How Auto picks a model for every message

Auto chooses a model for each message you send, so you can focus on the work instead of the model picker. A quick question can go to a faster model. Demanding work can go to a stronger model with more reasoning. Auto is available for every account in Babbily on the web, and named models work the same way they always have.

How Auto chooses

Before each reply, Auto 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 it is. Then it picks a model and a reasoning level suited to that work.

• It chooses from approved models. Auto only picks from models Babbily has approved for that kind of work. Which models those are, and how they’re ranked, can change as Babbily adds and tests models.

• It uses a dependable default when it can’t see everything. That happens when your message includes a file, when you work in a project, when you select a skill or when a Canvas artifact is open.

• Choosing is a separate, quick step. A dedicated model reviews the text of your message and recent conversation to classify the task. Babbily doesn’t charge your API usage budget for this step.

Auto chooses one message at a time. If you want the same model for every reply, or you want to set Fast mode or reasoning effort yourself, choose a named model instead.

Turning it on

Select the model pill in the composer, choose Auto at the top of the list and send your message. Each later message gets its own choice.

New chats open on your last choice in the picker, whether you start from the home screen, a project, News or Finance. If you haven’t chosen yet, or your last named model has been retired, new chats start on Auto. Your last choice is saved in your browser, and existing chats keep the model they used.

Seeing which models answered

Individual replies aren’t labeled with a model. Instead, an Auto chat shows Auto · Models used: below the composer, followed by every model that has written a reply in that chat. In the sidebar and Library, Auto chats show the Auto icon. Like every model, Auto can make mistakes, so double-check important answers.

What changes while Auto is selected

• Reasoning effort is hidden. Auto sets the reasoning level for each reply. Reasoning that the chosen model shows still appears in the thread.

• Fast mode isn’t available. Choose a named model that supports it when speed matters most.

• Tool mode works as usual. Auto, Manual or Off decides whether Babbily can use tools such as web search. It’s a separate setting from choosing Auto in the model picker.

• Explicitly selected tools need a named model. Commands such as /create-image, /create-video, Manual > Deep research and Plugin slash commands ask you to choose a named model. Turning on Manual > Image generation or Manual > Video generation, or simply asking in Auto tool mode, still works.

• Skills are available. When you select a skill, Auto uses its dependable default for that reply.

Retry, edit and long chats

Retry and Edit message choose again, so a new reply can come from a different model. Changing the picker while a reply is running only affects later replies, and choosing Stop while Auto is still choosing cancels the reply without a fallback.

A change of answering model doesn’t make Babbily summarize your chat. The context indicator measures against the smallest context window among the models Auto can use. In Temporary Chat, Auto picks a model with enough room for the whole conversation.

Auto and your usage budget

• Choosing a model doesn’t use your API usage budget.

• The reply does. Replies, tools and file reading count at the rates of the model Auto picked, so usage can vary from message to message.

• When your remaining budget runs low, Auto favors models that use less of it, still choosing only among approved models.

Where Auto isn’t available

Translate has its own model list, and Scheduled tasks run on the named model you choose. Everywhere else in Babbily on the web, including mobile browsers and Babbily added to your iPhone or iPad Home Screen, Auto sits at the top of the picker.

If a named model you were using leaves the picker, Babbily switches that chat to Auto and tells you. Nothing is resent automatically, so choose Retry if your last message didn’t get a reply.

Read the full docs ↗