2.1 Service message pricing

From October 1, 2026, service messages are billed per message, at rates comparable to utility. The volume discounts that apply to utility and authentication don't apply to them.

4 min

What changes

From October 1, 2026, service messages become billable. The free-form replies you send inside the 24-hour window, the ones that have always cost nothing, will be charged per message.

Here's the pricing object on a support reply, before and after:

"pricing": {  "billable": false,  "billable": true,  "pricing_model": "PMP",  "type": "free_customer_service",  "type": "regular",  "category": "service"}

The free_customer_service type from Chapter 1.3 stops being how your support replies come back. The category doesn't move, which is easy to miss: if you count billable messages by category, service has been there all along, always free, and from October it's billed.

What doesn't change

  • Inbound is still free. However many messages customers send you, they cost nothing.
  • The window still works the same way. It governs what you can send. It no longer makes what you send free.
  • Templates are unaffected. Marketing, utility and authentication carry on as they are.

Volume tiers

Service messages are priced at rates comparable to utility and authentication. The difference is in the discounts:

No volume tiers on service

The volume discounts that reduce your effective rate as you scale apply to utility and authentication only. Service messages don't tier.

The higher your volume, the worse service looks relative to everything else on your bill. At scale, service messages can end up costing more per message than the utility templates sitting next to them.

For a business whose support volume grows with its order volume (most businesses), the service line grows linearly while the rest of the bill grows sub-linearly.

Who this hits hardest

The exposure is roughly proportional to how many outbound messages you send inside the window. Three profiles are most affected:

  • Support-heavy businesses. Long human conversations, many turns each.
  • Automated agents. An AI assistant that asks clarifying questions can easily take ten or fifteen turns to resolve something. Every one of those turns becomes a charge, and conversational agents are chatty by design.
  • Anyone who split messages for readability. Sending three short messages instead of one long one has been a free style choice. From October it's three charges instead of one.

If you're none of these, and you're almost entirely outbound templates with little back-and-forth, the change may barely register. The next chapter is how to find out where you sit.