New WhatsApp API Pricing Update: Non-Template (Service) Messages Will Now Be Charged From October 2026
If you run customer support, sales, or notifications on WhatsApp, Meta just announced a big change to how it charges for messages. Starting October 1, 2026, WhatsApp will begin charging for service messages — the free-form replies your team or chatbot sends to customers. These have been free since November 2024, so this is a meaningful shift for anyone using WhatsApp for support conversations.
Here's what's changing, explained simply.
Message categories for non-template messages
Today, when your business replies to a customer within the 24-hour customer service window, that reply is treated as one single category: a service message. It doesn't matter who or what sends it — a support agent or a third-party AI chatbot — it's all lumped together and it's free.
That's about to change. Meta is splitting non-template replies into two categories:
- Service messages — this is the existing category. Any reply that is not powered by Meta's own AI agent falls here. That includes replies from a human support agent or a third-party AI/chatbot solution.
- Meta Business Agent messages — a brand-new category for replies powered by Meta's own AI agent platform, which officially launches July 1, 2026.
So going forward, the question "who (or what) sent this reply?" actually determines what you get billed and how.
Pricing updates
Here's the full rollout timeline:
| Date | What happens |
|---|---|
| July 1, 2026 | Meta Business Agent Platform launches. Businesses can start integrating. |
| August 1, 2026 | Meta starts charging for Meta Business Agent messages — billed per token (like an AI usage fee), not per message. |
| October 1, 2026 | Meta starts charging for service messages — billed per message, the same way template messages are billed today. |
| October 1, 2026 | Meta also starts charging for utility messages sent inside the 24-hour window (previously free since July 2025). |
In short: by October 2026, almost every reply you send a customer — human-written, AI-written, or automated — will carry some kind of charge.
Service messages
This is the part most businesses will feel directly.
- What counts as a service message: any free-form reply sent by a person or a third-party AI tool (not Meta's own AI agent) within the open 24-hour customer service window.
- How it's charged: per message, at the same rate as utility and authentication template messages in that country. There are no volume discounts for service messages (unlike utility/authentication templates, which do get volume tiers).
- When it kicks in: October 1, 2026. Exact rates for each market will be published by Meta before September 1, 2026.
- What's still free: service messages remain free inside the 72-hour "free entry point" window (i.e., conversations that start from a Click-to-WhatsApp ad or Facebook/Instagram CTA button).
If you want a broader primer on how WhatsApp pricing categories (marketing, utility, authentication) work today, Chakra's WhatsApp API pricing guide is a solid reference.
Pricing example
Meta shared a simple example to show how charges stack up across a single customer interaction, from a marketing template to a live agent handoff:
A business sends a marketing template, then uses Meta's AI agent to answer two quick product questions, then hands off to a human agent when the customer has a payment issue, and finally sends an order confirmation — all within an hour.
| Message | Type | Category | Charged as of Aug 1, 2026 | Charged as of Oct 1, 2026 |
|---|---|---|---|---|
| Cart abandonment reminder | Template | Marketing | ✅ Charged | ✅ Charged |
| AI answers "Do you have this in Small?" | Non-template | Meta Business Agent | ✅ Charged | ✅ Charged |
| AI answers "Do you have this in white?" | Non-template | Meta Business Agent | ✅ Charged | ✅ Charged |
| Human agent handles payment issue | Non-template | Service | ❌ Free | ✅ Newly charged |
| Order confirmation | Template | Utility | ❌ Free (in-window) | ✅ Newly charged |
Total charges: 3 today (as of August 2026) → 5 charges after October 1, 2026, for the exact same conversation.
Cost comparison: service and Meta Business Agent messages
Meta also published a cost comparison for 10,000 AI-powered replies sent to customers in Brazil, comparing a service message with a third-party AI tool against Meta's own Meta Business Agent:
| Service (simple queries) | Service (complex queries) | Meta Business Agent (complex queries) | |
|---|---|---|---|
| AI cost per message | ~2 cents | ~9 cents | ~4–5 cents |
| Message delivery cost | 0.68 cents | 0.68 cents | Included in the per-token charge |
| Total per message | ~2–3 cents | ~9–10 cents | ~4–5 cents |
| Total for 10,000 messages | ~$268 | ~$968 | ~$400–$500 |
A few things worth noting:
- If your existing AI/chatbot vendor is efficient and handles simple queries, sticking with a service message + third-party AI setup can be cheaper than switching to Meta's own agent.
- For complex, multi-turn queries, Meta's own agent can work out cheaper than an expensive third-party AI provider — but this varies a lot by vendor.
- Meta Business Agent bills in one combined charge (AI usage + delivery); service messages combine a separate AI provider bill (if you use one) plus Meta's own per-message delivery charge.
Since your total cost will now depend heavily on your AI/chatbot vendor's rates plus Meta's per-message delivery charge, it's worth revisiting how your current provider prices things — Chakra's detailed WhatsApp API pricing guide breaks down how vendor markups, volume tiers, and country-based rates typically stack up, which is useful context for estimating your real cost after October.
How to reduce costs
Since non-template messages will no longer be free after October 2026, it's worth building a few habits and vendor checks into your workflow now, before the charges kick in.
- Pick a BSP with low or zero markup on message charges. Some providers add a markup on top of Meta's standard rates for both template and non-template messages. Others — like Chakra Chat and a few similar providers — pass on message charges at Meta's standard rates with little to no markup. Since you'll now be paying for every service message, even a small per-message markup adds up fast at scale, so this is worth checking before October.
- Say more in fewer messages. Since service messages are billed per message (not per character), splitting one reply into three separate messages costs three times as much as saying the same thing in one well-written message. Combine context, answer, and next steps into a single reply wherever you can.
- Club multiple customer queries into one message. If a customer asks two or three quick questions back-to-back, resist the urge to reply message-by-message. Batch your response into a single message that addresses everything at once.
- Design your chatbot flows to capture more in fewer steps. A chatbot that takes 8 back-and-forth steps to collect a customer's name, order ID, and issue will cost more than one that captures the same information in 2–3 well-designed steps (e.g., using list messages, buttons, or forms instead of one-question-at-a-time flows).
- Use templates where the conversation is predictable. For repetitive, structured updates (order confirmations, appointment reminders, OTPs), a utility or authentication template is often cheaper and more consistent than a free-form service message, especially once volume discounts are factored in.
- Route simple queries to lower-cost automation, and reserve AI/human agents for complex ones. Based on Meta's own cost comparison, simple queries are cheap to handle with any method, but complex, multi-turn conversations are where costs spike. Use lightweight rule-based bots or FAQ automation for simple queries, and only escalate to a human agent or a more expensive AI setup when genuinely needed.
- Track cost-per-conversation, not just cost-per-message. Once multiple message types can appear in a single customer journey (template + AI agent + service message), monitoring total conversation cost will give a clearer picture of where money is actually going than looking at any one message type in isolation.
Important notes
- Every non-template reply is charged from October 1, 2026 — no exceptions. If it's a Meta Business Agent message, it's charged per token. If it's anything else (human agent, third-party bot), it's charged as a service message per message.
- You're only charged once per message. A message is never double-billed as both a service message and a Meta Business Agent message, and non-template messages with promotional content don't also get hit with a marketing template charge.
- Meta Business Agent costs scale with complexity. A quick "what time do you open?" answer costs far less (roughly 16–20 cents across a 4-message exchange) than a complex, multi-turn troubleshooting conversation (roughly 40–50 cents across 10 messages).
- The 72-hour free entry point window still protects message delivery. Conversations starting from a Click-to-WhatsApp ad or Facebook/Instagram CTA remain free for message delivery — though Meta Business Agent token charges still apply even inside this window, starting August 1, 2026.
- Rates for service messages aren't public yet. Meta has committed to publishing exact per-market rates by September 1, 2026 — a month before charging starts.
Bottom line: if your WhatsApp strategy leans on free-form support replies to keep costs down, October 2026 changes the math. It's worth auditing your support volume now, understanding which of your replies will fall into the "service" bucket, and deciding whether Meta's own AI agent or your existing chatbot/BSP setup is the more cost-effective path going forward.