guide
WhatsApp AI Pricing: Meta Charges $2 per Million Tokens
Meta started billing its own WhatsApp agent by the token on 1 August. Almost every explainer of it is wrong.
On 1 August 2026 Meta began charging for messages its own AI agent sends on WhatsApp, and it charges by the token. That makes it the only platform in this directory whose AI bill scales with how much the model reads and writes rather than with resolutions, conversations or seats.
Every explainer we found for this was written by a WhatsApp reseller, and the ones we read describe the change incorrectly. So here it is from Meta's pages.
The rate
Meta's documentation gives one global rate: $2.00 USD per million tokens. It states that one message typically consumes 20,000 to 25,000 tokens, which works out at roughly four to five cents a message, and that the single charge covers both the model's processing and the message delivery.
Four to five cents sounds cheap, and it is the wrong unit to plan with. Meta publishes its own worked examples and they are per interaction rather than per message.
| Interaction | Messages | Tokens | Meta's estimated cost |
|---|---|---|---|
| Simple inquiry — "At what time do you open?" | 4 | ~80,000 | ~16–20 cents |
| Complex — "walk me through step 3" | 10 | ~250,000 | ~40–50 cents |
A support conversation that costs forty to fifty cents is in the same range as the per-resolution platforms, not an order of magnitude below them. And unlike a per-resolution rate, it rises with how talkative the agent is.
Nothing switched, and this is where the explainers go wrong
Six reseller blogs describe this as WhatsApp switching to token-based pricing. Nothing switched. Meta added a fifth message category.
Marketing, utility and authentication templates are still billed per delivered message on the same country rate cards. Service messages — the non-template replies inside a customer service window — are still free. Token pricing applies to exactly one thing: messages generated by Meta's own agent.
A team told that WhatsApp moved to token pricing would model their bill wrong in both directions.
The same explainers also conflate this with Meta's AI Providers policy, which is a different policy with a different subject. That one governs third-party AI companies operating on WhatsApp, bills per message rather than per token, took effect in February 2026, and its geographic scope shrank in May to a single market. It has nothing to do with Meta Business Agent.
Who pays for what, today
| Who sends the reply | Charged by Meta today | What changes |
|---|---|---|
| A human agent | No | Charged per message from 1 October 2026 |
| A third-party AI tool | No — that vendor bills you | Charged per message from 1 October 2026 |
| Meta Business Agent | Yes, per token, since 1 August 2026 | — |
The October date is the part the explainers miss and it matters more than the token rate for most teams. Service messages have not been charged since November 2024, and utility templates sent in reply to a customer have been free since July 2025. Both start being charged on 1 October 2026, at the utility rate, and Meta's documentation states there will be no volume tiers for service messages.
If your support runs on WhatsApp with human agents, your bill goes from zero to per-message in October regardless of whether you ever touch an AI agent.
The clause that should decide your rollout
Meta's API exposes a budget object for the agent. It measures spend in tokens, over a rolling window of one, seven, fourteen or thirty days, with a maximum you set.
Its documentation also states what happens if you leave it alone: an empty budgets array means no cap is set and usage is unlimited.
So the default state of a Meta Business Agent is uncapped spend on a meter that rises with conversation length. That is one line in an API reference and it is the single most useful fact on this topic. No reseller explainer we read mentions it.
There is a second gate worth knowing: Meta states that Business Agent messages will not be delivered if a credit line is not attached to the account, and that billing is by monthly invoice, with cards targeted for September.
What it can actually do
Meta describes the agent as answering from your business knowledge, taking actions through your own systems, and handing off to your app when needed. The mechanism is documented rather than asserted: you register a connector with a base URL and an authentication type, then define each operation as a tool with an explicit request schema the agent uses to build valid calls.
Which places it firmly on the buyer-built side of the split we track. Meta ships no connectors — no Shopify, no Zendesk, nothing. You expose your own API, hand Meta the credentials, and declare every operation. The reasoning is Meta's; the reach is entirely yours.
One eligibility constraint matters for regulated buyers: Meta's documentation excludes Health, Finance, Government and several other verticals from the agent product outright, and the WhatsApp Business Terms disclaim suitability for entities with heightened confidentiality requirements such as healthcare.
One thing to read carefully
Meta's main pricing page carries a notice about these changes, and on 19 August it still read that pricing updates for Meta Business Agent, service and utility messages will launch on 1 August 2026 and 1 October 2026 — future tense, nineteen days after the first of those dates had passed.
The free-message statements in the table above it are unqualified, so a reader has to click through to discover that Meta's own agent is charged. Worth knowing if you are pricing this from the page rather than the documentation underneath it.
Read from Meta's own developer documentation on 19 August 2026 and matched against raw page source. Meta's pages require full browser headers to fetch; a short response body from them is a failed request rather than an empty page, which is how several third-party accounts of this appear to have gone wrong.
Frequently Asked
How much does Meta charge for WhatsApp AI?
One global rate of $2.00 per million tokens for messages sent by Meta Business Agent, effective 1 August 2026. Meta says a message typically consumes 20,000 to 25,000 tokens, about four to five cents. Its own worked examples are more useful: roughly 16–20 cents for a simple four-message inquiry and 40–50 cents for a ten-message complex one.
Did WhatsApp switch to token-based pricing?
No, and this is the most common error in explainers of it. Meta added a fifth message category. Marketing, utility and authentication templates are still billed per delivered message on country rate cards. Token pricing applies only to messages generated by Meta's own agent — not to human replies and not to third-party AI tools.
Are human agent replies on WhatsApp still free?
Until 1 October 2026. Service messages — non-template replies inside the customer service window — have been free since November 2024, and utility templates sent in reply to a customer since July 2025. Both start being charged per message on 1 October at the utility rate, with no volume tiers for service messages.
Is there a spending cap on Meta Business Agent?
Only if you set one. Meta's API has a budget object measuring token spend over a rolling one, seven, fourteen or thirty day window — and its documentation states that an empty budgets array means no cap is set and usage is unlimited. The default is uncapped spend on a meter that grows with conversation length. Messages also will not deliver at all unless a credit line is attached.
Can Meta Business Agent take actions in my systems?
Yes, through connectors and tools you build. You register a base URL with an authentication type, then define each operation with an explicit request schema. Meta ships no pre-built connectors — no Shopify, no help desk, nothing — so the reasoning is Meta's and the integration work is entirely yours. Health, Finance and Government verticals are excluded from the product.
Tools Mentioned
Full reviews, pricing tiers and where each one breaks.
WhatsApp Business Platform
Meta's own channel and, since July 2026, its own agent. The only platform here billing AI per token rather than per resolution, and the only one where the vendor is also the channel.
ManyChat
Priced per active contact rather than per seat, with a real free tier at 25 contacts. The AI answers and captures; external writes run through an External Request block you author, not through the AI.
Chatfuel
Social-commerce chatbot whose AI books, reschedules and cancels appointments end to end — into its own calendar. Publishes two mutually incompatible price lists on two of its own properties.
Tidio Lyro
One of the few genuinely cheap autonomous agents, bundled with live chat and a basic help desk.
You Can Also Look Into
Voice AI Pricing in 2026: Two of Nine Publish a Rate
Voice is the only category in this directory where every platform can act on a system of record, and it is near the bottom on whether anyone will quote you. Here is what the two that publish actually charge, the meters that sit underneath a per-minute rate, and why a quoted price per minute is not comparable across vendors.
Does Your AI Support Vendor Bill the Attempt or the Outcome?
HubSpot moved from charging per conversation to charging per resolved conversation, and Kustomer bills when the AI merely generated a response, sent or not. Same year, same category, opposite directions — and the difference decides whether a bad agent costs you more or less than a good one.
Only 38 of 83 AI Customer Service Vendors Publish Their Pricing
The full count across 83 platforms, the categories where transparency collapses entirely, and what a hidden price actually costs you.
SOURCES
WRITTEN BY AR · UPDATED 2026-08-20
I read the fine print. Vendor pricing pages, billing definitions, terms, funding filings and acquisition notices — then I do the arithmetic nobody publishes: what a platform actually costs at your volume, what its headline metric is really counting, and who owns it now. I do not run benchmarks, and no page here pretends otherwise.