Meta Business Agent: token-based billing has arrived, what to plan for before 1 October

Since 1 August 2026, Meta has been billing token-based fees for replies generated by its AI agent, Meta Business Agent, on WhatsApp, Instagram and Messenger. A second, bigger change lands on 1 October: service messages, free since 2024, will become chargeable too. If your business relies on WhatsApp to answer customers, here is what actually changes for your budget.

1. What changed on 1 August

Meta now charges for every message processed by Meta Business Agent using a flat, token-based fee that covers both AI processing and message delivery: $2 per million tokens. A typical message uses between 20,000 and 25,000 tokens, roughly 4 to 5 cents per message. Cost mostly depends on how complex the request is: a simple question, such as opening hours, might take 4 messages for 15 to 20 cents total, while a more involved request spanning several exchanges can reach 40 to 50 cents.

2. The real shift lands on 1 October

Starting 1 October 2026, Meta will also charge for every service message, meaning any reply sent within the 24-hour window that opens once a customer messages you, whether written by a person or a third-party AI tool. These messages have been free since November 2024. Meta must publish exact regional rates by 1 September 2026 at the latest, a date worth marking before you close your Q4 budget.

3. A free channel becomes a real cost line

For an SME running customer support on WhatsApp, the consequence is direct: every inbound conversation, free on Meta's side until now, will carry a cost. Businesses will then face a choice: pay the service message at the standard regional rate, or route the conversation through Meta Business Agent, where cost depends solely on token usage, regardless of where the customer is based. The two fees never stack: a single message is billed once, under one category or the other.

4. How to prepare before October

Three simple steps make this change manageable. First, measure the volume of service messages your WhatsApp Business number has sent over recent months, that's the baseline for your future budget. Second, test Meta Business Agent while it's still possible to compare its actual token consumption against your current support costs. Third, block time for a budget review in early September, once regional rates are published, to adjust your conversation scenarios before they take effect.

The AI view, humans first

Answering an opening-hours question, confirming an order, routing a simple request: these are tasks where an AI agent is efficient and, for now, inexpensive. But every reply now carries a measurable price, and the more complex a conversation gets, the more tokens it consumes. Deciding which flows to automate, which to keep human, and where to set the balance between cost and service quality remains an expert judgment call. That's our stance: AI executes. Expertise decides, and watches over it.