WhatsApp API Pricing 2026: Per-Message vs. Flat Rate

Ask three providers what a WhatsApp API costs and you get three different units: per message, per conversation, per number, per seat. None of them are wrong. They are just answering different questions, and the one that matters — what will my bill be at my volume? — rarely gets a straight answer.

This is our attempt at one. Every number below carries a date, because pricing in this market moves, and we run one of the flat-rate services, so read the last section knowing that.

The two models, in one paragraph each

Meta's Business Platform bills per message. Since 1 July 2025 the official Cloud API charges for each template message it delivers, priced by category and by the recipient's country. Regular replies inside an open customer-service window are free. Your bill scales with how much you send.

Third-party providers bill per connected number. Most of the services built on the web client — ours included — charge a flat monthly price for each phone number you connect, and nothing per message. Your bill scales with how many numbers you run, not with how much they send.

Which one is cheaper depends entirely on where you sit on two axes: message volume and number of senders.

What Meta actually charges

The official pricing page is precise about the model and silent about the numbers — it links out to per-currency rate cards. As of the rate cards effective 1 July 2026, here is what the model looks like:

A customer-service window opens for 24 hours whenever a user messages you. A free entry point window opens for 72 hours when someone reaches you through a Click-to-WhatsApp ad or a page call-to-action, and everything you send inside it is free.

Meta's own page gives no per-message figures, so the numbers below are from published 2026 rate-card summaries by Unipile and respond.io. Treat them as the order of magnitude, not an invoice:

Two things the table hides. First, you rarely get Meta's rate as your rate: access goes through a Business Solution Provider, and their markup is either a few thousandths per message or a flat monthly licence. Second, marketing is the category that costs, and it is also the category most people mean when they say "send messages to my customers".

What that means at volume

Take a business in Brazil sending 10,000 marketing messages a month — a modest promotional list.

• Meta's rate alone: 10,000 × $0.0625 ≈ $625 per month
• Plus a provider licence or per-message markup on top

The same 10,000 messages in the United States land around $250, in India around $94. Per-message pricing is a bet on your volume staying small, and a bet on your customers being in the cheap countries.

What flat-rate providers charge

We surveyed the unofficial-API market on 24 August 2026 — 32 providers, of which 16 were actively selling. Prices are per connected number, per month, list price in USD:

The market has settled into three bands: a $5–15 developer tier (API only, minimal UI, hard session limits at the cheap end), a $24–39 standard tier with documentation, SDKs and a sandbox, and a $40–120 tier where you are really buying a team inbox and the billing shifts from per number to per seat.

Fourteen of the sixteen live providers advertise "no per-message fees". Per-message billing has all but disappeared from this side of the market; the one provider we found still using it was roughly fifteen times more expensive at volume than the flat-rate alternatives.

One thing almost nobody publishes: a volume discount. Whapi is the exception, with a tier that drops from $35 to $25 per number above 20. Everyone else, us included, publishes no tier at all — several providers simply refer you to sales past ten numbers. That matters for the next section.

The break-even nobody wants to calculate

Both models above have a ceiling: self-hosting. The open-source web-client libraries are free, and a small VPS — two vCPUs, 4 GB, anywhere from $5 to $25 a month depending on the host — comfortably holds fifty sessions.

So why does anyone pay $10 a number? Because the server is not the cost. The cost is the person who keeps it running.

Here is the calculation we use internally, and we would rather you see it than not. Assumptions: ten connected numbers, a developer at $120 an hour, a $10 VPS, a one-off six hours of setup spread over two years, and three hours a month of operational work — a realistic figure for normal operation once the protocol breaks, a session needs re-linking, or a library release changes behaviour.

Below roughly forty numbers, paying a provider is cheaper than paying a developer to babysit a server. Above it, self-hosting wins on cash — and at that scale you usually have the engineer anyway.

The honest caveats:

• The three hours are the whole argument. If you already run a Docker fleet and the on-call rota exists, your marginal cost is lower and the break-even moves down — somewhere between ten and twenty numbers for a team with that muscle.
• The libraries are not stable. The most-used one spent most of 2026 in release candidates, with a security fix along the way and at least one release that exists only because the protocol changed underneath it. That is where the three hours come from.
• A provider removes infrastructure and session risk, not compliance risk. Nobody — not us, not anyone on the list above — can promise a number will never be restricted. Anyone who does is selling you something.

How to pick

Run the two questions against your own situation:

How many messages, and to whom? Under a few thousand marketing messages a month to customers in cheap countries, Meta's per-message model can be the lowest bill, and it comes with the official platform's stability. Past that — or with a Brazilian or European audience — the per-message line grows faster than any flat rate.

How many numbers? One to three numbers: a flat-rate provider almost always wins; the annual saving from self-hosting would not cover a single afternoon of setup. Four to ten: this is where the price-sensitive markets start doing the maths, and where a published volume tier would matter. Twenty and up: self-hosting is cheaper on paper, and the question becomes whether you want the operational job.

If your traffic is inbound — customers messaging you, and you replying — Meta's model is close to free, because service messages inside the window cost nothing. If your traffic is outbound — you initiating, at volume — that is exactly where per-message billing gets expensive and per-number billing does not care.

Where we stand, stated plainly

WaAPI charges $10 per connected number per month, with unlimited messages, no per-message fees, no setup fee and no contract. Checkout shows it in your local currency, and there is currently no volume discount, which puts us in the developer tier above — and, by our own analysis, is where we lose customers to self-hosting once they pass ten numbers.

We are not the cheapest on the list. WasenderAPI is. We are cheaper than the standard tier by a margin we have chosen on purpose: the product is the API, the SDKs and the MCP server, not an inbox, and we would rather be priced like an API.

What you get for the difference between us and a server of your own is the three hours a month, every month, for as long as the protocol keeps changing. Whether that is worth $10 a number is a calculation only you can run — but now you have the numbers to run it.

Meta's rates and the provider prices above are as published on 2 September 2026. We will keep the date on this article honest; if a number here looks stale, the pricing page is always current.