Get the best AI model for every task with Auto

Auto is a model router that reads each task and sends it to the model that best balances quality, speed, and cost. For each task, Auto chooses a model from multiple AI providers, based on Notion’s evaluations and the models your workspace allows. It uses faster models for quick questions and edits, and more capable models for complex work.

Auto is on by default in the model picker, in Notion Agent chat, and in a Custom Agent’s settings. When you use Auto, Notion picks the best model for each task. Here’s how it works:

• Auto matches the model to the task. Quick questions, edits, and summaries go to fast, efficient models. More complex tasks, such as in-depth research, writing code, or building Custom Agents, can go to more capable models.

• Auto can adapt mid-task. During a multi-step task, Auto can switch to a more capable model for demanding steps, then switch back for simpler work.

• Auto keeps up with new models. Notion evaluates new models for quality, speed, cost, and reliability before making them available. The models available through Auto may change. Learn how Notion evaluates models against common knowledge work →

• Auto follows your workspace settings. It only uses models your workspace allows that meet its requirements, including model controls and provider restrictions.

• Auto needs enough models switched on to work well. If an admin turns off too many models for the workspace, Auto can become less efficient. Admins will see a warning before a restriction impacts Auto.

You can take over model selection at any time. Pick one yourself when:

• You want to use the same model throughout your chat instead of letting Auto choose or switch models as the work changes.

• You prefer a specific model for a particular kind of work.

• Auto didn’t give you the result you need, and you want to try another model.

To choose a model

In Notion Agent chat:

1. Select Auto to open the model picker.

2. Hover over a model to compare speed, intelligence, and cost.

3. Select a model.

Once selected, Notion Agent will use that model for your chat. To switch back to Auto, open the model picker and select Auto.

In Custom Agent settings:

1. Open your agent’s Settings.

2. Go to the model section and open the model picker.

3. Select a model.

4. Select Save.

Once selected, the Custom Agent will use that model until you change it.

Workspace owners and admins can control which models members choose. If a model isn’t listed, you can ask an admin to turn it on. Learn how admins can manage AI models and member credit spend →

On Business and Enterprise plans, Notion Agent uses Standard credits when you use Auto and Select credits when you choose a model yourself. If you run out of either, your admin can allow you to keep using AI features with Notion credits.

Learn more about managing your AI credits →

For most everyday work in Notion, Auto can help your credits go up to 10 times further than picking a top model yourself. Auto matches the model to the job and spends less per task, so you can put your credits toward getting more done. In our testing:

• Auto's average cost to run was 87% lower than Opus 5.5.

• Polishing a doc or email used about 8 times fewer credits than building a detailed strategy doc or quarterly plan.

This table compares estimated credit use for common tasks with Auto and Opus 5.5. Your actual use may vary.

Credit use varies by scenario. For example, asking Notion Agent to polish an email in a long thread with complex info may use more credits. Even for tasks that typically use more credits, Auto produced results comparable to frontier models.

We evaluated Auto by replaying thousands of common, real-world tasks in a simulated Notion environment, then comparing how different models did. Here's how it worked:

1. We compared models two at a time. We showed a group of top models the work from two different models, side by side.

2. The group of judges picked the better result each time, using a set scoring guide.

3. We gave each model an Elo score based on all these head-to-head matchups.

An Elo score rates skill based on how a player does against others over many one-on-one matchups, like stack ranking chess players. It's a simple, reliable way to score models on real-world work.

Running every matchup by hand is expensive, so we trained an AI judge to closely match these rankings, letting us run many comparisons at once. Our scoring comes from that method.