Model Orchestrator was called Auto before. Only the name changed: it chooses models the same way, and your chats and settings carry over.Orchestrator is available for every account in Babbily on the web. If Orchestrator is not at the top of your model picker, refresh the page. Named models work the same way they always have.
How Model Orchestrator chooses
Before each reply, Orchestrator reads your latest message along with the recent conversation. It judges what kind of task it is, such as a quick question, writing, coding, math, data analysis, or research, and how demanding the task is. Then it picks a model and a reasoning level suited to that kind of work.- Orchestrator chooses from models Babbily has approved for that kind of work. The models Orchestrator can use, and how it ranks them, can change as Babbily adds and tests models.
- When two models do about as well, Orchestrator uses the one that uses less of your budget. Babbily tests each model on answer quality, speed, and how much of your API usage budget it uses. A kind of work moves to a model that uses more only when that model does clearly better.
- When Orchestrator cannot see everything that shapes the task, it still judges how demanding it is. When your message includes a file, when you work in a project, or when you select a skill, Orchestrator chooses before the file, project, or skill is read. It judges how demanding your request is from what you wrote and picks a model for general work at that level, so a hard request can still get a stronger model. When a Canvas artifact is open, when a file comes with no text, or when your remaining budget is low, Orchestrator uses its dependable default for these messages instead.
- Choosing is a separate, quick step before the reply starts. A dedicated model reviews the text of your message and recent conversation to classify the task. Babbily does not charge your API usage budget for this step.
Model Orchestrator on the Free plan
On the Free plan, every message uses Orchestrator, and tools use Auto tool mode. The model picker shows Choose any model with a paid plan, and choosing a named model, Manual tool mode, Deep Research, and video generation are part of the paid plans. Free accounts run one reply at a time. See plans.Use Model Orchestrator
1
Open the model picker
Select the model pill in the composer. Its tooltip reads Select model.
2
Choose Orchestrator
Orchestrator appears at the top of the list, above the named models.
3
Send your message
Orchestrator picks a model for this reply. Each later message gets its own choice.
Which model a new chat starts on
New chats open on your last choice in the model picker. This applies to new chats you start from the home screen, a project, News, or Finance.- Opening an older chat is not a choice. The composer follows that chat’s model, and your next new chat goes back to your last choice.
- Your last choice is saved in your browser. Logging out or switching accounts clears it, so new chats start on Orchestrator until you choose again.
- Existing chats keep their model. A chat that used a named model stays on that model until you change it. Your choice is saved with each chat.
- Your named-model settings are kept. Babbily remembers your last named model and reasoning effort. If you switch back to that same model, its Fast mode setting is also as you left it, so check the lightning control before you send if you do not want the faster path, which uses more of your API usage budget. Choosing a different named model turns Fast mode off.
See which models answered
Individual replies are not labeled with a model. Instead, an Orchestrator chat shows Orchestrator · Models used: below the composer, followed by every model that has written a reply in that chat. Replies that failed without any output are not listed. In the sidebar and Library, Orchestrator chats show the Orchestrator icon instead of a model icon. Like every model, Orchestrator can make mistakes. Double-check important answers.Controls while Model Orchestrator is selected
Files, projects, and other context
- Attachments. You can attach any supported file. Orchestrator doesn’t read the file when it chooses, so for a message that includes one, it judges how demanding the request is from your text and picks a model for general work at that level. Later messages in the chat are judged like any other, from the conversation, even though the file stays in the chat. If you send a file with no text, Orchestrator uses its dependable default. See file attachments.
- Projects and skills. Project instructions, project knowledge, and selected skills load after the model is chosen. In project chats and when you select a skill, Orchestrator judges how demanding your request is from what you wrote and picks a model for general work at that level. See Projects.
- Canvas. When a Canvas artifact is open, Orchestrator uses its dependable default so the artifact’s content is taken into account.
- Temporary Chat. Orchestrator picks a model with enough room for the whole conversation, because Temporary Chat does not summarize older messages.
- Voice Mode. A voice call on Orchestrator picks one model as the call starts and keeps it for the whole call, for lookups and detailed answers. The call bar names the model it picked. See voice conversations.
- Long chats. A change of answering model does not make Babbily summarize your chat. The context indicator measures against the smallest context window among the models Orchestrator can use. See context window.
Retry, edit, and stop
- Retry and Edit message choose again, so the new reply can come from a different model.
- Changing the model picker while a reply is running affects only later replies.
- Choosing Stop while Orchestrator is still choosing cancels the reply. Babbily does not send a fallback reply.
API usage budget
- Choosing a model does not use your API usage budget.
- The reply does. Replies, tools, and file reading count toward your budget at the rates of the model Orchestrator picked, so usage can vary from message to message. Documents, code, and sheets in Canvas are written by GPT-6.1 Terra, at its rate.
- When your remaining budget runs low, Orchestrator favors models that use less of it. It still chooses only among approved models.
Where Model Orchestrator is not available
- Translate. Translate has its own model list. Continue in chat opens with the Translate model selected. See Translate.
- Scheduled Tasks. Scheduled Tasks run on the named model you choose. See Scheduled Tasks.
When a model you used is retired
If the named model in your chat leaves the model picker, Babbily switches the chat to Orchestrator and shows a notice such as:Troubleshooting
I do not see Orchestrator in the model picker
I do not see Orchestrator in the model picker
Refresh the page. Orchestrator appears in the chat model picker on the web, but not in Translate or Scheduled Tasks. If the composer says
Models are temporarily unavailable., the model list could not load; refresh again in a moment. Named models keep working in the meantime.The reply came from a model I did not expect
The reply came from a model I did not expect
Orchestrator chooses for each message based on the task, how demanding it is, and your remaining API usage budget. Check Orchestrator · Models used: to see which models answered. To keep one model for the whole chat, choose that model in the picker.
I want Fast mode or a specific reasoning level
I want Fast mode or a specific reasoning level
Choose a named model that supports the control you want. Fast mode and Reasoning effort appear again as soon as you leave Orchestrator.
Model capabilities
Compare named models, reasoning effort, and Fast mode.
Chat and models
Learn the basics of choosing models and continuing threads.

