AI agent cost has three parts: the build, monthly usage and upkeep. Most price guides cover only the first. This one covers all three, using our own research: 183 client job posts from Upwork, Freelancer.com, PeoplePerHour and Hacker News, mostly posted between June and September 2026, plus package prices and platform rate cards we recorded in September 2026. The Upwork posts came from targeted searches, so treat the shares below as indicative. Any Myrado prices below match our published pricing.
AI agent pricing at a glance (2026 market ranges)
The table shows what the market charged in September 2026. “Marketplace packages” are entry and top tiers on Fiverr and the Upwork Project Catalog; “Upwork hiring guides” are ranges from Upwork’s hiring pages for AI agent, n8n and AI voice agent developers.
| Agent type | Marketplace packages (entry → top tier) | Upwork hiring guides (Sept 2026) | Myrado starting price |
|---|---|---|---|
| FAQ or support chatbot | $250 → $1,400 (two RAG chatbot packages) | $500–2,000 | AI support agent from $2,000 |
| Workflow automation with AI steps | $25 → $2,500 (n8n and AI automation) | $500–2,000 per n8n workflow; $2,000–5,000 for a workflow agent | From $800 per workflow |
| Knowledge-base (RAG) agent | Fiverr starting prices: median $150, middle half $90–250 | $5,000–15,000 | Knowledge-base assistant from $2,500 |
| Voice agent or AI receptionist | $50 → $2,225 | $600–6,000 for one scenario; $2,400–24,000 in production | AI receptionist (voice) from $2,000 |
| Multi-agent system | Rare; one Fiverr gig in our sample had a $345 “multi-agent” tier | $10,000–30,000 | Custom from $8,000 |
One name covers very different products. A $50 voice package is usually one agent on a hosted platform with a basic script. Upwork’s $2,400–24,000 production range assumes telephony, a knowledge base and analytics. Even Upwork’s chatbot-developer guide quotes just $150–300 for a RAG bot. Compare what is being built before you compare prices.
On the Western agency price lists we reviewed, custom builds typically started at $5,000–8,000, and care ran $750–2,500 a month.
Why some guides say $40,000–400,000
Price guides ranking for this query in September 2026 quote $40,000–80,000+ and even $10,000–400,000+; Reddit threads put installations at $3,000–15,000. Figures that high usually assume enterprise scope: a full team, a paid discovery phase, security and compliance reviews, legacy integrations, sometimes custom model training. A small-business agent on one to three channels, built on existing models and your current tools, is a different purchase. If its quote runs to five figures, ask what the extra scope is.
What marketplace budgets buy
Of the 183 posts in our sample, 57 stated a budget: the median was about $160, 68% were under $500, and none went above $5,000. Larger datasets point the same way, with a longer tail: Marathon Insights found 50.7% of AI fixed-price Upwork posts under $500 (February–August 2026), and Upwatcher found a $150 median across 4,089 AI posts (the 30 days ending May 16, 2026). For hourly rates, see our guide to hiring AI agent developers in Ukraine.
What those budgets bought, summarized from posts in our sample:
| Budget | What clients asked for |
|---|---|
| Up to about $300 | A website FAQ bot with lead capture; a menu-based WhatsApp bot; one Retell or Vapi agent set up; a fix for a broken workflow. One post wanted a “production-ready” LangGraph system for $30. |
| About $500–1,500 | A voice receptionist for a trade business, with several call modes and email summaries; a voice sales agent that qualifies leads into the CRM; a helpdesk AI agent |
| About $3,000–5,000 | Bilingual voice and chat lead qualification with hand-off and follow-up; document extraction linked to a legal platform |
| $15,000 and up | No post in our sample stated a budget this high. Projects this size tend to come through direct channels, not job boards. |
A typical low-budget post draws 100–300 proposals, and at those prices a seller can afford a few hours of setup, not weeks of testing.
What the packages don’t mention
We read 18 Fiverr gigs and 9 Upwork Project Catalog listings line by line. In Fiverr search, median starting prices were $95 for “ai voice agent” and $100 for “ai agent” (first 48 results for each search). Most voice gigs follow the same ladder: a basic agent, then booking and CRM sync, then inbound plus outbound calls with reporting. No voice package said how many call minutes were included, what knowledge-base limits applied or which platform fees were extra. Only one seller stated support days, and none offered acceptance criteria or monitoring. A cheap setup can be fine for a low-stakes FAQ widget, but its listed price is rarely the full cost.
What drives AI agent development cost
Build price follows scope. These drivers move a quote the most:
| Cost driver | Cheaper end | Pricier end |
|---|---|---|
| Channels | One website chat widget | Phone, WhatsApp, website and Instagram, each with its own rules |
| What the agent does | Answers questions | Takes actions: books, updates the CRM, creates tickets, sends payment links |
| Integrations | A Google Sheet or one calendar | CRM, practice-management or field-service software, helpdesk, custom APIs |
| Knowledge | 30 clean FAQ answers | Thousands of pages of PDFs, some of them outdated |
| Quality bar | “Looks fine in the demo” | Tested on a set of your real questions with written pass criteria |
| Voice | One language, simple script | Interruptions, accents, live transfers, after-hours and overflow modes |
| Control | Public information only | Role-based access, human approvals, audit logs, your own hosting |
Two drivers are easy to underestimate. Actions cost more than answers because each needs error handling: a full calendar, a CRM timeout, a customer who changes their mind. Testing takes real hours, but it separates a demo from something you can leave running overnight.
Subscription, DIY or custom build?
| Option | What you pay | Good fit | Watch out for |
|---|---|---|---|
| Done-for-you AI receptionist subscription | $99–149 a month plus $1,500 setup (one public price list, Sept 2026) | Standard call answering and booking | Call caps on lower tiers; check which integrations are included |
| DIY on a hosted voice or chat platform | Per-minute or per-message fees, plus your time | Simple flows and a hands-on team | Testing, updates and fixes are on you |
| Custom build | From $2,000 with us, usage at cost, optional care | Your own rules, CRM, hand-offs, several channels | Needs a clear scope and acceptance criteria |
If your flows are fixed (pick a service, pick a time, confirm), a button-based chatbot may be all you need. We compare the two in AI agent vs chatbot.
Run costs: model tokens, voice minutes and hosting
Run costs grow with volume, not with how clever the build is. Get a usage estimate before you sign, and ask whose account it’s billed to.
Model tokens (chat and text agents)
Model providers bill per token (a token is about three-quarters of an English word), counting what the model reads and what it writes:
Monthly model cost ≈ conversations per month × average tokens per conversation × price per token
Per-token prices change often, so check the provider’s pricing page. For a rough budget, one public studio price list we reviewed (September 2026) bills model usage separately at 1,500–15,000 hryvnias a month (roughly $34–335), and a published chatbot price guide puts hosting from about $10 a month. Long instructions, knowledge-base passages attached to each answer (how RAG works) and long chats push token use up. The only reliable way to know your average tokens per conversation is to measure it on a demo with your real questions.
AI voice agent cost per minute
Voice platforms bill per minute, so AI voice agent cost is easy to estimate once you know your call volume. List prices we recorded in September 2026:
| Voice platform | Published price at research time | Note |
|---|---|---|
| Vapi | $0.05 per minute | Platform fee; other providers billed on top |
| Retell | $0.07–0.31 per minute | Listed range; see the pricing page for what it includes |
| Bland | $0.12–0.14 per minute | Listed rate |
For scale, take an illustrative 1,500 call minutes a month (say, 500 three-minute calls). At Retell’s listed range that’s $105–465 a month; at Bland’s, $180–210. Vapi’s platform fee would be $75, plus the other providers’ charges. Then add a phone number, carrier minutes and SMS fees if the agent texts back missed calls. Texts should go only to people who agreed to hear from you, under local rules (for example, A2P 10DLC registration for business texting in the US). More on phone agents: AI voice agents.
Messaging and hosting
Meta bills WhatsApp Business messages on its own rate card, and messages your business starts need the customer’s opt-in. At research time, n8n listed its cloud plans at €20, €50 and €667 a month. The self-hosted Community edition has no license fee: you pay for a small server instead, and for someone to keep it patched. More in n8n and AI automation.
AI agent maintenance cost: what care plans cover
An agent isn’t launch-and-forget. Models get retired, APIs change, your prices and policies change, and customers ask things nobody planned for. In our sample, 8 of the 183 posts came from clients who needed a broken agent or workflow fixed or taken over.
Market ranges for ongoing care in 2026:
- Agencies: $750–2,500 a month on the US-dollar price lists we reviewed.
- Voice agents: Upwork’s hiring guide for Vapi developers lists $1,500–4,000 a month for ongoing management.
- Our care plans: Starter from $250 a month, Growth from $450 and Custom from $900. They start after the 30 days of tuning included in every build.
A care plan is worth paying for when it names specific work:
- Monitoring and alerts when a flow or integration fails
- Regular reviews of real conversations or call recordings, with fixes
- Knowledge-base updates when prices or policies change
- Re-testing after a model or platform update
- A stated response time for incidents
- Small improvements every month
You can skip care if you get documentation, an editable knowledge sheet and access to every account; your team then owns the upkeep. Don’t accept a care fee that quietly hides a markup on your usage.
Three illustrative budgets
These are illustrative scenarios, not client cases. Volumes are assumptions; build prices are our published starting prices.
| A. Online store support chat | B. After-hours AI receptionist | C. Internal knowledge assistant | |
|---|---|---|---|
| What it does | Website chat for product, shipping and returns questions, with hand-off; Growth adds WhatsApp and order status | Answers after-hours calls, books jobs, texts back missed callers who opted in, emails a summary | Answers staff questions from SOPs in Slack or Teams, with citations and access control |
| Build (starting price) | $2,000 (AI support agent, one channel); $3,500 on Growth | $2,000 (AI receptionist); $3,500 on Growth with chat and CRM | $2,500 (Knowledge-base assistant); $8,000 on Custom with approvals |
| Time to launch | 1–2 weeks; 2–3 weeks on Growth | 1–2 weeks; 2–3 weeks on Growth | 2–3 weeks; 3–6 weeks on Custom |
| Illustrative volume | 1,000 chats a month | 1,500 call minutes a month | 40 staff asking a few questions a day |
| Main usage costs | Model tokens, hosting; WhatsApp fees on Growth | Voice minutes ($105–465 a month at Retell’s range), phone number, SMS | Model tokens, hosting, indexing, vector storage |
| Optional care after 30 days | From $250 a month; from $450 on Growth | From $250 a month; from $450 on Growth | From $250 a month; from $900 on Custom |
| Illustrative first-year total (smaller build, with care) | About $5,300–8,900 | About $6,000–10,300, plus phone number and SMS | About $5,800–9,400, plus indexing and storage |
How to read the last row: build + 11 months of care (month one is the included tuning) + 12 months of usage. For B, that’s $2,000 + 11 × $250 + 12 × $105–465 in call minutes; for A and C, the $34–335 token band plus $10 hosting. It’s arithmetic on published prices, not a quote. Without care, subtract $2,750. We measure real token use on your demo and estimate it before you sign. Scenario C is a typical knowledge-base assistant (RAG) project.
Hidden costs and how to compare quotes
Costs that don’t show up in the headline price
- Usage markups. “I will NOT accept hidden monthly developer markups,” one client wrote in a post we collected. Ask whether tokens and minutes are billed at cost on your own accounts.
- Vendor-owned accounts. If the phone number, workflows or model keys belong to the vendor, leaving can mean a rebuild.
- Your team’s time. Someone has to collect FAQs, fix outdated documents and review hand-offs in the first weeks.
- API access. Some CRM and helpdesk plans limit it. Check yours before the build starts.
- Change requests. Ask how post-launch changes are priced: hourly or a fixed fee per change.
A checklist for comparing quotes
| Question to ask | Good answer | Red flag |
|---|---|---|
| What exactly is in scope? | Channels, integrations, actions and knowledge sources are listed | “A custom AI agent” and nothing else |
| How will we know it works? | Written acceptance criteria and a test set of your real questions | “You’ll see it in the demo” |
| Can I test it before the main payment? | A working demo on your data first | Full payment up front |
| Whose accounts are used? | Yours: model, telephony, workflows | The vendor’s, bundled into a monthly fee |
| How is usage billed? | At cost, with a written estimate | “Unlimited,” with no numbers |
| What happens when the agent is unsure? | Hand-off to a person, with a summary | No escalation path |
Compare first-year cost, not the headline price. A cheap build with a vague care fee and marked-up minutes can cost more than a pricier build with usage at cost.
How we would approach it
Every Myrado project is led by a senior engineer based in Ukraine. Hold any vendor, including us, to the checklist above. Here is how we work:
- Brief. We reply within one business hour and send a plan, stack and fixed quote within 24 hours.
- Demo on your data. You test a working prototype before the main invoice.
- Fixed price, written acceptance criteria. We agree on a test set of your real questions. If the agent fails it, we fix it at no cost.
- Milestone payments. The larger part is paid after acceptance.
- Usage at cost. Model and telephony accounts are in your name, and we estimate usage up front.
- 30 days of tuning included. Care is optional after that.
No defined sales or support process yet? We’d start with an audit, not a build. An agent that misbehaves? A rescue audit starts at $400 and is credited toward further work. You’ll find the full process in the How we work section, and you can send a brief from the pricing section.
FAQ
How much does it cost to build an AI agent?
Upwork’s September 2026 hiring guide for AI agent developers puts chatbots and simple assistants at $500–2,000, workflow agents at $2,000–5,000, custom RAG agents at $5,000–15,000 and multi-agent systems at $10,000–30,000. Marketplace packages start far lower, but they usually cover a basic setup and rarely state testing, support days or usage limits. Our fixed-price builds start at $2,000.
How much does an AI voice agent cost per minute?
At the time of our research (September 2026), Vapi listed a $0.05 per-minute platform fee, with other providers billed on top. Retell listed $0.07–0.31 per minute and Bland $0.12–0.14 per minute. Add a phone number and SMS fees, and check each pricing page, because rates change.
How much does an AI receptionist cost?
One done-for-you AI receptionist service we reviewed in September 2026 charged $99–149 a month plus a $1,500 setup fee, with the lower tier capped at 100 calls a month. Marketplace setups start lower but rarely say how many minutes or how much support is included. Our AI receptionist build starts at $2,000, with call minutes billed at cost on your own accounts.
How much does an AI agent cost per month to run?
Expect two monthly costs: usage (model tokens, voice minutes, messaging, hosting) and optional care. The US-dollar agency price lists we reviewed charge $750–2,500 a month for care, and Upwork’s hiring guide for Vapi developers lists $1,500–4,000 a month for ongoing management. Our care plans start at $250 a month, with usage billed at cost on your own accounts.
Why are AI agents on Fiverr so cheap?
Entry packages usually cover one agent set up on a hosted platform with a basic script. None of the voice-agent packages we reviewed in September 2026 said how many call minutes were included, what knowledge-base limits applied or which platform fees were extra, and none offered written acceptance criteria. That can be enough for a low-stakes FAQ widget, but not for a phone line that books real jobs.
How much do AI agent developers charge per hour?
Upwork data compiled by Upwatcher for the 30 days ending May 16, 2026, shows a median of $30 an hour, $40 at the 75th percentile and $60 at the 90th. Upwork’s own hiring guide quotes $30–150 an hour for AI agent developers. One August 2026 industry overview puts experienced engineers with production work on several systems at $100–180+ an hour.