A chatbot on a no-code platform costs from $0 to a few dozen dollars a month per plan, before per-conversation AI charges. A custom AI chatbot trained on your company’s knowledge base costs from $700 to build at Strymba, plus a monthly bill for the language model: in our worked example below, roughly $2 to $190 per thousand conversations, depending on the model. The gap between those numbers comes from the knowledge base, integrations, human handoff, and the testing done before launch.
This guide breaks the budget down: what you pay once, what you pay every month, how to work out the cost of one conversation from the official OpenAI, Anthropic, and Google price lists, and when a chatbot pays for itself. All prices were checked on November 16, 2026. If you first want to know how a chatbot finds its answers, read How do AI chatbots work?
How much does a chatbot cost? The short answer
| Option | One-time cost | Monthly cost | Good fit for |
|---|---|---|---|
| No-code chatbot platform | $0, you set it up yourself | from $0 (free tiers with limits) to $19–99 per plan or per user, plus per-seat or per-AI-answer fees | menus, bookings, lead capture, simple FAQs |
| Custom AI chatbot on your knowledge base | from $700 at Strymba, from 7 days | AI model usage, hosting, knowledge base updates | a website where customers ask in their own words about terms, products, and services |
| AI chatbot with integrations (CRM, catalog, order status) | more than a basic bot: each integration and scenario adds work | the same, plus integration upkeep | sales and support flows where the bot creates leads or tickets and checks statuses |
Chatbot platform vendors’ own pricing guides put fully custom enterprise builds much higher: Holy Grail’s chatbot pricing guide, for example, puts a fully custom chatbot at $10,000–80,000 to build (checked November 16, 2026). The spread is wide because “chatbot” covers very different jobs, from a widget that answers a few questions to a bot wired into your CRM that hands conversations to your team.
Chatbot pricing: platform plans vs a custom build
A no-code platform gives you a ready-made builder where you assemble flows and connect channels yourself. You pay a subscription, and the setup is your own time. Here’s what popular platforms charge on their official pricing pages (November 16, 2026):
| Platform | Plan | Price |
|---|---|---|
| Tidio | Free | $0: 50 billable conversations, 50 Lyro AI conversations one-time |
| Tidio | Starter | $24.17/month billed annually, 100 billable conversations |
| Tidio | Growth | from $49.17/month billed annually, from 250 billable conversations |
| ChatBot.com | Essential: 1 AI agent, 10 AI resolutions | $19 per user/month billed annually ($25 monthly) |
| ChatBot.com | Growth: 10 AI agents, 200 AI resolutions | $79 billed annually ($99 monthly) |
| Intercom | Essential with Fin AI Agent | from $0.99 per Fin outcome, plus a per-seat fee |
Chatbot pricing models
Platforms charge in one of three ways, often combined: a flat plan with usage limits (Tidio), per seat (Intercom, ChatBot.com), or per AI answer or resolved conversation (Intercom’s Fin AI Agent). A custom build adds a fourth model: you pay the model provider per token. So price the plan for your real volume, not the entry tier: at $0.99 per Fin outcome on Intercom, 2,000 outcomes a month cost about $1,980 before any seats.
A platform is enough when the bot walks people through buttons, books appointments, collects contact details, or answers a dozen standard questions. A custom build makes sense when:
- customers ask in their own words, and answers must come from your documents, price list, or catalog;
- the chatbot has to work with your CRM, inventory, or order statuses through an API;
- you want control over which model answers, what data it receives, and where conversation logs are stored;
- answers must be tested on real customer questions before the bot speaks for your company;
- per-outcome fees would outgrow the cost of paying for the model directly.
What drives AI chatbot pricing
The technical core of a chatbot is fairly standard, so the price comes from the work your business needs on top of it.
- The size and state of the knowledge base: how many documents, pages, and tables the bot must know. A current price list and ready answers to common questions connect quickly. Scattered PDFs, outdated terms, and conflicting data have to be sorted out and approved first.
- Integrations. Every system the chatbot exchanges data with adds work: a CRM for leads and tickets, a product catalog, order status checks, a helpdesk. Messaging channels such as WhatsApp, Messenger, Instagram, or Telegram are connected separately, and some carry their own usage fees.
- Human handoff: when the bot calls a person, who gets the conversation, in which channel, and what context travels with it. More rules mean more setup and more testing.
- Languages. Each one needs its own knowledge base or a checked translation, plus its own test questions, so a bilingual chatbot costs more to prepare than a single-language one.
- Testing before launch. We build an evaluation set from real customer questions and check whether the bot answers from its sources, how fast it responds, what a conversation costs, and how often it escalates to a person. Topic limits, protection against prompt manipulation, and logging belong here too.
AI chatbot cost per month
After launch, the running budget has four parts.
- AI model usage. Language models are billed by tokens, the units of text they read and write. The bill depends on the number of conversations, the model you choose, and how many knowledge base passages the bot pulls into each answer. The next section shows how to estimate it.
- Hosting for the bot, its search index, and the conversation log. If they share the server that already runs your website, the extra cost can be small. A separate server or a managed cloud service adds its own bill.
- Knowledge base updates. When prices, terms, or product ranges change, the bot should know the same day. That takes hours from someone on your team or a contractor, more of them the more often things change.
- Maintenance: reviewing conversation logs, adjusting instructions, updating integrations when your CRM or website changes, and moving to newer model versions.
How much does one conversation cost? A formula and a worked example
The cost of one conversation = (input tokens × input price + output tokens × output price) ÷ 1,000,000.
Input tokens are the bot’s instructions, the knowledge base passages it retrieved, the conversation so far, and the question itself. Output tokens are the bot’s reply. The model receives all of this again with every message, so input is usually several times larger than output.
Our example assumptions: a conversation with 4 customer questions; for each question the model reads 3,500 tokens (1,000 of instructions, 2,000 of knowledge base passages, 500 of history and the question) and writes a 250-token answer. Per conversation, that’s 14,000 input tokens and 1,000 output tokens. Token counts for the same sentence vary by language and model, so on a real project we measure them on your own conversations.
| Model | Price per 1M tokens: input / output | One conversation | 1,000 conversations a month |
|---|---|---|---|
| OpenAI GPT-6 Luna | $0.10 / $0.50 | ≈ $0.002 | ≈ $1.90 |
| Anthropic Claude Haiku 5.5 | $0.10 / $0.50 | ≈ $0.002 | ≈ $1.90 |
| Google Gemini 3.8 Flash | $0.75 / $3.75 | ≈ $0.014 | ≈ $14 |
| OpenAI GPT-6.1 Terra | $1.00 / $5.00 | ≈ $0.019 | ≈ $19 |
| Anthropic Claude Sonnet 5.5 | $2.00 / $10.00 | ≈ $0.038 | ≈ $38 |
| Anthropic Claude Opus 5.5 | $4.00 / $20.00 | ≈ $0.076 | ≈ $76 |
| OpenAI GPT-6 Astra | $10.00 / $50.00 | ≈ $0.19 | ≈ $190 |
Prices come from the official OpenAI, Anthropic, and Google pages, standard processing, checked on November 16, 2026. Caching the repeated part of the prompt (instructions and fixed passages) lowers the input cost: all three providers charge several times less for cached input.
Models differ by up to a hundred times in price, so the model is chosen from test results. A smaller model often handles standard knowledge base questions well, and a larger one goes only where it is clearly more accurate on your evaluation set.
Compare this with per-outcome platform pricing. At $0.99 per Fin outcome, 1,000 outcomes cost $990 a month. Paying a model provider directly, the same volume costs $2–190 in model usage in our example, plus hosting and maintenance. Per-outcome plans buy you a ready product and support; a custom build buys you control and a lower variable cost at higher volumes.
Hidden chatbot costs people forget to budget
- Preparing the content. Your team collects and approves the prices, terms, and exceptions the bot may quote, even when a contractor builds the bot.
- Testing before launch: an evaluation set, test runs, and instruction fixes. Skip it, and you learn about the bot’s mistakes from customers.
- Reviewing logs after launch. The first weeks bring questions that weren’t in the tests, and the knowledge base gets filled in to cover them.
- Model price changes. Providers revise their prices: for example, Gemini 3.8 Flash doubles in price on January 1, 2027: input goes from $0.75 to $1.50 and output from $3.75 to $7.50 per million tokens (Google’s pricing page). Build the bot so you can switch models without rebuilding it.
- Channel fees: some messaging channels and platforms charge per message, per contact, or per subscriber on top of everything else.
- Seat creep on platforms. A per-seat plan grows every time another person needs access to the inbox.
How to lower your AI chatbot costs
Part of the budget can be cut during the build, and you can check the effect on the same evaluation set.
- Start with one channel and one language, such as your website in English. A second channel or language is easier to add once the first one works and you can see real questions.
- Choose the model through testing: compare 2–3 models on the same set of your questions. If a smaller model is just as accurate, the monthly bill can drop by an order of magnitude or more.
- Send the model less text. Short, well-labeled knowledge base passages instead of whole documents cut input tokens in every message.
- Cache the fixed part of the prompt. Instructions and standing passages repeat in every conversation, and cached input costs several times less.
- Defer background work. Log reviews, ticket classification, and reports don’t need real-time answers. Batch processing at OpenAI, Anthropic, and Google costs 50% less than standard.
- Name an owner for the knowledge base. A person who updates prices and terms the day they change costs less than cleaning up after customer complaints.
How to tell whether a chatbot will pay off
Only your own data can show the exact payback, but you can estimate the order of magnitude in advance.
Monthly savings = repeat inquiries per month × share the bot resolves without a person × minutes per answer ÷ 60 × hourly cost of the employee.
Subtract the monthly running costs: model usage, hosting, maintenance, and hours spent on knowledge base updates. Divide the one-time build cost by the difference, and you get the number of months to payback.
The containment rate (the share of conversations resolved without a person) shows up on an evaluation set of your real questions before launch, and in the conversation log afterwards. A figure from someone else’s slide deck won’t tell you yours. Also count what’s hard to put a price on: answers at night and on weekends, and leads that reach your team with the request already clarified.
A chatbot won’t pay off if repeat questions are rare, every answer needs a manager’s calculation, or your data changes daily with nobody responsible for updating it. We covered those cases in the “When you don’t need a chatbot” section of our guide to how chatbots work.
What an AI chatbot costs at Strymba
An AI chatbot for your website starts from $700 at Strymba and takes from 7 days. The final price depends on the size of the knowledge base, CRM or helpdesk integrations, and the number of scenarios. The bot works from an approved knowledge base and hands complex conversations to a manager together with the history. Before launch, we test it on an evaluation set of real questions: grounded accuracy, speed, cost per conversation, and escalation rate.
We’ve been building AI solutions since 2023. For Growexa, we’ve developed and supported the platform since 2023, including AI forms that call OpenAI and Claude models through the API. For Würth Ukraine, we built AI search across the product catalog that asks clarifying questions and helps buyers find the right item. In LiteLang, Claude splits a tutor’s PDF into sections and lessons, and the tutor reviews the result.
AI chatbot pricing FAQ
01How much does it cost to create a chatbot?
On a no-code platform you can build one yourself on a free tier or a plan such as ChatBot.com Essential at $19 per user a month, paying with your own setup time. A custom AI chatbot on your knowledge base starts from $700 at Strymba, and the price grows with the knowledge base, integrations, languages, and handoff rules.
02How much does it cost to use an AI chatbot?
The model usage in our example ranges from about $2 to $190 per 1,000 conversations, depending on the model. Hosting, knowledge base updates, and maintenance come on top. We measure the real cost per conversation while testing on your questions.
03Are chatbots free or paid?
Platforms such as Tidio have free tiers with limits: Tidio’s free plan includes 50 billable conversations. AI answers based on your knowledge base are never free: every conversation is billed by the model provider, either directly or through the platform’s per-answer fee.
04Can I buy my own AI chatbot?
Yes. With a custom build, the chatbot, its instructions, and its knowledge base are set up for your business, and you pay the model provider directly for usage. On a platform, you rent the product and pay for seats or AI answers for as long as you use it.
05How long does it take to launch a chatbot?
At Strymba, from 7 days. The timeline depends on the size of the knowledge base, integrations, and the number of scenarios, as well as how ready your material is and how many approvals it needs.
Next step
Let’s estimate your chatbot budget
Tell us which customer questions the chatbot should answer and which systems it has to work with, and we’ll propose a format, budget, and timeline.



