What AI moderation costs and how not to overpay

AI moderation is the only Scarabeus block that costs money as it works. The other rules run within your plan; here you top up a balance and spend it on messages that get checked.

The good news is that the spend is manageable, and in a busy chat the difference between "leave the defaults" and "set it up properly" is measured in multiples.

What makes up the bill#

You pay not for switching the block on but for every message that reaches the model. Which means the whole conversation about saving money is a conversation about which messages not to send there.

You can follow the spend on the history tab of the AI moderation card: the latest transactions, the average cost per message, how much went out over thirty days, and an estimate of how many more messages and days the remainder covers.

Settings that cut the spend#

Minimum message length#

The most underrated setting. In a living chat a huge share of messages are "ok", "+", "thanks" and emoji. Checking them is pointless: three characters cannot be spam.

Set a minimum length and that whole stream stops reaching the model.

Skip what has already been caught#

By default, if a message was already cut off by phrases, links, regex or length, the model is not called — the punishment is assigned, there is nothing to pay for twice.

There is a checkbox that makes it check anyway and possibly issue a second, independent punishment. It is rarely needed: switch it on deliberately, it doubles the spend on caught messages.

Who to check#

A separate setting lets you skip members with high, or only very high, trust. Regulars who have been writing without violations for a long time stop spending the balance, and checking stays on newcomers and suspicious accounts — where it does some good.

In a chat with an established core this is usually the biggest saving.

Albums#

Image checking is switched on separately, and it has a sub-option for albums. Keep the arithmetic in mind: an album of ten photos is ten checks, not one. If albums are common in your chat while spam arrives as text, there is no reason to enable them.

How many rules of your own#

Every rule of yours takes part in checking every message and makes processing more expensive. The dashboard warns about this in the interface. Three precise rules are cheaper and work better than ten kept just in case.

What saves money by itself#

Repeats are not billed twice. Spammers send the same message in dozens of copies; a repeat gets the ready answer and never reaches the model. Those checks are marked separately in the balance history — on a mass mailing you can see how much it saved.

Administrators and the allowlist do not go through the model at all.

So the balance does not run out at 3 a.m.#

A balance that ran out at night means a chat without AI moderation until morning. Hence auto top-up: you set a threshold and an amount, and when the remainder approaches the lower bound the balance tops itself up. A low-balance notification arrives either way.

New chats get their first dollar from us — enough to look at your own traffic and estimate the real spend before paying anything.

Where to start in a new chat#

1. Switch on AI moderation with the built-in topics and no rules of your own.
2. Set a minimum message length.
3. Enable image checking, leave albums off.
4. Live with it for a week and look at the history: average cost and spend over the period.
5. Then decide whether to skip trusted members and whether you need rules of your own.

Free blocks take part of the load off the model: how to protect a Telegram group from spam.