AI App Development Cost in 2026: Chatbots, Assistants and Running Costs

Updated October 2026 by the Cliffex team. Prices are in USD.

Short answer: Cliffex budgets a first version with one focused AI feature at $4,000 to $10,000. An assistant that books, takes payments or connects several systems falls in the $10,000 to $20,000 or higher planning tiers; larger platforms start at $20,000+. These are scope-based Cliffex budgets, not market averages. Budget separately for model usage, hosting, messaging and maintenance. The text-only example below calculates $5, $100 or $500 a month for model usage at 10,000 conversations, under stated token assumptions. Those figures are not a complete operating budget or a ceiling.

This guide is for business owners and founders budgeting for AI in a product. It separates build cost from ongoing costs, shows how to calculate a usage budget, and draws on documented Cliffex assistant projects.

FURTHER READING

> How Much Does It Cost to Develop an App? (2026 Pricing Guide)

> Machine Learning and AI Mobile App Development (2026 Guide)

> AI App Development: Features and Assistants

How much does an AI app cost to build?

Build cost depends on the workflow, data, quality requirements and what the AI is allowed to do. Even an answers-only chatbot needs access controls and testing. An assistant that changes bookings, charges cards or handles calls adds integrations, authorization, confirmation steps and human handover. The planning tiers below follow our pricing page and AI development service.

What you are building	Example	Cliffex planning budget
One focused AI feature in a first version	A website or in-app chatbot that answers from your own content and hands over to staff	$4,000 to $10,000
An assistant that takes actions	A booking assistant that checks availability, books, reschedules and takes payment by text message	$10,000 to $20,000 or more
An AI-led platform	A product with several AI workflows, many data sources, admin tools and user accounts across web and mobile	$20,000+, scoped individually
AI added to an app you already have	A summary, search or assistant feature inside your existing product	Quoted after we review your code and data

These are budgets for the first version, not fixed quotes. Your written quote is your price unless the scope changes. AI model usage, SMS and phone charges are billed separately by those providers, and taxes are not included.

Why AI development quotes vary so much

Two AI quotes can cover very different work. Before comparing the totals, check whether each includes:

• Training your own model. Training from scratch adds data collection, labeling, compute and evaluation. Fine-tuning an existing model is a different scope; compare it with prompting and retrieval before committing.
• Enterprise integration. Connecting internal systems, single sign-on and formal security reviews adds implementation and approval work.
• Hourly rates. The same scope costs very different amounts depending on where the team works and how it is staffed.
• Ongoing costs counted as build cost. Check whether the total includes hosting, model usage and support for a stated period.

This guide focuses on an app that calls an existing hosted model through an API, supplies relevant business content and controls access and actions. Training a foundation model from scratch is outside these build ranges.

What moves the price up or down

• Where the answers come from. Document quality, permissions and update frequency matter. Scanned PDFs may need text extraction; changing catalogs and databases need retrieval and refresh processes.
• Answers or actions. Changing records adds authorization, validation and recovery requirements. Confirm consequential actions, enforce permissions in application code, prevent duplicate charges and keep an appropriate audit trail.
• Channels. A web chat avoids telecom integration. SMS adds a messaging provider, consent/opt-out handling and sender requirements; US application-to-person messaging on local ten-digit numbers requires A2P 10DLC registration. Phone adds speech, call handling and tuning.
• Integrations. Each calendar, CRM, payment provider or practice system the assistant talks to adds work and testing.
• Testing and handover. A good build includes a set of test questions, checks for wrong or unsafe answers, and a clear path to a person when the AI cannot help.
• Sensitive data. Health, financial or children’s data needs a review of which providers can receive it and on what terms before any data is shared. For health data in the US, that includes business associate agreements where they apply; see our healthcare app development page.
• Platforms. Separate web, iOS and Android interfaces add implementation and testing; compare the required channels and shared-code options in the quote.

What does an AI feature cost to run each month?

Ask for a running-cost estimate alongside the build quote. Text APIs commonly charge for input and output tokens: pieces of text, not a fixed number of words. The following examples use standard, uncached text rates checked on 29 October 2026, before taxes and extras. They are price examples, not equivalent capabilities or recommendations: Anthropic, OpenAI and Google.

Model tier (examples)	Input per 1M tokens	Output per 1M tokens
Small (Claude Haiku 5.5 at prompts ≤100k tokens; OpenAI gpt-6-luna at short context)	$0.10	$0.50
Mid (Claude Sonnet 5.5; OpenAI gpt-6.1-sol at short context)	$2	$10
Mid (Google gemini-3.1-pro-preview, prompts up to 200k tokens)	$2	$12
Flagship (Claude Fable 5.1; OpenAI gpt-6-astra at short context)	$10	$50

Here is illustrative arithmetic, not a quote or calculator output. Assume 10,000 conversations a month, each totaling 3,000 billable input tokens and 400 billable output tokens across all model calls. Count instructions, retrieved content and any conversation history sent again on each turn. The output allowance includes any billable reasoning/thinking tokens, not only the visible reply. That produces 30 million input and 4 million output tokens a month. Monthly model cost = input millions × input rate + output millions × output rate.

Model tier	Input cost	Output cost	Model cost per month
Small ($0.10 / $0.50)	$3	$2	About $5
Mid ($2 / $10)	$60	$40	About $100
Flagship ($10 / $50)	$300	$200	About $500

Test candidate models against your actual tasks before choosing one on price. Model changes can require prompt, integration and evaluation work. Longer context, retries and multi-step actions can increase usage; caching, batch processing, service tier and regional processing can change rates. These totals exclude retrieval/embedding fees, storage, tool calls, hosting, messaging and support. Application limits and monitoring help control spend, but confirm enforcement and provider billing behavior rather than treating an alert as a hard cap.

SMS and phone can cost more than the AI

For an illustrative US long-code SMS budget, Twilio lists $0.0083 per inbound or outbound SMS segment (29 October 2026). If 10,000 conversations each contain three incoming and three outgoing single-segment messages, 60,000 × $0.0083 = $498 in base charges. Add carrier fees for the actual network and direction, plus number rental, registration/campaign charges, applicable failed-message charges and taxes. Long messages or different character encoding can create multiple segments. This is not an all-in SMS quote.

Budget voice separately using expected call minutes, caller/assistant speaking time, model choice and telephony rates. OpenAI lists gpt-realtime-2.1 audio at $32 per million input tokens and $64 per million output tokens. Audio tokens and text tokens are different units, so those rates alone do not tell you how much more a call costs than a text chat. Include phone charges and any separate transcription or speech services used by the design.

Chatbot, assistant or AI-led app: which do you need?

	Answers-only chatbot	Assistant that takes actions	AI-led app
What it does	Answers questions from your content and passes harder ones to staff	Books, reschedules, takes payments or updates records, with confirmation first	The AI is the core of the product, across several workflows
Where it runs	Your website or app	Website, app, SMS or phone	Web and mobile apps with accounts and admin tools
Cliffex planning budget	$4,000 to $10,000	$10,000 to $20,000 or more	$20,000+
Main running costs	Model usage, hosting	Model usage, hosting, SMS or call charges, payment fees	Model usage, hosting, storage, monitoring
Main risk	Wrong answers	Wrong actions	Cost and quality at scale

Choose the smallest workflow that solves the business problem. An answers-only pilot can test content and demand, but accurate answers alone do not prove that actions are safe. Before enabling booking or payments, test permissions, confirmation, duplicate prevention, failures and human escalation. Our MVP cost guide explains first-version planning.

What we learned building AI assistants for clients

From 2023 to 2025 we built an AI patient assistant for dental practices: a website chatbot on the OpenAI API under a signed business associate agreement, plus a Twilio phone assistant. We handed it over in January 2025. The project illustrates three budgeting questions.

1. Price the running cost early. In December 2023, pricing tiers based on tokens per patient chat were agreed for the product. If your own pricing depends on usage, specify measurement and reporting before launch.
2. Budget for tuning after the first release. December 2024 and January 2025 work included tuning the greeting and listening window. Include evaluation and revisions in the project budget.
3. Plan data access before using patient information. The AI-provider BAA was approved in March 2024. For new work, confirm applicable agreements and safeguards before sharing PHI; development can start with test data. A BAA alone does not establish compliance.

Our September 2026 project records also describe an SMS receptionist for hair salons, still in progress, covering booking, rescheduling and payments. The design includes confirmation before cancellation or charging and a salon-defined approval ceiling. Live SMS testing was awaiting provider verification at that point. Permissions, payment controls and provider onboarding belong in the scope and launch plan, not just the model integration.

Has AI made app development cheaper?

Our engineers work with frontier AI coding agents at every step of planning, coding and review, and every release goes through automated tests and manual QA. That is how a small team delivers quickly without cutting corners. This does not imply a fixed percentage saving on an AI product. Workflow decisions, evaluation, failure handling and data obligations still need engineering work. Our AI-built app production checklist covers launch preparation, and our AI-built app development service helps with that work.

Costs after launch

• Model usage. Billed by the AI provider each month, usually to your own account.
• Hosting and storage. The app, database and any document index.
• Model updates. Providers release new models and retire old ones, so prompts and test questions need re-checking from time to time.
• Content updates. The assistant is only as current as the content it reads.
• Support. Every Cliffex launch includes 30 days of support. After that, maintenance is $25 to $49 an hour with an estimate first; see our app maintenance guide.

How to get an accurate quote for an AI app

1. Write down the one job the AI should do first, and what it must never do.
2. List where the answers come from: documents, a website, a database or another system.
3. Decide the channel: web chat, in-app, SMS or phone.
4. Note any sensitive data, such as health or payment details.
5. Estimate monthly volume, even roughly, so running costs can be modeled.
6. Collect 20 to 30 real questions customers ask. Use them as an initial test set, then add failure, permission and adversarial cases; remove personal data you do not have permission to share.

Try our app cost calculator for an indicative build range, or send your project details for a written quote. Ask for separate build and operating-cost estimates with usage assumptions. Learn more about our AI app development work.

Frequently asked questions

What do AI apps cost?

Cliffex planning tiers start at $4,000 to $10,000 for a first version with one focused AI feature, $10,000 to $20,000 or more for an integrated assistant, and $20,000+ for larger platforms. Scope determines the quote; model usage and other operating costs are separate.

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

Under this article’s assumption of 30 million billable input and 4 million billable output tokens per month, the illustrated text rates produce $5, $100 or $500 in model charges. These are examples, not typical bills or upper limits. Hosting, retrieval, tools, support and any SMS or voice charges are additional.

How expensive is it to build an AI agent?

Our $10,000 to $20,000 or higher planning tiers cover integrated assistants, subject to scope. Booking and payment actions add permissions, confirmation, failure handling and testing. The label “agent” alone is not enough to estimate the work.

Can I build an AI chatbot for free?

Some tools and model APIs offer limited free tiers for prototypes. Check availability, usage limits and data-use terms before sending business content; use synthetic data where appropriate. A production budget must also cover access controls, evaluation, human handover and operations. Free access is not a commitment to free operation at scale.

Do I need to train my own AI model?

Start by testing an existing model with suitable instructions and retrieval from approved content. Fine-tuning an existing model may be worth evaluating for quality, consistency, latency or cost; it is different from training a foundation model from scratch. Decide using measured results and the cost of data preparation, evaluation and operation.

How do I compare two AI app quotes?

Compare the same workflow, platforms, data sources, integrations and acceptance tests. Check who pays for evaluation, hosting, model usage, monitoring and post-launch fixes. A lower build price may exclude work the other quote includes. Our app cost guide covers broader build budgeting.

Will you sign a business associate agreement for a healthcare AI app?

Yes. Cliffex signs a BAA before accessing protected health information when the work involves PHI. A BAA alone does not make an app compliant. We also review which providers receive the data, their eligible services and applicable agreements before sharing it. See HHS guidance on business associates.