.

Name	Type	Description	Notes
thinks	Boolean	Whether the model can think at all. False hides the whole control.	[required]
canDisable	Boolean	Whether off really turns thinking off. False means the model thinks always and off only drops to its lowest depth (or leaves the default depth, where there is no knob).	[required]
depths	List	Depths the model distinguishes, lowest first. Empty when thinking is an on/off switch with no depth (or the model doesn't think). A level not listed is clamped to the nearest one — see clampReasoningLevel.	[required]
defaultDepth	AiReasoningDepth	The depth the model runs at when nothing asks for one — what a stored off means on a model that cannot be switched off. Known only where a catalogue reports it (OpenRouter's default_effort); otherwise DEFAULT_REASONING_LEVEL clamped to depths is assumed.	[optional] [enum: low, medium, high, max]