WhatsApp · Pricing
For nearly two years, answering a customer on WhatsApp cost you nothing. That ends on 1 October 2026, when Meta begins charging for every service message and for utility templates sent inside an open 24-hour customer service window. Here are the rates, the dates, and the five things worth doing before the deadline.
繁體中文版本：閱讀中文版
Key takeaways
As of 1 October 2026, every message a business sends on the WhatsApp Business Platform is billable. Templates and service messages are charged per message at rates that vary by market; Meta Business Agent messages are charged per token at one global rate of $2.00 USD per 1 million tokens.
| Category | Who sends it | Meter | Rate | Billable from | Volume tiers | Free in 72-hour entry window |
|---|---|---|---|---|---|---|
| Marketing template | Business initiates | Per message | By market | Live | No | Yes |
| Utility template | Business initiates | Per message | By market | Live — also charged in the 24-hour window from 1 Oct 2026 | Yes | Yes |
| Authentication template | Business initiates | Per message | By market | Live | Yes | Yes |
| Service message | Human agent, or your own / third-party AI | Per message | Same as utility & authentication, by market | 1 Oct 2026 | No | Yes |
| Meta Business Agent | Meta Business Agent Platform | Per token | $2.00 USD / 1M tokens, globally | 1 Aug 2026 | Not applicable | No — tokens still charged |
Source: Meta for Developers, Pricing for non-template messages, as published 17 September 2026. Meta may update rates up to once a quarter.
Meta is ending the exemption that made replies free. Until now, businesses paid to start a WhatsApp conversation but could reply free of charge inside the 24-hour customer service window. From 1 October 2026 those replies are billable, because Meta now sells its own AI layer that generates them.
Until now, WhatsApp pricing rewarded conversation. You paid to start one — a marketing, utility or authentication template — and once the customer replied, the 24-hour customer service window opened and everything you sent back inside it was free. Support was effectively a free channel bolted onto a paid one.
Meta is closing that gap in two steps, and the reason is the arrival of its own AI layer. In July 2026 Meta launched the Meta Business Agent Platform, which lets Meta generate the reply on your behalf. Charging for AI-generated replies while human replies stayed free would have been an odd place to land, so Meta is now charging for both — just on different meters.
Any message you send in response to a customer, without using a template, is a non-template message. Non-template messages can only be sent inside an open 24-hour customer service window, which opens and resets with each customer message. Those replies are now one of two things:
category: service). Metered per message. Billable from 1 October 2026.One rule worth committing to memory: Meta's category depends on who generated the message, not on what it says. If the reply did not come from the Meta Business Agent platform, it is a service message — whether a person wrote it, or your own model did.
Meta's 2026 changes land on three dates: the Meta Business Agent Platform launched 1 July 2026, its messages became billable 1 August 2026, and service messages plus in-window utility templates become billable 1 October 2026.
One global rate: $2.00 USD per 1 million tokens, covering both AI usage and message delivery as a single charge. A typical message consumes about 20,000–25,000 tokens, which works out to roughly 4–5 US cents per message.
Tokens are counted both for ingesting the customer's message — the prompt — and for generating your reply. Because the meter is tokens rather than messages, cost scales with how much work the reply takes:
| Interaction | Example customer message | Replies | Tokens | Estimated cost |
|---|---|---|---|---|
| Simple inquiry | “At what time do you open?” | 4 | ~80,000 ~20,000/message |
~16–20 cents |
| Complex interaction | “I'm stuck on step 3 of this assembly — can you walk me through it?” | 10 | ~250,000 ~25,000/message |
~40–50 cents |
Source: Meta. Quick answers cost less; rich multi-turn problem-solving costs more.
Service messages are charged per message, and by market the rate is the same as that market's utility and authentication message rate. Meta does not offer volume tiers on service messages.
Three details matter for your budget:
Only one charge per message
Every message you send has exactly one category and is charged once. A non-template reply is billed either as a Meta Business Agent message or as a service message, never both. This holds even when a non-template reply contains promotional content — Meta does not add a marketing template charge on top.
In Meta's own worked example, a single one-hour conversation that generated one charge before August 2026 generates three charges from 1 August 2026 and five charges from 1 October 2026.
A cart-abandonment reminder goes out, the customer asks two questions that the Meta Business Agent answers, a payment problem escalates to a human agent, and an order confirmation closes it out — all inside one hour.
| Sender | Message | Category | Charge from 1 Aug 2026 | Charge from 1 Oct 2026 |
|---|---|---|---|---|
| Business | Cart abandonment reminder template | Marketing | 1 marketing charge | 1 marketing charge |
| Customer | “Do you have this in a Small?” | — | Opens 24-hour service window | Opens 24-hour service window |
| Business | Agent answers the question non-template | Meta Business Agent | 1 agent charge | 1 agent charge |
| Customer | “Do you have this in white?” | — | — | — |
| Business | Agent answers the question non-template | Meta Business Agent | 1 agent charge | 1 agent charge |
| Customer | “I'm ready to check out, but my card is expired” — handed off to a human rep | — | — | — |
| Business | Human rep resolves the payment issue non-template | Service | No charge | new 1 service charge |
| Business | Order confirmation template | Utility | No charge | new 1 utility charge |
| Total | 3 charges | 5 charges |
Source: Meta. The same conversation in June 2026 would have produced a single charge — the marketing template.
It depends on conversation complexity. In Meta's published comparison for 10,000 AI-powered replies in Brazil, a lower-complexity third-party or in-house AI setup costs about $268, a higher-complexity one about $968, and Meta Business Agent about $400–500.
This is the question the pricing change is really asking, and the answer is not a single vendor. Meta published the comparison below for 10,000 AI-powered replies to customers in Brazil; it is worth reading carefully.
Estimated cost of 10,000 AI-powered replies — Brazil
Total cost including both AI usage and message delivery
| Per message | Service lower complexity |
Service higher complexity |
Meta Business Agent higher complexity |
|---|---|---|---|
| AI agent cost | ~2¢ | ~9¢ | ~4–5¢ combined $2.00/1M × ~25,000 tokens, charged once |
| Message delivery | 0.68¢ | 0.68¢ | |
| Total per message | ~2–3¢ | ~9–10¢ | ~4–5¢ |
| 10,000 messages | ~$268 | ~$968 | ~$400–500 |
Source: Meta. Third-party AI estimates are Meta's, based on publicly available information and benchmarks. The 0.68¢ delivery rate reflects the published utility and authentication rate for Brazil at the time of writing — substitute your own markets' rates, which differ substantially across APAC.
Read the shape rather than the numbers. For simple, high-volume deflection — opening hours, order status, return policy — running your own AI still comes out cheaper, because the per-message delivery charge is small and a cheap model handles the reply. For complex, multi-turn problem-solving, Meta's bundled token price undercuts a premium third-party model plus separate delivery.
The practical conclusion is that the economically correct architecture routes by conversation complexity — and that requires knowing, per conversation, what it costs you and what it earns you. Which brings us to the real problem.
Support stops being a free channel, longer conversations cost proportionally more, post-purchase utility flows see the sharpest increase, and Click to WhatsApp ad economics stay broadly intact.
The thing nobody has an answer for
When every reply carries a price, you need to know which conversations are worth paying for. That means joining message-level cost to customer-level value — spend per conversation against order history, lifetime value and segment. Most businesses cannot do this today, because the conversation log lives in a support tool, the orders live in Adobe Commerce or Shopify, and the campaign data lives somewhere else again. Data silos were an analytics inconvenience when replies were free. On 1 October 2026 they become a budgeting problem.
pricing_category: "SERVICE" to get the count for the last 90 days, broken out by market. Multiply by each market's utility rate. That is your new monthly line item — and the figure to take to whoever owns the budget.Service messages already exist as a category, so you can measure your exposure today: query the Pricing Analytics API with pricing_category set to SERVICE, or read the same category from message status webhooks.
Billable messages carry "billable": true, "pricing_model": "PMP", "type": "regular" and "category": "service" in the pricing object of the status webhook. The analytics response returns volume and cost for the timeframe, filterable by phone number and country.
Analytics and webhook values for Meta Business Agent messages are a separate set; Meta stated these would be published before charging began on 1 August 2026, so confirm the current reference before you build reporting against them.
category: service on the platform. Charged per message from 1 October 2026.In two steps. Meta Business Agent messages have been charged since 1 August 2026. Service messages — replies written by human agents or by your own or a third-party AI — are charged from 1 October 2026, along with utility templates sent inside an open 24-hour customer service window.
Service messages are charged per message, and by market the rate is the same as the utility and authentication message rate for that market. Meta does not offer volume tiers on service messages, so the rate does not fall as volume rises. Meta may update rates up to once a quarter, so confirm the current figure on Meta's rate card for every market you message.
A service message is any non-template message you send in reply to a customer that was not generated by Meta's Business Agent platform. It covers replies typed by a human agent and replies generated by your own AI or a third-party AI vendor. Meta's category is determined by who generated the message, not by its content.
Meta Business Agent is Meta's own AI agent platform, launched 1 July 2026, which generates replies to customers on your behalf. It is charged per token at one global rate of $2.00 USD per 1 million tokens, covering AI usage and message delivery as a single charge. A typical message consumes about 20,000 to 25,000 tokens, roughly 4 to 5 US cents per message.
No, not after 1 October 2026. Utility templates sent in response to a customer inside an open 24-hour customer service window have been free since 1 July 2025. From 1 October 2026 Meta charges for them on a per-message basis.
No. Meta does not offer volume tiers for service messages. Volume tiers continue to apply to utility and authentication template messages, so a high-volume support operation gets no discount curve on its replies.
Yes, for message delivery. Messages sent within the 72-hour free entry point window opened by Click to WhatsApp ads or Facebook call-to-action buttons remain free to deliver, and that is unchanged. The one exception is that Meta Business Agent messages are still charged for token usage inside that window.
No. Every message has exactly one category and is charged once. A non-template reply is billed either as a Meta Business Agent message or as a service message, never both. This holds even when a non-template reply contains promotional content — Meta does not add a marketing template charge on top.
It depends on conversation complexity. In Meta's published comparison for 10,000 AI-powered replies in Brazil, a lower-complexity third-party or in-house AI setup costs about $268, a higher-complexity one about $968, and Meta Business Agent about $400 to $500. Simple high-volume deflection favours your own AI; complex multi-turn problem-solving favours Meta's bundled token price.
Query Meta's Pricing Analytics API with a pricing_category value of SERVICE to get the count of delivered service messages for any timeframe. The same category also appears in message status webhooks, where billable messages carry billable true, type regular and category service in the pricing object.
Conversation 365 brings WhatsApp, email and your other channels into one inbox, on top of the same unified customer profile as your orders, campaigns and loyalty data. So when replies become billable on 1 October 2026, you can see spend per conversation next to the customer it belongs to — and route by complexity instead of guessing.
All rates, dates and examples in this article are Meta's, as published at the time of writing (). Meta may update rates up to once a quarter, and the rates effective 1 October 2026 should be confirmed against Meta's live rate card for each market you message. Rates vary significantly by market, including across Hong Kong, Singapore, Malaysia, Indonesia, Taiwan, Thailand, the Philippines, Vietnam, India and Brazil. This article is general information about platform pricing, not commercial, tax or legal advice.