Choose a model in the chat composer before sending a request.
Choose a Supported Model
- Open Choose model, or click the current model's name.
- Select an available option.
- Confirm Ask, Agent, or Plan and send your message.
The desktop does not choose a default for you. If no supported choice is active, attempting to send opens the picker instead. A previously saved choice can be remembered, but a retired choice is not silently replaced.
In compatible builds, the four choices are:
- GPT Astra
- GPT Sol
- Trace Genie 2.0 Flash
- Trace Genie 2.0
Model names do not establish that one is best for your circuit or that it has particular speed, accuracy, or image capabilities. Review the results regardless of your choice.
Choose for the Next Turn
You can select a model for your next prompt while a turn is running. That does not change the model executing the current turn.
A queued prompt retains the model it had when queued, along with its mode and attachments. Use the queue's editing or unavailable-model controls to change that prompt deliberately; changing the main picker is not enough. See Chat.
Images and PDFs
Attachment support and direct model vision are separate:
- In compatible builds, GPT Astra and GPT Sol support direct image input.
- Trace Genie 2.0 Flash and Trace Genie 2.0 are text-only. A supported image workflow may supply a description instead of the image directly.
Do not assume every host provides that description path or that a description preserves all small labels, measurements, or diagram details. If an image cannot be handled, use a supported choice or provide the necessary information as text.
PDF handling also depends on the attachment workflow; an accepted PDF is not proof that every page or figure was read. See Attachments.
If a Choice Is Unavailable
Check the displayed model and account condition. Select a supported option for a new unsent request, or for a queued message explicitly reported as not sent.
Do not replace or resend a possibly delivered prompt just to change models. Check the original task and design when delivery is unconfirmed; terminal errors and interrupted tasks have different recovery controls.
What the Picker Does Not Change
The picker selects the chat model for your request. It does not promise that component generation, indexing, layout, or every background utility uses that same model. Component Studio uses task-specific models selected by the service; it does not expose a separate user model picker. Plugin library AI review does have its own model picker: check that review's selection before starting.
The newer Ask Trace editor-tool finder uses Trace Genie 2.0 Flash independently of the chat picker where supported. It finds existing editor tools; it does not switch your active chat model or turn a lookup into approved CAD work.
Model selection does not grant project permissions or bypass Billing limits.