AI pricing
How much does an AI chatbot cost in 2026?
An AI chatbot for a business costs a one-off build plus a monthly fee for the AI model. Our custom chatbots start at EUR 5,000 excluding VAT, and for an e-shop with 3,000 conversations a month a low-cost model comes to roughly 200 to 600 CZK a month.
The price of an AI chatbot for a company is made of three parts that people often lump together: a one-off build, a monthly bill for the AI model, and ongoing operation. If you do not separate them, vendor quotes become impossible to compare. One agency offers a "chatbot for 990 CZK a month", another quotes a 150,000 CZK project, and a third charges for every resolved conversation. All three can be fair, they simply charge for different things. This article breaks down what you actually pay for, what the ChatGPT, Claude and Gemini models cost today converted to Czech crowns, and how to work out whether a chatbot pays off for your business.
Interest in AI among Czech companies is growing fast. According to the Czech Statistical Office (ČSÚ), 17% of Czech enterprises used artificial intelligence in 2025, up from 11% a year earlier, and among large firms the share reached 54%. A customer chatbot is often the first project, because the result is visible immediately and easy to measure.
What goes into the price of an AI chatbot
Before comparing offers, split the price into these items:
- Build and setup. Conversation design, preparing source material (terms, price list, FAQ, manuals), connecting a knowledge base and testing on real customer questions. This is a one-off cost.
- Integrations. A chatbot that only answers from documents is cheaper than one that checks order status in Shoptet, opens a return request or reads data from the Pohoda accounting system. Every system connection adds work.
- AI model usage. The model provider (OpenAI, Anthropic, Google) charges per processed token. A token is a piece of a word, so in practice you pay for the volume of text the model reads and writes.
- Operation and maintenance. Hosting, updating source material, reviewing answers, adjusting to new kinds of questions. With off-the-shelf SaaS tools this part is hidden in the monthly subscription.
- Human work behind the scenes. Someone has to take over the questions the chatbot cannot resolve. A well-built AI chatbot for your website hands complex cases to a person together with a summary, so the agent does not start from zero.
The biggest differences between offers are usually in items 1 and 2. For a typical business chatbot the model itself is surprisingly cheap, as the calculation below shows.
What the ChatGPT, Claude and Gemini models cost
Model prices are quoted in US dollars per million tokens, separately for input (what the model reads) and output (what it writes). For conversion we use the Czech National Bank rate of 22 October 2026: 21.651 CZK per dollar. Prices come from the OpenAI, Anthropic and Google Gemini API pricing pages on the same day.
| Model | Input per 1M tokens | Output per 1M tokens | Typical use |
|---|---|---|---|
| Claude Haiku 5.5 (Anthropic) | $0.10 | $0.50 | fast answers from FAQ and documents |
| gpt-6-luna (OpenAI) | $0.10 | $0.50 | low-cost model for high volume |
| Gemini 2.5 Flash-Lite (Google) | $0.10 | $0.40 | simple routing and short answers |
| Gemini 3.6 Flash (Google) | $0.75 | $3.75 | mid-range, price doubles from 1 January 2027 |
| Claude Sonnet 5.5 (Anthropic) | $2 | $10 | harder questions, tool use |
| gpt-6.1-sol (OpenAI) | $2 | $10 | harder questions, tool use |
The gap between the cheapest and the more capable models is roughly twentyfold. That is why it pays not to use one model for everything. Routine questions such as "when will my parcel arrive" can go to a cheap model, and only a complaint or an unclear case is passed to a stronger one. This is called routing and it is one of the main ways to keep the monthly bill low. Terms such as token or context window are explained in our AI glossary.
Worked example: an e-shop with 3,000 conversations a month
Take a Czech home goods e-shop running on Shoptet that handles 3,000 customer conversations a month. Our assumptions: an average conversation has 6 messages, for each message the model receives about 4,000 tokens (instructions, conversation history and excerpts from the shop's terms) and writes a reply of about 300 tokens. That makes 72 million input tokens and 5.4 million output tokens a month.
- Claude Haiku 5.5 or gpt-6-luna: $9.90 a month, about 214 CZK.
- Gemini 2.5 Flash-Lite: $9.36, about 203 CZK.
- Gemini 3.6 Flash: $74.25, about 1,610 CZK, double from January 2027.
- Claude Sonnet 5.5 or gpt-6.1-sol for every question: $198, about 4,290 CZK.
If the shop sends 90% of conversations to a cheap model and 10% to a strong one, the model bill comes to around 620 CZK a month. Prompt caching for the repeated parts of a prompt, which all three providers offer, lowers it further. The real number depends on the length of your documents and conversations, so we always measure it on a sample of real questions before launch. For online shops we offer a dedicated AI chatbot for e-commerce and Shoptet.
For comparison, off-the-shelf tools often charge per outcome. Intercom charges $0.99 for each case its Fin AI agent resolves, plus seat licences for human agents. At 1,800 resolved conversations a month that is about 38,600 CZK for the AI alone. For a company already on Intercom this is convenient. For a Czech e-shop that needs a Shoptet connection and good Czech answers, a custom chatbot with a one-off build and a low monthly bill often works out cheaper.
Off-the-shelf tool, ChatGPT licences, or a custom chatbot
In practice companies choose between three routes:
| Option | How you pay | When it makes sense |
|---|---|---|
| SaaS chatbot | monthly subscription or a fee per resolved case | standard questions, no special integrations, quick start |
| ChatGPT, Claude or Gemini licences for staff | per user per month, e.g. Claude Team at $25 billed monthly | internal work with text, not customer communication |
| Custom chatbot | one-off build plus model usage | connection to your systems, your own documents, data kept in the EU |
Staff licences and a customer chatbot are two different things. A team with ChatGPT or Claude subscriptions writes emails faster, but nobody on your website gets an answer. If you need both, start with people: a short AI training for companies shows which tasks are worth automating.
For a custom chatbot the knowledge base is what matters most. A chatbot answers only as well as its source material. If your information is scattered across PDFs, emails and an internal wiki, it is worth building an AI knowledge base for your company first, which both the chatbot and your staff can then draw on.
Hidden costs people forget
- Language quality. Cheap models sometimes write clumsy Czech or mix formal and informal address. Testing on real Czech questions is a requirement, not a bonus.
- Wrong answers. A chatbot that invents a discount or a return deadline costs more than the agent time it saves. Fixed rules, answering only from verified sources and handover to a person all help. We cover this in detail in our article on AI guardrails for agents.
- Regulation. According to the European Commission, people talking to a chatbot must be made aware that they are interacting with a machine, and the AI Act transparency rules apply from August 2026. Our guide AI Act for companies explains what that means in practice.
- Personal data. Customers type names, addresses and order numbers into the chat. EU hosting and a data processing agreement with the model provider belong in the budget from day one.
- Keeping sources current. When you change your carrier or price list, the chatbot has to know the same day. Plan for someone in the company to own the source material.
How to work out whether a chatbot pays off
The simplest calculation starts with time. Find out how many questions your team handles each month, how many minutes each one takes, and how many are repeats. An accounting firm in Brno that answers client questions every day, such as "when do I need to send my receipts" or "how is the flat-rate tax calculated", may find that half its questions repeat. If a chatbot resolves even part of them and hands the rest over with a ready summary, it frees several hours a week for work clients actually pay for.
Response speed is another good measure. Klarna announced in February 2024 that its assistant, built on OpenAI models, handled 2.3 million conversations in its first month, two-thirds of its customer service chats, and that resolution time fell from 11 minutes to under 2 minutes. These are figures the company reported itself, at a scale a small Czech e-shop will not reach. The principle is the same, though: a fast answer at night and at the weekend means fewer abandoned carts and fewer repeat questions.
Be conservative in your calculation. Do not plan for the chatbot to resolve everything. Assume it starts with a clearly defined group of questions and add more based on data from real operation.
What a chatbot costs with us
At Prahapp we build chatbots and custom AI agents for a fixed price you know upfront. A project for one clearly defined process starts at EUR 5,000 excluding VAT (about 121,000 CZK at the Czech National Bank rate of 22 October 2026) and goes live within 30 days. Larger projects are billed at EUR 25 to 49 per hour. You pay for model usage directly to the provider, usually a small monthly amount that we estimate upfront on your data. Actions with consequences, such as refunds or order changes, are always approved by a person. Data can stay in the EU or on your own server. All prices are on our AI project pricing page. Prahapp is the Prague studio of Buinsoft Technology.
The fastest way to find out what a chatbot will cost for you is to try one. Within 3 working days we build a free working prototype on your real source material. If you don't like it, you don't pay. Contact us and send a few typical questions from your customers.