Meta changed the model: agent conversations are no longer billed per conversation but by tokens consumed. Here is what that costs, with the published figures and the arithmetic done.
Last updated: September 2026
Since August 1, 2026 Meta bills Meta Business Agent conversations by tokens, at USD 2.00 per million. A delivered message consumes between 20,000 and 25,000 tokens, which puts the real cost at roughly USD 0.04 to 0.05 per message, with AI processing and delivery already bundled into the same rate.
Previously AI usage and message delivery were paid separately. Now it is a single line item, which simplifies the calculation but changes how you optimize it.
Sources: WhatsApp Business Platform pricing documentation and partner analyses, September 2026.
The meaningful change is not the price, it is the unit. You pay for what the agent processes, not for opening a conversation.
That is the date token billing replaced the previous model for agent conversations.
There is a second platform pricing change worth budgeting for, even though it is not exclusive to the agent.
The math is simple: messages the agent will answer per month, times the average cost per message. The hard part is estimating how many the agent resolves and how many end up with a person.
With token billing, knowledge quality stops being only an experience concern: it becomes a direct cost lever.
Based on USD 2.00 per million tokens and 20,000 to 25,000 tokens per delivered message.
Estimated monthly cost
$200 – $250
Indicative estimate, not a quote: real consumption depends on message length.
Open the full calculatorCalculated on 20,000 to 25,000 tokens per delivered message and USD 2.00 per million tokens.
Indicative estimate, not a quote. Real consumption depends on message length and on the context the agent processes.
What the agent is, how it is activated and what it takes to implement it.
Activation was free at launch, but since August 1, 2026 agent conversations are billed by tokens. Meta also announced WhatsApp Business Premium subscriptions for smaller businesses and token-based consumption for larger ones.
No. Meta merged both into the token rate: the price per million covers model processing and message delivery as a single billing line.
Between 20,000 and 25,000 per delivered message, according to published figures. It depends on the length of the reply and on how much context the agent needs to produce it, including the knowledge it consults.
The agent's token rate is global. What does vary by country and category is the price of conversations the agent does not handle, which follow the platform's usual pricing scheme.
By restricting the agent's audience to an allowlist of numbers. Volume stays bounded and measurable, so you can work out cost per resolved conversation before opening it to your whole base.
It depends on volume. With Meta's agent you do not separately pay an AI provider or maintain the model, but you pay per delivered message. With a custom agent the AI cost is yours and so is the build, though marginal cost per message can be lower at large scale.
With your real conversation volume and the share the agent can resolve, the calculation stops being an estimate.
What would it cost you?
We run the numbers with you