Pick a model, and control how hard it thinks
Picking a model by name
The model control under the composer lists every model your plan and trust mode make available, grouped and searchable by name. Each entry carries what it can do: Vision, Tools, Search, Image output, Video output, Reasoning.
Leaving the control on Auto picks the best model for each message. An explicit choice is never silently replaced with a different provider.
What the badges mean
- Upgrade: the model exists but your plan does not include it.
- Coming soon, not yet available: it is in the catalogue but not live.
- Unavailable right now: it is live but not answering at the moment.
- Auto (free) · community models, capacity varies: the free routing lane, whose capacity is not guaranteed.
Thinking and effort
How reasoning is controlled depends on the model. Some have no reasoning control, some always reason, some expose an on/off Extended thinking switch, and some a set of effort levels. The control under the composer says which applies: "Always on for this model", "Extended thinking is off for this model", or an effort picker whose hint is "Higher effort thinks longer before it answers."
Switching mid-conversation
Within the same trust boundary you can switch models in the middle of a conversation; the provider label updates before the next request. The control warns that switching here starts a new prompt cache, so the next message is processed fresh.
Continuing a Local chat on a BYOK provider is different: it is an explicit continuation with context selection, a payload preview, your consent, and a label naming the destination provider.
A model that was retired
If a conversation was saved with a model that no longer exists, the chat says so in place: that model is no longer available, new messages will use the named replacement, and you can pick a different one.
How many there are
Providers ship and retire models constantly, so the picker in the product is the current source of truth, never a number in a document. The provider-key catalog counts, and what they measure, are in Providers and models.