AI Chatbot Development Cost in 2026: What You Pay to Build, Run and Maintain

Ask five vendors what an AI chatbot costs and the answers will run from a few thousand dollars to half a million. They are not all wrong. They are pricing different things under the same name, and the word “chatbot” now covers everything from a website FAQ widget to an assistant that answers from ten years of company documents and hands the hard questions to a person. This guide breaks the AI chatbot development cost into the three bills you actually pay, building it, running it and keeping it accurate, using published ranges and the model providers’ own price lists.

The short answer

Vendor-published 2026 ranges put a simple rule-based or FAQ chatbot at roughly USD 3,000 to 30,000, an assistant that answers from your own documents at roughly USD 30,000 to 150,000, and an enterprise assistant connected to several business systems at USD 100,000 and well beyond. Annual maintenance typically adds 15 to 25 percent of the build cost. The model itself is often the smallest line: at November 2026 list prices, a typical support conversation costs between a fraction of a cent and about 26 cents in model fees, depending on which model you choose.

Build cost by type of chatbot

There is no independent survey of chatbot build prices, so the honest picture comes from comparing what development firms publish. The table puts three recent 2026 guides side by side.

Type of chatbot	What it does	Published build ranges	Typical build time
Rule-based or FAQ	Follows set paths and answers a fixed list of questions	USD 3,000 to 15,000 (Space-O); USD 5,000 to 30,000 (Monocubed)	2 to 8 weeks
LLM assistant on your documents (RAG)	Answers in natural language from your own knowledge base	USD 30,000 to 80,000 (Keyhole); USD 50,000 to 150,000 (Space-O)	8 to 20 weeks
Enterprise assistant across several systems	Reads and writes to CRM, ERP or helpdesk data, with compliance controls	USD 100,000 to 300,000+ (Keyhole); USD 150,000 to 500,000+ (Monocubed)	4 to 9 months

The overlap between rows is the useful part. A “simple” assistant that turns out to need a CRM connection, or a document assistant that has to meet a security review, moves up a row, and that move is where most budgets break. Keyhole makes the same point: projects often move from its basic tier to its mid-level tier once integration or compliance needs surface during discovery.

What it costs to run

Running costs are easier to pin down than build costs, because the model providers publish their prices. The table below works through one typical support conversation of six exchanges, where each exchange sends about 3,000 tokens to the model (your instructions, the passages retrieved from your documents and the conversation so far) and gets about 250 tokens back.

Model tier	Examples and list price per million tokens (input / output)	Cost per conversation	10,000 conversations a month	50,000 conversations a month
Small, fast models	Claude Haiku 5.5 or gpt-6-luna: USD 0.10 / 0.50	About USD 0.003	About USD 26	About USD 128
Mid-tier models	Claude Sonnet 5.5 or gpt-6.1-sol: USD 2 / 10	About USD 0.05	About USD 510	About USD 2,550
Top-tier models	gpt-6-astra: USD 10 / 50 (Claude Fable 5.2 is USD 8 / 40)	About USD 0.26	About USD 2,550	About USD 12,750

Two conclusions follow. First, model choice moves the running cost by a factor of about a hundred, so the right question is not “which model is best” but “which is the cheapest model that answers our questions well enough”, and that is something you test, not assume. Many good assistants send routine questions to a small model and only the hard ones to a larger one. Second, even at the top tier the model bill is modest next to the people cost of the conversations it handles.

The model is not the only running cost. Monocubed’s guide estimates USD 200 to 1,000 a month for hosting, USD 50 to 500 for a vector database that stores your searchable documents, and USD 50 to 200 for monitoring. On top of that comes maintenance, which Keyhole and Space-O both put at 15 to 25 percent of the build cost each year: keeping the knowledge base current, reviewing answers, and adjusting the assistant as your products and policies change.

What moves the price

1. The state of your knowledge base. An assistant can only be as good as the documents behind it. Outdated policies, duplicate versions and knowledge that lives only in people’s heads all have to be dealt with before launch, and this preparation is regularly underestimated.
2. Integrations. Answering questions from documents is one job. Looking up an order, changing a booking or opening a ticket means connecting to live systems, with the security and testing that comes with it.
3. How accurate it has to be. A wrong answer about opening hours is an annoyance. A wrong answer about a refund policy or a dosage is a liability. The higher the stakes, the more you spend on testing, answer checks and review before and after launch.
4. Handover to people. The best assistants know what they should not answer. Routing a conversation to a person, with the context attached, is a feature you have to design and build.
5. Channels and languages. Web chat is the cheapest channel. WhatsApp, a mobile app, email and especially voice each add design and testing work, as does every extra language.
6. Compliance. Keyhole estimates that regulated industries such as banking, healthcare and government add 25 to 35 percent, for access controls, audit trails and data handling reviews.

Build or buy

Off-the-shelf chatbot platforms are the cheapest way to start: Keyhole puts setup at anywhere from nothing to about USD 20,000, plus usage fees. They suit standard support questions on a website, where speed matters more than control. A custom build earns its cost when the assistant has to answer from material a platform cannot reach, work inside your own systems, follow your security rules, or behave in a way that is part of your product rather than bolted onto it.

A sensible test is to price both for the same twelve months, including usage fees, the people who will maintain the platform’s content, and what happens to your data and setup if you leave. If you are already planning AI features in a product, our guide to adding AI to an existing mobile app covers the same build-or-buy decision from the product side.

Chatbot or AI agent?

A chatbot answers. An agent acts: it can complete a task across several systems, such as processing a return from start to finish. Agents cost more to build and to govern, and the label is being stretched. In a 2025 survey of 265 customer service leaders, Gartner found AI agents ranked outside the top ten most valuable service technologies, with leaders wary of “agent-washing”, rule-based products marketed as agentic (Gartner, August 2025). For many companies, an assistant that answers well is the right first step, and the agent comes later. When you get there, our breakdown of what an AI agent costs to build picks up where this guide ends.

Accuracy is where the money goes

The cheapest part of an assistant is getting it to talk. The expensive part, and the part that decides whether people keep using it, is making its answers trustworthy. For Pfizer, Appson built an AI platform, to HIPAA and GxP standards, that puts the expertise of senior researchers within reach of every junior researcher. Every answer is checked against the senior researcher’s own material and given a match percentage and a rating of good, average or below average, so the team always knows how far to rely on it.

For Wordibly, a US transcription company, the same principle runs the other way: AI does the first pass and people correct it, so speed comes from the machine and accuracy from the humans. Neither design is the cheapest possible build. Both are why the systems are used. Whichever route you take, budget for checking answers, not just generating them. If you plan to prove the idea first, our guide to what an AI pilot should prove sets out what to measure before you fund the full build, and our comparison of RAG and fine-tuning explains the two main ways to teach a model your material.

Frequently asked questions

How much does it cost to build an AI chatbot in 2026?

Published 2026 ranges run from about USD 3,000 to 30,000 for a rule-based or FAQ chatbot, USD 30,000 to 150,000 for an assistant that answers from your own documents, and USD 100,000 and up for an enterprise assistant connected to several systems. Where a project lands depends mostly on integrations, accuracy requirements and the state of the knowledge base.

How much does an AI chatbot cost to run each month?

Model fees depend on traffic and model choice. At November 2026 list prices, 10,000 typical support conversations a month cost roughly USD 26 on a small model, USD 510 on a mid-tier model and USD 2,550 on a top-tier model. Hosting, a vector database and monitoring add a few hundred to around USD 1,700 a month, and maintenance adds 15 to 25 percent of the build cost a year.

How long does it take to build an AI chatbot?

Published estimates put a rule-based chatbot at two to eight weeks, an assistant on your own documents at eight to twenty weeks, and an enterprise assistant across several systems at four to nine months. Preparing the knowledge base is often what sets the pace.

Is it cheaper to use a chatbot platform than to build one?

To start, yes. Platforms can be set up for little or nothing plus usage fees. A custom build pays off when the assistant needs your own data, your systems, your security rules, or behavior that a platform cannot offer. Compare twelve months of total cost, not the first invoice.

Which AI model should a chatbot use?

The cheapest one that answers your real questions well enough, which you find by testing several against a set of your own questions. Many assistants use a small model for routine questions and send only the difficult ones to a larger, more expensive model.

Budgeting for the assistant you actually need

The price of an AI chatbot is set less by the technology than by three decisions: what it must know, what it must be allowed to do, and how sure you need to be of its answers. Make those three calls first and the estimates stop disagreeing, because every vendor is finally pricing the same thing.

If you are scoping an assistant now, our AI chatbot development team can help you set those three decisions and cost the build around them. Talk to our team to start.