Notion's agents now cost money: which ones earn their credits

Bottom line. Notion’s Custom Agents left free beta on 4 May 2026: they now run on credits at $10 per 1,000, with nothing carried over month to month. If you built agents during the beta, the question is no longer how many you can build; it is which recurring agents are worth their monthly credit draw. For most small teams, that is a short list.

When Notion shipped Custom Agents on 24 Feb 2026, they were free to run. As of 4 May 2026 they are not: usage now bills against Notion credits at $10 per 1,000 credits, bought monthly on top of a Business plan that lists at $15 per user per month on annual billing, and unused credits do not roll over. Billing started on each workspace’s first billing date on or after 4 May 2026.

The milestone that turned the meter on came from Notion itself:

“In two months, you’ve created more than a million Custom Agents.”

— Akshay Kothari, Co-founder, Notion, in the 4 May 2026 post.

A million agents in two months is exactly the kind of free-tier enthusiasm that does not survive contact with a bill, which is the point: the credit model is Notion asking you to decide which of those agents are real.

What actually drives the cost

Notion does not publish a per-action price, and it is honest about why: a run’s cost depends on the task’s complexity, the data it touches, and the model it uses. So the useful mental model is not a price list. It is run frequency. A search agent you trigger occasionally barely moves the meter; a daily-brief agent that fires every morning draws steadily, every day, whether or not you read the output that day.

The table is qualitative on purpose, because Notion’s pricing page gives you a credit unit, not a per-agent rate. Your bill is the sum of how often each surviving agent runs, so the lever is frequency, not headcount.

The economics for a small team

A monthly credit pack at $10 per 1,000 is generous or tight depending entirely on how many of your agents run on a schedule. The beta taught a lot of small teams to automate everything; the credit model rewards the opposite. The agents worth keeping are the few that run often and save real time, a daily brief, inbox triage, a recurring status roll-up. The ones to kill are the high-frequency agents whose output you do not actually use, because those drain credits quietly.

This is the same discipline the bootstrapped-SaaS AI cost read applies to a stack: pay for what produces value at the frequency it runs, not for the comfort of having built it. If you set agents up during the beta, the Notion agents hub walkthrough and the no-code agent-building read are the build-side companions; the workspace-AI seat decision sits in the Notion AI versus ClickUp Brain comparison.

What changes this verdict

Cadence on this piece is 30 days, because Notion is actively tuning credit pricing and the developer platform around it. The three changes that would move the recommendation: Notion changing the $10-per-1,000 rate or introducing rollover; Notion publishing a clearer per-action cost that makes budgeting precise rather than estimated; or a bundled credit allowance landing inside the Business plan that changes the break-even. We re-test against Notion’s published pricing on or before 5 Jul 2026; the Holding-up record for OPS-092 carries any change, dated.