Meta Meters WhatsApp Agent Replies and Restores Service Fees

Meta began charging WhatsApp businesses $2.00 per million tokens on August 1, 2026 for every Meta Business Agent reply. A typical reply burns 20,000 to 25,000 tokens, which Meta prices at 4 to 5 cents. A second bill lands on October 1, 2026, when free-form service replies and in-window utility templates lose the waiver that has kept them at zero.

The token meter is already running on the WhatsApp Business Platform. The October change is a different product: it puts a per-message price on human and third-party bot replies that have been free since November 1, 2024, and it does so in markets where Meta’s own agent is not allowed to take the traffic.

Token Billing Already Started on August 1

On the Cloud API path, Meta Business Agent is no longer a free test. Meta’s developer docs set one global rate of $2.00 USD per 1 million tokens and fold AI processing plus message delivery into a single per-token charge for Agent messages. Those August token rates on WhatsApp now sit on live invoices, not on a preview card.

The small-business app path is still a separate product. Meta’s June 3, 2026 newsroom note said getting started is free there, with paid subscriptions coming later. Platform traffic is the lane that already pays tokens.

THE AGENT COST EXAMPLES

A four-bubble hours question is not a 4 to 5 cent event. It is a 16 to 20 cent thread if the agent answers four times at the typical burn. A ten-bubble repair chat lands at 40 to 50 cents. Cache tokens are not in Meta’s public worked examples, so the published range is the number that should go into a budget.

The 72-hour free entry point window still waives marketing, utility, authentication, and service delivery when a person taps through an ad or a call-to-action. Agent token use does not ride that waiver. Delivery can be free in that window; the thinking is not.

The October Bill Hits Every Other Reply

From October 1, 2026, any non-template reply that is not generated by Meta Business Agent is a service message, billed per delivery at the same market rate Meta already uses for utility and authentication templates. The 24-hour customer service window still controls whether a free-form reply is allowed. It will no longer control whether that reply is free.

Utility templates sent inside that same window have been free since July 1, 2025, the day WhatsApp moved from conversation billing to per-message billing. That in-window utility waiver ends on October 1, 2026 too. Order updates and shipping notes that a bot fires while a customer is still chatting become ordinary utility lines again.

Marketing templates and login-code templates were already billed. The new money is the reply lane that support teams and outside chatbots have treated as a free pipe for almost two calendar years of service traffic, and for 15 months of in-window utility templates.

Banks and Hospitals Cannot Switch to the Agent

Meta’s own agent docs shut whole industries out of the token bundle. A number is eligible only if the WhatsApp Business account is on the Cloud API, in good standing, in an allowed country, and in an allowed vertical. Finance, government, health, alcohol, gambling, over-the-counter drugs, and matrimony are out.

VERTICALS THE AGENT WILL NOT SERVE

• Finance: Banks, lenders, and insurers that already run OTP and support traffic on WhatsApp cannot move that desk onto Meta Business Agent.
• Government: Public agencies on Cloud API numbers stay on per-message rates after October 1, 2026.
• Health: Clinics, hospitals, and related care accounts are blocked from the bundled agent.
• Alcohol, gambling, OTC drugs, matrimony: The same eligibility gate applies, even if the account is otherwise clean.

Those are not niche senders. They are the desks that already buy authentication templates and that staff 24-hour reply queues. The product Meta is selling as round-the-clock support is closed to them, so October 1, 2026 is not an off-ramp. It is a new line item on the same numbers they already have.

What a Reply Costs in India

India is WhatsApp’s largest market, and industry estimates put its business-messaging take above $1 billion a year. Meta’s India utility rate, which the October service rate is built to match, is ₹0.115 per delivered message, about $0.0014. That is the unit cost for a human bubble or a rule-based bot bubble after the free tier.

WHO IS BILLED FOR WHAT

A typical Agent reply at 4 to 5 cents costs about 30 times India’s $0.0014 service rate. In a cheap market the “one charge that covers AI and delivery” pitch is a premium product, not a discount. In a high-rate market, a short Agent thread can undercut a long stack of billed service bubbles. The same global token price therefore does two jobs: it sells convenience where human replies are already expensive, and it leaves India paying for volume on the old per-message card.

Partners on X have been selling the Agent as a way around the October service fee. That pitch collides with the rate card. Agent tokens never sat inside the free service window, and they do not become cheaper than ₹0.115 just because service messages start billing. A shop that can answer with a template or a short scripted reply still wins on unit cost in India. A shop that wants the agent’s catalog, booking, and handoff tools is buying a different thing.

Meta Waived These Fees After the 2025 Switch

The October clawback only makes sense against the waivers Meta used to push businesses onto per-message billing. Conversation-based pricing, the old 24-hour session charge, ended on July 1, 2025. Meta then priced each delivered template by category and country, and it left two reply types at zero so inboxes would keep moving.

THE PATH TO THE TWO BILLS

1. November 1, 2024: Service messages, the free-form replies inside an open customer window, stop drawing a Meta charge.
2. July 1, 2025: Conversation billing ends. Per-message pricing starts, and utility templates inside the 24-hour window are free.
3. June 3, 2026: Meta expands Business Agent globally on WhatsApp and Messenger and says it will add Instagram, with a platform that can talk to Shopify, Zendesk, and Shopee.
4. August 1, 2026: Platform Agent replies move to the $2.00-per-million-token meter.
5. October 1, 2026: Service messages and in-window utility templates become billable again, with a new monthly free slice for service only.

Each waiver trained a habit. Support teams learned to answer in open windows without watching a meter. Utility bots learned to fire confirmations while the customer was still in the thread. The Agent launch then offered a bundled AI lane with a global price. The October date pulls the free habits back onto a rate card without pulling excluded industries onto the agent.

One Thousand Free Messages per Number

Meta is not billing the first service bubble on every number. Its pricing page adds a monthly slice of 1,000 free service messages per business phone number, unused units do not roll over, and the count resets each month. Charge starts at message 1,001 on that number.

THE NEW METER AT A GLANCE

• Agent rate: $2.00 per 1 million tokens, live since August 1, 2026, typical reply 4 to 5 cents.
• Service free slice: 1,000 messages per number per month, then the local utility-level rate.
• Utility in-window: billed again from October 1, 2026, with no matching free slice in Meta’s note.
• Installed base: Meta says more than one million businesses already run a Business Agent, on more than 1 billion daily business threads across WhatsApp, Messenger, and Instagram.

A single Cloud API number that sends 1,000 service replies in a month still pays zero on that lane. A shared inbox that sends 8,000 agent bubbles on one number pays for 7,000 of them, plus every in-window utility template, plus any Agent tokens if that number also runs Meta’s model. The cushion is real for a quiet line. It is thin for a national support desk that parks volume on one sender.

Third-Party Bots Face Two Meters at Once

Outside Meta’s agent, an AI reply has always been two products. The model vendor bills for inference. WhatsApp bills for delivery once service messages leave the free list. Meta’s pitch is that Business Agent collapses those into one token invoice. After October 1, 2026, that gap gets wider for anyone still on a Zendesk bot, a BSP chatbot, or a home-grown script, because the WhatsApp leg of that stack becomes a service line.

AI that lets every business show up for every customer as if they had an infinite team behind them.

Meta Newsroom, June 3, 2026

The same newsroom post said more than one million businesses were already using a Meta Business Agent on WhatsApp and Messenger, and that the platform can connect into Shopify, Zendesk, and Shopee so the agent can take actions rather than only talk. That is the lock-in in plain language: keep the inbox, the catalog, and the booking tools inside Meta, and the bill is one token rate. Keep a third-party brain, and WhatsApp delivery is a second invoice again.

Token pressure also changes how those agents will be written. A business that pays for every extra 20,000 tokens has a reason to shorten prompts and loosen guardrails, and the agent still sits on the business node, reading customer text in the clear rather than inside WhatsApp’s user-to-user encryption. That is a product choice, not a side note, once metered AI replies for businesses are the default rather than a trial.

On October 1, 2026, the 24-hour window still opens when a customer writes. It just stops being a free window after 1,000 service bubbles on each number, and it never was a free window for Meta Business Agent tokens. Banks, clinics, and public desks that cannot turn the agent on will pay the restored per-message rates on the same numbers they already use.