TL;DR
A realistic AI voice agent costs about $0.11 to $0.16 a minute to run on a hosted platform in 2026. Advertised rates of $0.05 to $0.08 cover only one part of the bill.
Every voice minute has the same five parts: the platform fee, speech-to-text, the language model, the voice, and the phone line. The model and the voice cause most of the price spread.
Retell AI pricing starts at $0.07 a minute for its voice layer, and a typical support setup lands near $0.114 a minute once the model and telephony are added.
A custom AI voice agent costs roughly $15,000 to $150,000 to build, depending on integrations and compliance. Lower per-minute costs alone rarely repay that below about 50,000 minutes a month.
Compare cost per resolved call, not cost per minute. A cheap agent that fails a third of its calls costs more than a dearer one that finishes them.
The pricing page says $0.07 a minute. The first invoice says $0.13. Nobody lied. The page showed one line of a bill that has four or five.
That gap is why AI voice agent pricing is so hard to compare. Every platform publishes the number that makes it look cheapest: a platform fee here, a bundled rate there, a monthly plan with minutes included somewhere else. Put three quotes side by side and you are comparing three different things.
This guide puts them on the same footing. It breaks one voice minute into its parts, shows Retell AI pricing line by line, compares the main platforms at the same call volume, and covers what it costs to build a custom agent and when that build pays for itself.
How much does an AI voice agent cost in 2026?
An AI voice agent costs about $0.11 to $0.16 per minute to run on a hosted platform in 2026, with the full published range running from $0.05 to over $0.30. A custom build adds a one-time cost of roughly $15,000 to $150,000. Subscription platforms bundle minutes into monthly plans starting around $49.
There are three ways to pay, and each prices a different thing.
| Pricing model | How you pay | Typical cost | Who it suits |
|---|
| Subscription platform | Monthly plan with minutes included, overage per minute | From about $49 a month; overage around $0.10 to $0.15 a minute | Small teams that want a working agent without engineers |
| Pay-as-you-go platform | Per minute, often split into separate components | About $0.11 to $0.16 a minute all-in for typical setups | Developers who want to choose each part of the stack |
| Custom build | One-time build cost plus your own per-minute stack | Roughly $15,000 to $150,000 to build, then usage | Businesses with complex integrations, compliance needs or very high volume |
The rest of this guide covers each one, starting with what is inside a minute.
What goes into the price of a minute of an AI voice agent?
Every AI voice agent minute is built from five parts. Some platforms bill each one separately; others bundle them into one rate. Either way, you pay for all five.
| Component | What it does | Typical cost per minute | What moves the price |
|---|
| Platform or orchestration fee | Manages turn-taking, interruptions and the call flow | About $0.05 to $0.08 | The platform you choose |
| Speech-to-text | Turns the caller's words into text | About $0.01 | Accuracy needs and language |
| Language model | Decides what the agent says next | About $0.003 to $0.16 | Model size and prompt length |
| Text-to-speech | Turns the reply into a voice | About $0.015 to $0.04 | Voice quality and provider |
| Telephony | Connects the agent to the phone network | About $0.015 in the US, or $0 with your own carrier | Country and carrier |
Two of these lines do most of the work. The language model ranges more than fifty-fold depending on which one you pick, and premium voices cost two to three times the cheapest options. Everything else moves by a cent or two.
There is also a sixth design choice: speech-to-speech models, such as OpenAI's Realtime API, that replace the separate speech-to-text, model and voice steps with one model. OpenAI bills these by audio token rather than by minute. Working from OpenAI's July 2026 rates, analyses put a typical agent at about $0.06 to $0.11 a minute on the flagship model and $0.02 to $0.05 on the smaller one, before telephony and orchestration.
Retell AI pricing explained
Retell AI pricing is pay-as-you-go with no platform fee, and each minute is billed as separate parts. Retell publishes a voice agent range of $0.07 to $0.31 per minute.
Here is how a Retell minute builds up:
Voice infrastructure: $0.055 a minute. This is charged on every call, whatever else you choose.
Voice: from $0.015 a minute. Retell's own voices and Cartesia sit at the bottom; premium voices cost more. Voice infrastructure plus the cheapest voice is the $0.07 headline.
Language model: from a fraction of a cent to well over $0.10 a minute. Retell lists per-minute rates for each model. GPT-5 mini sits near $0.012; Claude Sonnet sits near $0.08. Bring your own model and this line drops to zero.
Telephony: $0.015 a minute on a Retell-managed US number, or nothing at Retell if you bring your own SIP trunk. Phone numbers cost $2 a month each.
A realistic support setup, with GPT-5.4 mini, Retell's own voice, US telephony and a knowledge base, comes to about $0.114 a minute. At 10,000 minutes a month, that is roughly $1,144.
Retell also charges for some extras. The first 20 concurrent calls are included and each extra slot costs $8 a month. Its AI quality review adds $0.10 a minute after 100 free minutes. New accounts get $10 of credit.
The trade-off is control. Retell suits teams who want to pick each component and can track a bill with several lines. Teams that want one number on the invoice usually prefer a bundled rate.
AI voice agent pricing compared: Retell, Vapi, Bland, ElevenLabs and Dialora
The main platforms price in three different ways: unbundled per-component billing, a single blended rate, or a monthly plan with minutes included.
| Platform | Pricing model | Advertised rate | Typical all-in per minute |
|---|
| Retell AI | Per minute, billed by component | From $0.07 | About $0.11 to $0.15 |
| Vapi | $0.05 platform fee, providers at cost | $0.05 | About $0.13 to $0.16 |
| Bland AI | Blended rate by plan tier | $0.14 (Start), $0.12 (Build), $0.11 (Scale) | $0.11 to $0.14, plus $299 or $499 a month on paid plans |
| ElevenLabs Agents | Per minute, model and telephony extra | $0.08 | About $0.10 to $0.14 |
| Dialora | Monthly plan with minutes, all-inclusive overage | From $49 a month | $0.10 to $0.15 per overage minute |
| Synthflow | Enterprise contracts | From $30,000 a year | Scoped per contract |
A few details change the comparison:
Vapi's saving comes from your own keys. If you already hold contracts with a model or speech provider, Vapi charges nothing on those lines. Vapi is also the only major platform with a published HIPAA price, at $2,000 a month.
Bland includes speech, model and voice in its rate, but bills call transfers at $0.05 a minute and dropped its old $0.09 rate in December 2025.
ElevenLabs charges double, $0.16 a minute, for calls above your concurrency limit. Busy hours can cost more than the headline suggests.
Synthflow no longer sells self-serve plans, so it fits large contracts rather than first projects.
What 10,000 minutes a month costs on each platform
This illustrative comparison uses each platform's published pricing for a typical inbound support setup. Your numbers will move with model and voice choices.
| Platform | Estimated monthly cost at 10,000 minutes | How it is calculated |
|---|
| Retell AI | About $1,144 | $0.114 a minute support setup |
| Dialora (Agency plan) | About $1,197 | $297 plan with 1,000 minutes, plus 9,000 at $0.10 |
| Vapi | About $1,350 | Typical hosted stack |
| ElevenLabs Agents | About $1,400 | Scale plan, overage, model and telephony |
| Bland AI (Build plan) | About $1,499 | $299 plan plus $0.12 a minute |
At this volume, the spread between the cheapest and dearest option is about $350 a month. That is smaller than most buyers expect, and it is the reason the per-minute rate should rarely decide the choice on its own.
Subscription AI voice agent pricing: when a monthly plan makes sense
Subscription pricing bundles minutes, agents and features into a fixed monthly fee, with an all-inclusive rate for minutes beyond the plan. It trades a little per-minute efficiency for a predictable bill and no engineering work.
Dialora, Codiste's AI voice agent platform, uses this model. Its published plans, at the time of writing:
| Dialora plan | Monthly price | Included minutes | AI agents | Extra minutes |
|---|
| Starter | $49 | 200 | 2 | $0.15 |
| Pro | $97 | 400 | 10 | $0.14 |
| Business | $149 | 600 | Unlimited | $0.12 |
| Agency | $297 | 1,000 | Unlimited | $0.10 |
Annual billing saves 20%, and enterprise pricing covers volume-based minute rates, 100 or more concurrent calls, SIP trunk integration and compliance support. The overage rate already includes the model, voice and phone line, so there is no second invoice to reconcile.
A subscription plan makes sense when:
Nobody on the team writes code. Agents are configured in a dashboard and tested live before going on a phone line.
The job is standard. Answering calls, qualifying leads, booking appointments and sending reminders are covered out of the box, including calendar and CRM connections.
You resell to clients. The Agency plan includes white-label and multi-client management, which per-minute platforms leave you to build.
The bill must be predictable. A fixed plan with a known overage rate is easier to budget than four separate component lines.
It makes less sense when the agent must follow complex rules inside your own systems, which is where a custom build comes in.
Most teams compare per-minute rates and stop there. The real question is which of the three models fits your calls, your systems and your volume. Codiste will map your call types against all three, and point you to a Dialora plan where a subscription already does the job.
How much does it cost to build a custom AI voice agent?
A custom AI voice agent costs roughly $15,000 to $35,000 for a single-purpose agent, $35,000 to $80,000 for a production agent handling several call types, and $80,000 to $150,000 or more for an enterprise system with deep integrations and compliance requirements. A build from the ground up, including your own speech infrastructure, can run far higher.
Development firms publish these ranges, and they agree on the main driver: integration, not the voice. Connecting the agent to your CRM, booking system, payment provider or patient records is typically 40 to 60% of the engineering work.
| Build type | Typical cost | Typical time | What it includes |
|---|
| Single-purpose agent | $15,000 to $35,000 | 6 to 8 weeks | One call type, one language, one or two integrations |
| Production multi-intent agent | $35,000 to $80,000 | 2 to 4 months | Several call types, CRM and calendar actions, transfers, analytics |
| Enterprise or regulated system | $80,000 to $150,000+ | 4 months or more | Deep back-office integration, multiple languages, audit trails, compliance controls |
There are really three ways to get a custom agent:
Build on an existing voice platform. A development team writes your call flows, integrations and rules on top of a platform such as Retell, Vapi or LiveKit. This is the most common route, usually four to ten weeks.
Assemble your own stack. Choose speech, model, voice and telephony providers separately and write the orchestration. More control, and lower per-minute costs once built.
Build every layer in-house. Only sensible when voice is your core product, and analyses put year-one cost in the hundreds of thousands to millions of dollars.
Codiste builds custom voice agents around a client's own phone systems, data and approval rules, on whichever stack fits, and hands the system over to the client's team. Our AI agent development services cover scoping, integration, testing and handover.
When does a custom AI voice agent pay for itself?
A custom build rarely pays back through lower per-minute costs alone unless volume is high. It pays back through what a platform cannot do.
Here is an illustrative calculation. A hosted platform costs $0.13 a minute. A custom stack costs $0.07 a minute. That saves $0.06 a minute. A $60,000 build therefore needs about one million minutes to repay itself on per-minute savings alone. At 50,000 minutes a month, that is about 20 months. At 10,000 minutes a month, it is more than eight years.
So the honest case for a custom build is rarely the per-minute rate. It is:
Integration depth. The agent can look up an order, change a booking or take a payment inside your own systems, instead of only collecting details for a person to act on.
Rules a platform cannot express. Regulated scripts, consent checks, verification steps and escalation paths specific to your business.
Ownership. No platform margin, no lock-in, and the ability to change any part of the stack as prices fall.
Outcome value. If the agent resolves calls that previously needed a person, the saving is measured in staff hours, not cents per minute.
If none of these apply, a platform or subscription is the cheaper choice, and we say so in first calls.
Hidden costs of AI voice agents
The costs that surprise buyers are rarely on the pricing page. Watch for these:
Concurrency limits. Retell includes 20 simultaneous calls and charges $8 a month per extra slot. ElevenLabs bills calls above your limit at $0.16 a minute. Plan for your busiest hour, not your average.
Transfers and silence. Bland bills call transfers at $0.05 a minute. Flat per-minute pricing also bills the silence while a caller looks for an account number.
Phone numbers and compliance. Numbers cost around $2 a month each, and outbound calling in the US involves registration, caller ID and consent rules. Healthcare work may need paid compliance tiers, such as Vapi's $2,000 a month HIPAA price.
Premium model or voice upgrades. Swapping a small model for a frontier one can add $0.05 to $0.10 a minute overnight.
Quality review and testing. Automated call review, evaluation and re-testing after model updates cost money and engineering time.
Maintenance. Prompts drift, integrations change and models get replaced. Budget ongoing time to keep the agent working, whichever route you choose.
Why cost per minute is the wrong number to compare
Cost per minute tells you what a minute costs, not what a finished call costs. An agent at $0.08 a minute that fails 30% of its calls is more expensive per resolved call than one at $0.12 that finishes them, because every failure goes to a person anyway.
Three numbers matter more than the headline rate:
Resolution rate. The share of calls the agent completes without a person. This depends on your call types and integrations more than on the platform.
Average call length. A slower agent, or one that repeats itself, uses more minutes for the same outcome.
Cost of a failed call. A missed booking, a lost lead or a frustrated customer who calls back.
Divide total monthly cost by the number of calls resolved. That figure, not the per-minute rate, is the one to compare between options. For a fuller view of how voice agents affect handling time and staffing, see our guide to AI call center automation.
How to estimate your AI voice agent budget
Five steps give you a realistic monthly number before you talk to any vendor:
Count your minutes. Take monthly call volume and multiply by average call length. Add 10 to 20% for retries and longer AI calls during tuning.
Pick a stack tier. Budget about $0.10 a minute for a lean setup, $0.13 for a typical one, and $0.20 or more for premium voices and frontier models.
Size your busiest hour. Check how many calls happen at once at peak, and price the concurrency you need.
Add the fixed costs. Phone numbers, plan fees, compliance tiers and quality review.
Add build and upkeep. Zero for a subscription platform, a one-time build for custom work, and ongoing maintenance either way.
Then compare that total against what the same calls cost today in staff time, missed calls and callbacks.
Which AI voice agent option should you choose?
Match the option to your situation, in this order:
Codiste: for custom AI voice agents that must act inside your own systems, follow regulated or business-specific rules, or run at a volume where ownership matters. Codiste scopes, builds and hands over the agent, and starts by checking whether a product you could buy already does the job.
Dialora: for businesses that want a working voice agent this week without engineers, with predictable monthly pricing, booking, CRM connections and white-label options for agencies.
Retell AI or Vapi: for developer teams that want to choose every component and are comfortable tracking a multi-line bill.
Bland AI: for teams that want one blended rate and can commit to a monthly plan at volume.
ElevenLabs Agents: for use cases where voice quality and language range matter most.
OpenAI Realtime API: for engineering teams building their own speech-to-speech agent and willing to manage token-based billing.
If you are deciding whether to build in-house or bring in an outside team, our agency vs in-house AI development guide walks through the trade-offs.
Final verdict
AI voice agent pricing looks chaotic because vendors publish different slices of the same bill. Once every option is reduced to an all-in cost per minute, the field narrows quickly. Typical hosted setups land between $0.11 and $0.16 a minute in 2026, and at 10,000 minutes a month the gap between the main platforms is a few hundred dollars.
That makes the per-minute rate a poor deciding factor. The choices that move your real cost are elsewhere: which model and voice you run, how many calls happen at once, whether the agent can act in your systems, and how many calls it finishes without a person.
Start with the simplest option that can do the job. If your calls are standard and nobody on the team writes code, a subscription platform like Dialora will be cheaper and faster than anything custom. If you have engineers and want control over each component, a pay-as-you-go platform fits. Move to a custom build when the agent must work inside your own systems, follow rules no platform can express, or handle volume large enough that platform margins add up.
Whichever route you take, measure cost per resolved call from the first week. It is the only number that tells you whether the agent is saving money or moving work around. Price the minutes, then price the outcomes, and let the second number make the decision.
If your calls involve a CRM, booking system, or regulated workflow, explore Codiste’s AI Voice Agent Development Services to find the right solution for your business requirements and integration needs.
FAQs
How much does an AI voice agent cost per minute?
+
On hosted platforms, a typical AI voice agent costs about $0.11 to $0.16 per minute in 2026 once the platform fee, model, voice and telephony are added. Advertised rates of $0.05 to $0.08 cover only part of that. Premium models and voices can push the cost above $0.30 a minute.
How much does Retell AI cost?
+
Retell AI charges $0.07 to $0.31 per minute for voice agents, with no monthly platform fee. The $0.07 covers its voice infrastructure and standard voice. A typical support setup with a mid-sized model and US telephony costs about $0.114 a minute, or roughly $1,144 for 10,000 minutes.
What is the cheapest AI voice agent platform?
+
On headline rate, Vapi's $0.05 platform fee is lowest, but its all-in cost lands near $0.13 to $0.16 with hosted providers. At 10,000 minutes a month, Retell, Dialora's Agency plan, Vapi, ElevenLabs and Bland fall within about $350 of each other. The cheapest option depends on your model, voice and volume.
How much does it cost to build a custom AI voice agent?
+
A single-purpose custom voice agent typically costs $15,000 to $35,000. A production agent with several call types and CRM integration runs $35,000 to $80,000. Enterprise or regulated systems cost $80,000 to $150,000 or more. Integration with your existing systems is usually the largest share of the work.
Is it cheaper to build or buy an AI voice agent?
+
Buying is cheaper for most businesses. A custom build saves a few cents per minute, so it rarely repays the build cost on usage alone below about 50,000 minutes a month. Build when the agent must act inside your systems, follow rules a platform cannot handle, or meet strict compliance needs.
Are there hidden costs with AI voice agents?
+
Yes. Common extras include concurrency limits and burst pricing, call transfer fees, phone number rental, compliance tiers such as HIPAA, quality review add-ons, and ongoing maintenance. Premium model or voice upgrades can also raise the per-minute cost sharply. Size costs for your busiest hour, not your average.
