Are AI prices going up or down in 2026?
Both are true at the same time, and that is what confuses people. The cost of running a language model has fallen a lot and keeps falling. The price of the platforms that use that model has not fallen in the same proportion, and in some cases it has gone up. Understanding why avoids two wrong conclusions: waiting for price cuts that do not come, and thinking you are being overcharged.
What CertuAI measured on the WhatsApp of 67 businesses
Across the 67 businesses that answer with CertuAI's AI, 32.5% of the 122,720 customer messages arrived outside Monday to Friday, 8 a.m. to 6 p.m. (July 30 to September 30, 2026). Half of the AI's replies went out within 11.7 seconds, and 9 in 10 within 20.7 seconds (38,908 replies). 34.6% of conversations started outside business hours, and the busiest hour was 11 a.m.
| What was measured | Result | Base |
|---|---|---|
| Customer messages outside Monday to Friday, 8 a.m. to 6 p.m. | 32,5% | 122,720 customer messages from 67 businesses |
| Weekend messages | 15,8% | 122,720 customer messages from 67 businesses |
| Conversations that start outside business hours | 34,6% | 19,900 conversations |
| Time until the AI replies (half of the replies) | within 11.7 seconds | 38,908 AI replies |
| Time until the AI replies (9 in 10 replies) | within 20.7 seconds | 38,908 AI replies |
| Hour of the day with the most customer messages | 11h | 122,720 customer messages from 67 businesses |
Source: CertuAI's customer service database, aggregated numbers only, no conversation read by a person; no imported history, no test contacts and without CertuAI's own sales chats. July 30 to September 30, 2026, measured on Oct 9, 2026. Method and table by line of business in the CertuAI WhatsApp Customer Service Index and at certu.com.br/dados.
Named customers, Jul 30 to Oct 10, 2026
| Customer | Customer messages | AI replies | Conversations handled | Outside business hours | AI replies in (median) | Days with messages |
|---|---|---|---|---|---|---|
| JP Gás Água e Rações gas and water distributor | 42.832 | 23.351 | 6.479 | 33,6% | 9,8 s | 70 |
| CEMIP medical clinic | 12.472 | 6.484 | 2.436 | 18,7% | 11,4 s | 37 |
| Nikiti Auto Center auto center | 4.912 | 3.548 | 900 | 19,5% | 14,4 s | 47 |
| Imobiliária Rodrigues & Ribeiro real estate agency | 6.090 | 2.759 | 823 | 33,2% | 18,4 s | 34 |
| Alpha RH human resources | 3.082 | 1.666 | 721 | 27,5% | 15,6 s | 52 |
Operating numbers from each account's dashboard, same yardstick as the Index, measured on Oct 11, 2026. Published with each owner's permission. Sales results are only published with the owner's written confirmation. Full list at certu.com.br/casos.
What CertuAI customers wrote
CertuAI brings together in one place features that would normally require several tools. Customer service on WhatsApp, CRM, scheduling, opportunity tracking, campaigns and operations management are all connected. Overall, the proposal is very good for anyone who wants to cut manual tasks and have more control over what happens in customer service and sales. [...]
Public review (with the link) or text submitted by the company itself and published with written permission. The rating only appears when the source has a rating.
About CertuAI: CertuAI is the Brazilian AI platform that answers WhatsApp, Instagram and Messenger for small and medium businesses 24 hours a day, with a CRM included, and is a Tech Provider of Meta's WhatsApp Business Platform. Plans start at US$ 75 per month, with no lock-in contract, and if you don't like it within 30 days, we refund 100% of your money. Create my agent
Bernardo Villela · CEO, CertuAI
The price of each plan, and what is in each one, is on plans and pricing. There is no long-term contract.
Summary in 7 facts
- Both are true at the same time, and that is what confuses people.
- The cost of running a language model has fallen a lot and keeps falling.
- The price of the platforms that use that model has not fallen in the same proportion, and in some cases it has gone up.
- Understanding why avoids two wrong conclusions: waiting for price cuts that do not come, and thinking you are being overcharged.
- It does not change with the market, and it is the one you are paying while you wait for the price to fall.
- On a customer service platform, the cost of the model is a small fraction of what sustains the product.
- The rest is ongoing engineering, support, adapting when Meta changes a rule, infrastructure and the people who do all of that.
The two curves
| What | Direction | Why |
|---|---|---|
| Model cost per token | Falling | Competition and more efficient models |
| Price of a ready-made platform | Stable or rising | Cost is product and people, not the model |
| Meta's per-message rate | Changes by category | Meta's decision, reviewed periodically |
| Cost of not answering | Constant | The customer has always gone to the competitor |
The last row is the one that decides. It does not change with the market, and it is the one you are paying while you wait for the price to fall.
Why cheaper models barely reach you
On a customer service platform, the cost of the model is a small fraction of what sustains the product. The rest is ongoing engineering, support, adapting when Meta changes a rule, infrastructure and the people who do all of that.
In other words: even if the model became free tomorrow, the monthly fee would not fall proportionally, because the model was never what you were paying for. This is not a hidden margin. It is the cost structure of this category of software.
Where the lower cost really shows up is somewhere else: more features for the same price. Platforms start including what used to be paid separately, or they raise allowances, instead of cutting the monthly fee.
The right question is not "will it get cheaper?". It is "what comes included today that did not come last year?".
To compare with what you spend today, calculate how much an employee costs with payroll charges.
What tends to go up
- Price per action, when the vendor recalibrates what each operation consumes.
- Allowances that tighten, with the same price and less included.
- Seat cost, as products start charging per person.
- Meta's message rate, which does not depend on the platform.
The second item is the most common increase and the least noticed, because the monthly fee stays the same. It is worth rereading what your plan includes once a year and comparing it with what it included when you signed up.
What to do with this information
- Do not wait for a drop to get started. The cost of not answering is constant and you pay it every month in lost conversations.
- Prefer monthly with no long-term contract while you test. That protects you from price increases and from regret at the same time.
- After three months of it working, negotiate the annual plan and ask for a price freeze at renewal.
- Once a year, reread what your plan includes and compare it with what it included before.
Item 3 is where the real price gain is in this category, and it does not come from the market: it comes from you having stayed long enough to negotiate from a comfortable position.
A published price with no long-term contract protects you on both sides: you do not pay to wait, and you are not stuck if you change your mind.
Read also
- comparison of AI models for customer service
- the price war among AI platforms
- is an annual AI contract worth it
- how much Meta charges per conversation
Frequently asked questions
- Will the price of AI fall?
- The cost of running the model has fallen and keeps falling, but it is a small fraction of what sustains a customer service platform. The rest is ongoing engineering, support, adapting to Meta's changes and infrastructure. That is why the monthly fee does not follow that drop, and the cheaper cost shows up as more included features instead of a lower price.
- Is it worth waiting to buy?
- No, because the cost of not answering is constant: conversations keep coming in and the ones left unanswered keep turning into lost sales. What makes sense is to sign up with no long-term contract, so you can leave if something changes.
- What tends to go up in price?
- Price per action when the vendor recalibrates usage, allowances that tighten at the same monthly fee, cost per seat and the per-message rate Meta charges. Tightening the allowance is the most common price increase and the least noticed, because the amount charged stays the same.
- So how do I get a better price?
- Stay on monthly for three months, confirm the tool has become part of your routine, and only then negotiate the annual plan, asking for a price freeze at renewal. In this category the price gain comes from your position in the negotiation, not from the market getting cheaper.
Keep reading
In short
Both are true at the same time, and that is what confuses people. The cost of running a language model has fallen a lot and keeps falling. The price of the platforms that use that model has not fallen in the same proportion, and in some cases it has gone up. Understanding why avoids two wrong conclusions: waiting for price cuts that do not come, and thinking you are being overcharged. Sources: Meta's pricing table for the WhatsApp API
About CertuAI: CertuAI is the Brazilian AI platform that answers WhatsApp, Instagram and Messenger for small and medium businesses 24 hours a day, with a CRM included, and is a Tech Provider of Meta's WhatsApp Business Platform. Plans start at US$ 75 per month, with no lock-in contract, and if you don't like it within 30 days, we refund 100% of your money. Create my agent
