What Meta's August 1 WhatsApp AI Pricing Actually Costs an Indian SMB

Short answer: Since 1 August 2026, Meta has billed its WhatsApp Business AI agent by the token, $2 (about ₹189) per million tokens, which works out to roughly 4–5 cents (₹3.80–₹4.70) per message at Meta’s typical usage. Smaller businesses get it through WhatsApp Business Premium tiers instead of raw token billing. Since 1 October 2026, ordinary service messages are billed too, after the first 1,000 per number each month, but an agent reply is charged once, by the token, not twice.

Neither change is a reason to panic. Both are a reason to do arithmetic before your bill does it for you.

What changed on 1 August 2026?

Meta moved its Business AI agent onto usage-based pricing at a global rate of $2 per million tokens, counting both what the agent reads and what it writes, with message delivery included in the same charge (Meta for Developers). A token is roughly three-quarters of a word. Meta puts a typical message at 20,000 to 25,000 tokens, which is where the 4–5 cents per message figure comes from.

Business Today converted the rate at about ₹189 per million tokens (Business Today), so a typical agent message costs roughly ₹3.80 to ₹4.70.

Two things follow from that. Cost scales with conversation volume, and it scales with complexity, a long back-and-forth where the AI re-reads your catalogue costs more than a one-line hours-and-address reply.

Do small businesses pay per token too?

No, not directly. Meta has said smaller businesses will get the agent through select tiers of its WhatsApp Business Premium subscription, while larger businesses are billed on token usage. That is genuinely better for a small shop: a subscription is predictable, and a busy month does not produce a surprise.

The catch is the ceiling. A subscription tier that suits you at 300 conversations a month is a different conversation at 3,000, and the moment you outgrow the tier you are back in usage-based territory.

What does this actually cost my business per month?

Run it yourself, the arithmetic is not hard, and anyone quoting you a single confident number is guessing. Take your monthly customer conversations, multiply by the number of replies the AI sends in a typical thread, and multiply that by the per-message cost. Here it is at the top of Meta’s typical range.

Conversations / month	AI replies each	AI messages	At ~₹4.70/msg (upper bound)
300	5	1,500	~₹7,050
1,000	5	5,000	~₹23,500
3,000	5	15,000	~₹70,500

Treat that table as a ceiling, not a forecast. It assumes every message hits the 25,000-token upper bound, which most will not, short factual replies cost a fraction of that. The honest use of this table is to see the shape: your cost is a straight line through conversation volume, with no step-down for scale.

What changed on 1 October, and does it stack with the agent charge?

Since 1 October 2026, service messages, the replies you send inside a customer-initiated conversation, are billed per message after the first 1,000 per phone number each month, and utility templates sent inside that window are billed too (YCloud). It does not stack with the agent: a reply from Meta’s Business Agent is charged by the token only and is not also billed as a service message (Meta for Developers).

In India the service rate is ₹0.115 per message, the same as utility and authentication (Storyboard18), and unlike those two it gets no volume discount (Meta for Developers). For a business doing support on WhatsApp with its own team or its own bot, that is the line that now moves with volume. We cover it in detail in WhatsApp service messages are no longer free.

Does this make WhatsApp AI a bad deal for SMBs?

No. It makes it a priced deal instead of a free one, which is a different thing. The comparison that matters is not the monthly bill versus zero, it is the monthly bill versus what those same conversations cost you in salaried hours today, and versus the leads that die overnight because nobody replied until morning.

What changes is that sloppiness now has a line item. An agent that rambles through six paragraphs to say “we open at 10” costs real money at 5,000 messages a month. Under the old free tier nobody cared. Now, concision is a budget decision.

How do I keep the bill down without giving up the agent?

Four things move the number, in rough order of impact. None of them require you to be technical.

• Answer short. Token cost tracks length. An agent tuned to answer in two sentences instead of six costs materially less at the same volume.
• Do not send the whole catalogue every turn. Cost includes what the AI reads. Retrieving the three relevant products beats re-reading four hundred.
• Route the trivial stuff around the AI. “What are your hours?” does not need a language model. A rule handles it for nothing.
• Know your own numbers. If you cannot say how many conversations you handled last month, you cannot tell whether a subscription tier or usage billing is cheaper for you.

Should I use Meta's native agent or a custom one?

That depends on whether your customers’ questions need to touch your own systems, and we have written the honest version of that comparison separately, see Meta’s WhatsApp Business AI Agent: use it or build custom?. The short version: the native agent is a good fit for high volumes of simple, WhatsApp-only queries, and the wrong fit the moment it needs to read your inventory or write to your CRM.

On pricing specifically, the useful thing to understand is that a custom agent does not escape token costs, it moves them somewhere you can see and control. You choose the model, you choose how much context each reply reads, and you can use a cheaper model for the easy 80% of messages. That is not automatically cheaper. It is legible, which over a year tends to become the same thing. We worked the numbers side by side in Meta Business Agent vs your own AI: cost per conversation.

What should I do now?

Three things, and none of them take a week. Count last month’s WhatsApp conversations and replies, so the pricing lands on a number instead of a feeling. Separate the questions that genuinely need an AI from the ones a rule answers for free. Then check your October bill against your volume and decide whether a Premium tier, usage billing, or your own agent fits best. And confirm a payment method is on your WhatsApp Business Account: since 1 October, accounts without one have had service messages stopped (here is the fix).

If you want to see what a tuned agent actually sounds like before committing to anything, try our live WhatsApp AI demo, it answers the way an agent should: short, correct, and out of the way. Or book a free 15-minute call and we will do the arithmetic with you.