WhatsApp AI agents: Meta starts billing by the token — what it costs your SME

Since August 1, 2026, Meta has switched billing for AI-generated replies on the WhatsApp Business Platform — what it calls Business AI conversations — to a token-based model: $2.00 per million tokens, bundling what used to be billed separately as AI processing and message delivery. A free build-and-test window ran from July 1 to July 31, 2026; it is now over. A typical AI reply consumes roughly 20,000 to 25,000 tokens, which works out to about 4 to 5 cents (US) per message — a figure confirmed across independent trade coverage of Meta’s own developer documentation. The same token pricing also applies to AI agent conversations on Messenger and Instagram, but WhatsApp is where most SMEs running a sales or support agent will feel it first.

What changes, and when

• →Since August 1, 2026: every AI-generated reply sent through a WhatsApp Business AI agent is billed at $2 per million tokens — roughly 4 to 5 cents per message, replacing the previous free test period.
• →By September 1, 2026: Meta has committed to publish the exact regional rates for the next wave of charges — figures not yet public region by region as of this writing.
• →From October 1, 2026: service messages (free-form replies sent by a business or a human agent inside the 24-hour customer window) and utility message templates sent in that same window become chargeable again — they had been free since November 2024 and July 2025 respectively.
• →Marketing and authentication template pricing, unaffected by this change, continues to follow the per-message rates already in place market by market.

Why this is more than a line-item change

An SME that plugged a WhatsApp AI agent into its sales or support flow now depends on a pricing decision it does not control, made unilaterally by the platform it built on. That dependency is exactly what a 2021 study by Donato Cutolo and Martin Kenney, Platform-Dependent Entrepreneurs: Power Asymmetries, Risks, and Strategies in the Platform Economy, published in Academy of Management Perspectives, describes: businesses built on top of a digital platform face a structural power asymmetry, because the platform can unilaterally change the rules of engagement — including pricing — while the business bears the full cost of adapting. Meta’s August-to-October sequence is a textbook case: a channel that was cost-free during testing, then billed per token, then extended to service messages that had themselves been free for nearly two years.

What it means for your SME

If your business runs a WhatsApp commercial or support agent, two separate cost lines are moving. First, every AI-generated message your agent sends today already carries a token cost — a chatty agent that pads its answers, repeats context, or fails to close a conversation quickly is now directly more expensive, not just less efficient. Second, from October 1, any human reply your team sends inside the 24-hour window — a follow-up after the agent hands off to a person, a clarification typed by an advisor — resumes costing money after nearly two free years, on top of the AI-agent charge. Budgeting for a WhatsApp channel in 2026 now means tracking two variables at once: message volume and token consumption per message, not just conversation count.

Concrete steps before October 1

• →Estimate your current WhatsApp AI agent volume (messages per month) and multiply by roughly 4 to 5 cents to get a first order-of-magnitude monthly cost at the new token rate.
• →Ask your WhatsApp Business Solution Provider whether it passes Meta’s token rate through as-is or adds its own margin on top — the gap between the two can be significant at volume.
• →Review your agent’s prompts and conversation flows to cut unnecessary verbosity: shorter, more targeted replies cost fewer tokens and read better besides. A commercial WhatsApp agent designed to qualify and close efficiently, rather than chat at length, directly limits this new cost line.
• →Watch for Meta’s regional rate announcement, due by September 1, 2026, to refine your budget before the October 1 service-message charges take effect.
• →Recount how many of your team’s WhatsApp replies today fall inside the free-form "service message" category — those are the ones about to start costing money again from October 1.

None of this makes WhatsApp a bad channel for an SME — it remains where a large share of French customers already are. But a channel billed per token needs the same operational discipline as any other paid API: monitored, budgeted, and designed to say what it needs to say without padding — not run on the assumption that yesterday’s free tier is still today’s price.