It’s a quarter to six on a Friday. Your front desk has gone home, and the phone starts ringing. Nobody picks up. The caller leaves a voicemail, or hangs up and tries someone else.
An AI voice agent can answer that call. It can answer questions, book a time, or pass the caller to a person. Some work out of the box. Others are building blocks that a developer wires together.
That split matters more than any feature list. This guide covers the build-your-own kind: custom AI voice agent platforms. They suit teams with a developer, an agency partner, or the patience to configure a builder.
Want a ready-made receptionist with no build? That is a different product. We point to it near the end.
We compared eight platforms on published capabilities, pricing and security documentation. We placed no test calls. Cost and review ratings are not in the scores yet, so treat the order as a starting point.
Start with the quick picks. Then use the table to narrow the list to two.
Our top picks
- Vapi: best for developer teams. $0 to start, then $0.05 a minute for hosting, with model and carrier costs on top.
- ElevenAgents: best for self-serve builds with broad integrations. Free plan with 15 minutes, then about $0.08 a minute.
- Retell AI: best for a fast self-serve start. Pay as you go starts at $0 with a $10 credit. Rates run $0.07 to $0.31 a minute.
- LiveKit Agents: best when you want to own the stack. Open source (Apache 2.0). The hosted Build plan is $0 and Ship is $50 a month.
- Synthflow: best for an enterprise contract. Custom quote on an annual agreement.
- echowin: best for small teams and agencies. Prepaid bundles from $45 a month for 250 minutes.
Prices are for the entry plan, in US dollars, checked November 13, 2026. Per-minute rates are not total costs. This is a research-based ranking: we read documentation, prices and published reviews, and placed no test calls. How we scored them.
Quick comparison table
| Provider | Best for | Entry price and billing | Usage rate | Key capability | Main trade-off | Our rating |
|---|---|---|---|---|---|---|
| 1. Vapi | Developer teams that want full control | $0 to start (4 concurrent calls); Core $29/month | $0.05/min hosting, plus model and carrier costs at cost | API-first platform; Vapi’s own SIP and WebSocket transport is free | Model and carrier costs add to the bill; HIPAA add-on is $2,000/month | 63.0 / 100 |
| 2. ElevenAgents | Self-serve builds with broad integrations | Free ($0, 15 min); Starter $6/month (75 min) | $0.08/min after included minutes | Dashboard plus API and SDKs; SIP and Twilio numbers | LLM and telephony are billed separately | 61.0 / 100 |
| 3. Retell AI | A fast self-serve start | Pay as you go $0 (with $10 credit) | $0.07 to $0.31/min, depending on voice, model and telephony | No-code builder plus API; 20 concurrent calls free | Fees are itemized; its pricing page lists the verified-number fee two ways | 58.0 / 100 |
| 4. LiveKit Agents / Cloud | Teams that want an open-source framework | Build $0 (1,000 agent minutes); Ship $50/month | $0.01/min after included minutes (Ship and Scale plans) | Open-source framework with SIP and US numbers | Built for developers; setup and support was its weakest area in our research | 55.0 / 100 |
| 5. Synthflow | A contracted, no-code enterprise rollout | Custom quote; annual enterprise agreement | Not published | No-code editor with SIP telephony | No free plan or trial; no public prices | 52.0 / 100 |
| 6. Telnyx Voice AI Agents | Carrier and AI on one network | Pay as you go $0; Growth from $2,000/month | $0.05/min voice engine | Carrier-owned network with numbers and SIP | LLM and telephony are extra; voice engine rounds to 60-second increments | 51.0 / 100 |
| 7. Bland AI | One per-minute rate that includes LLM and voice | Start $0 platform fee; Build $299/month | $0.14/min (Start), $0.12/min (Build) | LLM, speech-to-text and text-to-speech included in the rate | Telephony billed separately; BAA, SSO and data residency are Enterprise only | 50.0 / 100 |
| 8. echowin | Small teams and agencies that want published bundles | Pay as you go $0; bundles from $45/month (250 min) | $0.20/min pay as you go; 14 to 18 cents/min in bundles | No-code builder plus REST API; agency plans | Phone numbers cost $10/month each; LLM and telephony inclusion is not stated | 43.0 / 100 |
Table note: Prices were checked on each provider’s pricing page on November 13, 2026, and exclude taxes. Per-minute rates are not total costs. Scores are points out of 100 on the evidence we could confirm. Gaps of 1 to 3 points are small. Do not compare these scores with scores from other guides.
What is an AI voice agent?
An AI voice agent answers or places phone calls and holds a spoken conversation. It listens, works out what the caller wants, and replies in a synthetic voice.
Most can also act: check a calendar, look up an order, transfer to a person or send a text.

Under the hood, most platforms chain three parts. Speech-to-text turns the caller’s words into text. A language model decides what to say or do. Text-to-speech speaks the answer. A phone connection (a number or a SIP line) carries the call.
This matters for buying. Some vendors bill each part separately. Others bundle them in one rate. Our table and reviews show which is which.
The platforms in this article are build-your-own. You configure an agent, connect a number and tune the prompts. Ready-made receptionists are a different product. We cover them near the end of this guide.
Is an AI voice agent worth it?
It depends on call volume, call type and who sets it up. A few facts help you size it.
The U.S. Bureau of Labor Statistics reports median receptionist pay of $38,010 a year in May 2025. That is $18.27 an hour, or about 30 cents per paid minute, before benefits and idle time. Source: BLS Occupational Outlook Handbook.
Do not compare that directly with the per-minute prices in this article. A person is paid when the phone is quiet. A voice agent bills mostly when a call is live. A person also handles walk-ins, mail and other work.
An agent tends to fit when:
- Calls are frequent, repetitive and easy to script.
- You miss calls after hours or at peak times.
- Someone on your team can build and review the agent.
A person tends to fit better when calls are rare, sensitive or complex. A mixed setup also works: the agent answers first and hands off to a person.
How we picked
We looked at platforms you build on. You or a partner set up the agent. Each platform had to say how it is deployed, and its AI had to handle business phone calls.
We rated eight platforms from their published documentation, pricing pages and security pages. The rating has 23 checks across seven areas. Core capabilities, pricing and integrations carry the most weight.
We did not place test calls. Cost and review points are not in the ratings yet, so each rating rests on roughly 70 of the 100 possible points. The full methodology shows the weights and the gaps.

The 8 best AI voice agents for business phone calls
Listed in rank order, highest rating first.
1. Vapi: best for developer teams that want full control
Our verdict: Pick Vapi if you have a developer and want to choose your own speech and AI models. Expect a bill with several parts.

| Plan we priced | $0 to start (4 concurrent calls); Core $29 a month |
| Type | Developer platform (you build the agent) |
| Reviews | G2 3.0 (3); Trustpilot 2.5 (13); Gartner 5.0 (1) |
| Our rating | 63.0 / 100 |

What stands out:
- Built for developers. Vapi calls itself a developer platform, and it is API-first. You can sign up on your own and start at $0, so you can build before you talk to sales.
- You choose the parts. Speech-to-text, the language model and text-to-speech are passed through at cost. Vapi’s pricing page prices Deepgram, OpenAI and ElevenLabs components as examples.
- Flexible phone options. Vapi’s own telephony, SIP and WebSocket transport is free. You can also connect Twilio, Telnyx or Vonage, which charge their own carrier rates.
- Strong security documentation. Security was one of its best-documented areas. We read what Vapi publishes. We did not audit its security.
Pricing. Hosting is one part of a Vapi bill. You also pay for speech-to-text, the language model, text-to-speech and the carrier, each at its own rate.
Vapi’s pricing page lists speech-to-text near $0.0095 to $0.0099 a minute, the language model at $0.0077 to $0.0452, and text-to-speech at $0.0146 to $0.0238. Twilio charges $0.008 a minute inbound. Telnyx charges $0.0055.
Extra concurrent lines cost $10 each per month. The HIPAA add-on costs $2,000 a month. Entry plans are in the comparison table, and totals are in the cost section.

What customers say. Vapi’s ratings are mixed and the samples are small. G2 shows 3.0 stars from 3 reviews. Trustpilot shows 2.5 from 13. Gartner Peer Insights shows 5.0 from 1.
Reviewers like the setup, the developer tools and the customization. They raise cost, speech limits and support. Test it in the free plan before you commit.
Watch out for:
- The headline rate is only part of the bill. Model, speech and carrier fees vary with the providers you pick.
- HIPAA costs extra. The add-on is $2,000 a month.
- Concurrency is capped by plan. Start allows 4 calls at once, Core 10 and Pro 30. More lines cost $10 each per month.
Why Vapi ranks #1. Vapi leads at 63.0 / 100. It is strongest on integrations and security documentation, and weakest on core capabilities.
ElevenAgents is two points behind with the same integrations strength, so the lead is narrow. Cost and review points are not in the rating yet, so the order could move.
Choose Vapi if you have a developer or an agency partner. You want to pick your own speech and AI models, and you can manage a bill with several parts.
Skip Vapi if you want a no-code setup. Retell AI offers a no-code builder as well as an API.
Skip it too if you want one per-minute rate that already includes the language model and voice. Bland AI includes both in its rate.
2. ElevenAgents: best for self-serve builds with broad integrations
Our verdict: Pick ElevenAgents if you want to start free, build in a dashboard and connect your own business tools. Plan for language model and phone costs on top.

| Plan we priced | Free ($0, 15 minutes); Starter $6 a month (75 minutes) |
| Type | Self-serve agent platform (dashboard, API and SDKs) |
| Reviews | No agent-specific listing found |
| Our rating | 61.0 / 100 |
Prices checked November 13, 2026. Product name on the provider’s site: ElevenAgents (ElevenLabs Agents Platform).

What stands out:
- Start free. You can sign up on your own and build in a dashboard. An API and SDKs are there when you need them.
- Phone numbers via SIP or Twilio. ElevenAgents supports both routes to the phone network. Telephony is billed at cost.
- Integrations are its strongest area. It ties with Vapi at the top of the list. We read the integration documentation. We did not test any connection.
- Room to grow. Concurrent calls rise from 4 on the free plan to 40 on Business. Enterprise adds higher limits on request.
Pricing. Each paid plan bundles minutes at about the same per-minute rate. Higher plans add minutes and concurrent calls. They do not lower the rate. ElevenAgents’ pricing page also lists a burst rate of $0.16 a minute, double the standard rate.
The language model costs extra and varies by model. Telephony is billed at cost. Text messages cost $0.003 each. Prices exclude taxes. Totals are in the cost section.

What customers say. We found no review listing for ElevenAgents alone. The company-wide ElevenLabs pages cover all its products, so we do not use them here. Run a few test calls on the free plan.
Watch out for:
- Two costs sit outside the plan. The language model and the phone line are billed separately.
- The free plan is small. It includes 15 minutes, which is enough for a quick test only.
- Setup and support is its weakest area. Retell AI rates higher here.
Why ElevenAgents ranks #2. ElevenAgents sits second at 61.0 / 100. Integrations and audience fit are its best areas, and setup and support is its weakest. Its integrations are well ahead of Retell AI’s, but Retell rates higher on setup and support. Vapi leads by two points.
Choose ElevenAgents if you want to build in a dashboard and start at $0. You need to connect the agent to your own business tools. You can also handle language model and phone costs on top.
Skip ElevenAgents if you want to host the stack yourself. LiveKit’s framework is open source under Apache 2.0.
Skip it too if you want one rate that includes the language model and voice. Bland AI includes both.
3. Retell AI: best for a fast self-serve start with strong security documentation
Our verdict: Pick Retell AI if you want to sign up, test with free credit and scale on usage pricing. Expect to add language model and phone costs yourself.

| Plan we priced | Pay as you go, $0 with $10 credit |
| Type | Self-serve agent platform (no-code builder, API and SDKs) |
| Concurrency | 20 free concurrent calls, then $8 per extra call per month; Enterprise 50+ |
| Reviews | G2 4.8 (2,653); Trustpilot 4.9 (805); Gartner 4.7 (6); Capterra 4.4 (7) |
| Our rating | 58.0 / 100 |

What stands out:
- No plan to buy first. You start on pay as you go with a $10 credit. There is no monthly fee.
- Security is its strongest area. It has the most complete published security documentation of the eight. We read the documents. We did not audit anything.
- Support and setup rate well. It rates above ElevenAgents here.
- Generous concurrency. The first 20 concurrent calls are free. Enterprise starts at 50 or more.
Pricing. Retell builds its rate from parts. On Retell’s pricing page, voice infrastructure is $0.055 a minute and text-to-speech starts at $0.015. ElevenLabs voices cost $0.04 to $0.10.
The language model is billed on top. Telephony is $0.015 a minute through Twilio or Telnyx, and $0 if you bring your own SIP line.
Add-ons are small. The knowledge base, noise reduction and guardrails each add $0.005 a minute. Phone numbers are $2 a month and SMS is $20 a month. Totals are in the cost section.

What customers say. Retell AI has the largest review base. G2 shows 4.8 stars from 2,653 reviews, and some are marked as incentivized. Trustpilot shows 4.9 from 805. Gartner shows 4.7 from 6. Capterra shows 4.4 from 7.
On G2, reviewers most often praise ease of use (495 reviews). The most common complaints are missing features (116) and cost (87). Run a few test calls on the free credit.
Watch out for:
- The price is a range. Your language model choice can move the cost a lot.
- Silence and hold time are billed. Check how that applies to your call flow.
- Integrations are its weakest area. ElevenAgents is well ahead on business-tool integrations.
Why Retell AI ranks #3. Retell AI is third at 58.0 / 100. Security documentation and setup and support are its best areas, and integrations are its weakest.
ElevenAgents is three points ahead, mostly on integrations. LiveKit is three points behind, and Retell rates higher on setup and support.
Choose Retell AI if you want to start on credit with no plan fee. You care about published security documentation. You can handle language model and phone costs on top.
Skip Retell AI if you need the widest set of business-tool integrations. ElevenAgents rates higher there.
Skip it too if you want one rate that includes the language model and voice. Bland AI includes both.
4. LiveKit Agents: best for an open-source framework you can host
Our verdict: Pick LiveKit if your team writes code and wants to own the stack. Skip it if you want a dashboard and hands-on support.

| Plan we priced | Build $0 (1,000 agent minutes); Ship $50 a month |
| Type | Open-source agent framework (Apache 2.0) plus a hosted cloud |
| Reviews | G2 4.8 (2), a mixed LiveKit listing |
| Our rating | 55.0 / 100 |
Prices checked November 13, 2026. Product name on the provider’s site: LiveKit Agents / Cloud.

What stands out:
- You can host it yourself. The framework is open source under Apache 2.0. The hosted cloud is optional.
- Security and fit rate high. Its security documentation is strong, and it suits technical buyers. We read the documents.
- Plans scale to large concurrency. Scale supports up to 600 concurrent sessions.
- A free tier to test. Build includes 1,000 agent minutes and 5 concurrent sessions.
Pricing. Plans include agent minutes and charge past them. On LiveKit’s pricing page, sessions cost $0.01 a minute over the included amount on Ship and Scale. The Build plan lists no overage rate. Phone (SIP) minutes and recordings are billed after the included amounts.
LiveKit’s own calculator shows an example of about $0.0479 a minute. That is an example on its page, not a quote for your calls. Language model and voice providers are separate costs. Totals are in the cost section.

What customers say. LiveKit has very little review data. G2 shows 4.8 stars from 2 reviews, on a listing that also covers LiveKit’s wider audio and video products.
Two reviews is too few to judge. We found no Trustpilot page. Build a small test agent on the free plan.
Watch out for:
- It is a framework first. You or a developer will write and run code.
- Setup and support is its weakest area. Retell AI rates higher here.
- Providers add cost. Language model and voice services are billed on top.
Why LiveKit ranks #4. LiveKit is fourth at 55.0 / 100. Security documentation and technical-buyer fit are its best areas, and setup and support is its weakest. Retell AI is three points ahead and rates higher on setup and support. Synthflow is three points behind.
Choose LiveKit if you have developers and want to host or customize the stack. You want open-source code you can inspect.
Skip LiveKit if you want a no-code dashboard. Retell AI and ElevenAgents are easier starts.
Skip it too if you want a vendor to run the rollout. Synthflow sells that.
5. Synthflow: best for a contracted, no-code enterprise rollout
Our verdict: Pick Synthflow if you want a vendor-led enterprise deployment with contract terms. Skip it if you want public prices or a free trial.

| Plan we priced | Custom quote; annual enterprise agreement |
| Type | No-code platform sold on an annual enterprise agreement |
| Contract support | MSA and DPA support; security review |
| Time to launch | “Weeks,” per the provider |
| Reviews | G2 4.5 (1,016); Trustpilot 4.1 (215); Gartner 4.3 (4) |
| Our rating | 52.0 / 100 |
Checked November 13, 2026. We rated the Enterprise AI Receptionist use case.

What stands out:
- Core capabilities are its strongest area. That covers conversation design, phone functions and tools. We read the documentation. We did not test calls.
- Setup and support rate well. It matches the best on this list.
- Contract paperwork is part of the offer. Synthflow lists MSA and DPA support and a security review.
- Built for a managed rollout. The provider describes going live in weeks.
Pricing. Synthflow does not publish plans or rates on its pricing page. It is sold as an annual enterprise agreement, with no free plan and no trial.
That makes it hard to compare on cost. Ask for a quote and set it against the published rates for the others in the cost section.

What customers say. Synthflow has a large G2 base but lower scores elsewhere. G2 shows 4.5 stars from 1,016 reviews. Trustpilot shows 4.1 from 215. Gartner shows 4.3 from 4.
On G2, reviewers praise ease of use (364 reviews). The most common complaint is cost (145). Ask the vendor for references in your industry.
Watch out for:
- No public price. You cannot size the cost before a sales call.
- No trial. You cannot test before you sign.
- Four checks could not be verified. They are cost, review count, star rating and recording consent controls.
Why Synthflow ranks #5. Synthflow is fifth at 52.0 / 100. Core capabilities and setup and support are its best areas. The unchecked items held its rating down.
LiveKit is three points ahead and Telnyx is one point behind, so a small gap here can close once those are confirmed.
Choose Synthflow if you want a no-code build, contract terms and a vendor-led launch. You are comfortable with an annual agreement.
Skip Synthflow if you want to start on a card and test today. Retell AI and ElevenAgents let you do that.
Skip it too if you need public prices to budget.
6. Telnyx Voice AI Agents: best for carrier and AI on one network
Our verdict: Pick Telnyx if you want phone service and the voice agent from one provider. Plan for language model costs and a 60-second billing floor.

| Plan we priced | Pay as you go $0; Growth from $2,000 a month |
| Type | Voice AI agents run by a telecom carrier |
| Billing note | Calls round up to 60 seconds |
| Reviews | G2 4.8 (6) |
| Our rating | 51.0 / 100 |

What stands out:
- Phone service and AI in one place. Telnyx runs the network the calls travel on.
- Setup and support rate well. It also fits technical buyers.
- Low telephony rates. Telephony starts at $0.0032 a minute.
- A clear path up. Pay as you go is $0. Discounts start at the Growth minimum.
Pricing. The voice engine, telephony and language model are priced separately on Telnyx’s pricing page. Telephony starts at $0.0032 a minute, and the language model is billed per token. One example, Kimi, runs about $0.006 a minute.
Calls round up to 60-second increments, so short calls cost more. Growth and Enterprise carry monthly minimum spends in return for a discount. Totals are in the cost section.

What customers say. Only G2 has a page for Telnyx Voice AI Agents. It shows 4.8 stars from 6 reviews, which is a small sample. Telnyx’s other review pages cover its phone network product, so we do not use them. Test with a few real calls.
Watch out for:
- Short calls cost more. Each call rounds up to 60 seconds.
- Language model cost is extra. It varies by model.
- Integrations are its weakest area. ElevenAgents is far ahead on ready-made business integrations.
Why Telnyx ranks #6. Telnyx is sixth at 51.0 / 100. Setup and support and technical-buyer fit are its best areas, and integrations are its weakest. Synthflow is one point ahead and Bland AI is one point behind, so treat these three as close.
Choose Telnyx if you want carrier and AI from one vendor. You have volume that can reach the discount tiers.
Skip Telnyx if you need many ready-made business integrations. ElevenAgents rates far higher there.
Skip it too if most of your calls are very short.
7. Bland AI: best for LLM, speech and voice in one per-minute rate
Our verdict: Pick Bland AI if you want one per-minute price that already includes the language model and voice. Telephony is billed separately.

| Type | Voice agent platform with bundled per-minute pricing |
| Start plan | $0 platform fee; $0.14/min talk; $0.05 transfer |
| Build plan | $299/month; $0.12/min talk; $0.04 transfer |
| Included | Language model, speech-to-text and text-to-speech |
| Billed separately | Telephony (Start includes an inbound number) |
| Enterprise only | BAA, SSO, data residency, on-prem or VPC |
| Reviews | G2 5.0 (11); Trustpilot 2.9 (2) |
| Our rating | 50.0 / 100 |

What stands out:
- One rate covers the main parts. The language model, speech-to-text and text-to-speech are included.
- Setup and support rate well. Its security documentation is solid too.
- No fee on the Start plan. You pay per minute only.
- Enterprise controls are listed. They include single sign-on, data residency and on-prem or VPC options.
Pricing. Talk time includes the language model, speech-to-text and text-to-speech. Transfers are billed separately, at a lower rate on Build. Telephony is billed separately too, and Start includes an inbound number. Rates are on Bland’s pricing page.
The break-even between Start and Build is in the cost section.

What customers say. Bland AI has few reviews and they disagree. G2 shows 5.0 stars from 11 reviews. Trustpilot shows 2.9 from 2. Both samples are small. Run a few test calls on the Start plan.
Watch out for:
- Some controls need Enterprise. A BAA, SSO, data residency and on-prem or VPC options are not on the lower plans.
- Telephony is extra. The per-minute rate does not cover the phone line.
- Integrations are its weakest area. ElevenAgents is well ahead on ready-made business integrations.
Why Bland AI ranks #7. Bland AI is seventh at 50.0 / 100. Setup and support and security documentation are its best areas, and integrations are its weakest. Telnyx is one point ahead. echowin is seven points behind and has more unchecked items.
Choose Bland AI if you want a bundled per-minute price and a $0 start. You can add telephony yourself.
Skip Bland AI if you need a BAA or SSO below Enterprise.
Skip it too if you want a free tier to test. Retell AI and ElevenAgents offer one.
8. echowin: best for small teams and agencies that want published bundles
Our verdict: Pick echowin if you want prepaid minute bundles and white-label options. Skip it if you need verified security terms or unused minutes to roll over.

| Type | Self-serve voice agent platform with agency options |
| Pay as you go | $0.20/min |
| Bundles | $45 for 250 min (18¢); $80 for 500 (16¢); $150 for 1,000 (15¢); $350 for 2,500 (14¢) |
| Agency plans | Grey Label $149.99/month; White Label $499.99/month (limited-time prices) |
| Extras | Numbers $10/month; chat $0.03; SMS $0.02 per part |
| Reviews | None yet on G2 or Trustpilot |
| Our rating | 43.0 / 100 |

What stands out:
- Prices are easy to read. Pay as you go and four bundles are all published.
- Setup and support rate well. It also suits small teams.
- Agency options exist. White-label and grey-label plans suit resellers.
- Volume pricing goes lower. The page lists rates as low as $0.12 a minute.
Pricing. Bundles lower the rate as they grow. Minutes reset each month and do not roll over. On echowin’s pricing page, phone numbers are $10 a month, chat costs $0.03 and SMS costs $0.02 per part. Transferred portions of a call are billed at half rate.
Agency plans (Grey Label and White Label) were limited-time offers when we checked. The bundle break-even is in the cost section.

What customers say. We found no reviews for echowin on G2 or Trustpilot. A missing rating does not mean poor quality. Ask the vendor for references in your industry.
Watch out for:
- Minutes do not roll over. Unused bundle minutes reset.
- Agency prices are promotional. They may rise when the offer ends.
- Five checks could not be verified. They include cost, reviews, recording consent controls and security or contract assurance.
Why echowin ranks #8. echowin is eighth at 43.0 / 100. Setup and support and audience fit are its best areas.
Five checks could not be verified, the most of any provider, so part of the gap to Bland AI is missing evidence, not poor results. Bland AI is seven points ahead.
Choose echowin if you want clear bundles and a low entry price. You may resell to clients.
Skip echowin if you need verified security and contract terms. We could not verify its security and contract assurance, so it earned no points there. Retell AI and LiveKit documented theirs.
Skip it too if your volume swings month to month.
What the per-minute price leaves out
A platform’s headline rate is rarely the whole bill. Most calls pass through several paid parts. The table shows what each provider’s published rate does and does not cover, as of November 13, 2026.
| Provider | What the published rate covers | Billed on top |
|---|---|---|
| Vapi | Hosting at $0.05/min | Speech-to-text, language model, voice and telephony; HIPAA add-on $2,000/month |
| ElevenAgents | $0.08/min after included minutes | Language model and telephony |
| Retell AI | Voice infrastructure at $0.055/min | Text-to-speech, language model, telephony, add-ons |
| LiveKit | Session minutes by plan | Language model and voice providers; SIP and recordings after included amounts |
| Synthflow | Not published | Ask for a quote |
| Telnyx | Voice engine at $0.05/min | Language model (per token); telephony from $0.0032/min; 60-second call rounding |
| Bland AI | $0.14/min (Start) or $0.12/min (Build) including language model, speech-to-text and text-to-speech | Telephony; transfers; BAA and SSO only on Enterprise |
| echowin | $0.20/min or a prepaid bundle | Numbers $10/month; SMS and chat; half-rate transfer portions |

Two more costs apply to almost everyone. Phone numbers carry monthly fees. Short calls can cost more where a provider rounds up, as Telnyx does to 60 seconds.
Plans can also change the math. We worked out two crossover points from published rates:
- Bland AI: Build costs less than Start above about 14,950 talk minutes a month. That is 299 divided by 0.02, before telephony and transfers.
- echowin: The $45 bundle beats pay as you go above 225 minutes a month (45 divided by 0.20).
These are our calculations, not vendor claims.

What it costs at your volume
We priced three monthly volumes from each provider’s published rates: 1,000, 4,000 and 10,000 talk minutes. These are our estimates, not quotes. They are not the research score for normalized cost.
| Provider | 1,000 min | 4,000 min | 10,000 min | What the estimate covers | Still add |
|---|---|---|---|---|---|
| Vapi | $87 to $137 | $349 to $548 | $873 to $1,369 | Hosting, speech-to-text, language model, voice and telephony, from low to high published component rates | Plan fee if you need Core ($29/month) |
| ElevenAgents | $80 | $320 | $800 | Voice platform minutes at about $0.08 | Language model, telephony |
| Retell AI | $70 to $310 | $280 to $1,240 | $700 to $3,100 | Published headline range of $0.07 to $0.31 a minute | Check what your model and phone line add |
| LiveKit | $0 | $50 | $100 | Platform plan only (Build or Ship) | Language model, voice, phone minutes past the plan |
| Synthflow | Not published | Not published | Not published | Sold by annual contract | Ask for a quote |
| Telnyx | $59 | $237 | $592 | Voice engine, a low-cost language model and telephony at $0.0592 a minute | 60-second call rounding raises short calls |
| Bland AI | $140 | $560 | $1,400 | Start plan at $0.14 a minute with language model, speech-to-text and voice | Telephony |
| echowin | $150 | $580 | $1,400 | Prepaid bundles at 14 to 15 cents a minute | Numbers ($10/month each), SMS and chat |
How we built the table:
- Vapi: low end uses the lowest published rate for each part and Telnyx phone minutes at $0.0055. High end uses the highest rate for each part and Twilio inbound at $0.008.
- ElevenAgents: every paid plan works out to about $0.08 a minute, so we used $0.08.
- LiveKit: Build is free to 1,000 minutes. Ship is $50 for 5,000, plus $0.01 a minute over that.
- Telnyx: $0.05 voice engine, about $0.006 for the Kimi language model and $0.0032 for telephony.
- echowin: cheapest bundle mix, such as 2,500 plus 1,000 plus 500 minutes for $580 at 4,000.
Treat each row as a floor, not a total. The rows do not cover the same things, so do not rank providers on this table. Call length, concurrency, taxes, recordings and volume discounts are not modeled. Ask each vendor to price your own call pattern.

Side-by-side comparison
| Provider | Our rating | How you build | Phone options | Open source |
|---|---|---|---|---|
| Vapi | 63.0 / 100 | API-first developer platform | Own telephony, SIP and WebSocket; Twilio, Telnyx or Vonage | No |
| ElevenAgents | 61.0 / 100 | Dashboard, API and SDKs | SIP and Twilio numbers | No |
| Retell AI | 58.0 / 100 | No-code builder plus API | Twilio or Telnyx, or your own SIP line | No |
| LiveKit | 55.0 / 100 | Open-source framework you code | SIP and US numbers | Yes (Apache 2.0) |
| Synthflow | 52.0 / 100 | No-code editor, vendor-led rollout | SIP telephony | No |
| Telnyx | 51.0 / 100 | Platform on a carrier-owned network | Numbers and SIP from the carrier | No |
| Bland AI | 50.0 / 100 | Platform with the model and voice included | Billed separately | No |
| echowin | 43.0 / 100 | No-code builder plus REST API | Numbers as add-ons | No |
Only LiveKit offers an open-source framework among the eight. Bland AI lists on-prem or VPC deployment on Enterprise.

Which one fits your business?
| If you are | Start with | Why |
|---|---|---|
| A developer team that wants control | Vapi | Ranked first; API-first |
| A small team that wants a dashboard and many integrations | ElevenAgents | Widest integration documentation |
| Someone who wants to test on credit with strong security documents | Retell AI | Strongest published security documentation |
| A team that wants to host and customize the stack | LiveKit | Open source, Apache 2.0 |
| An enterprise that wants a vendor-led rollout | Synthflow | Contract support, no-code build |
| A business that wants phone service and AI from one vendor | Telnyx | Carrier-owned network |
| A buyer who wants one bundled per-minute rate | Bland AI | Language model and voice included |
| An agency or small team that likes prepaid bundles | echowin | Published bundles, white-label plans |
If none of these sound like you, you may want a ready-made receptionist instead. See Other options.

How to choose in six steps
- Count your calls. Estimate monthly talk minutes and the share that are short. This sets which pricing model suits you.
- Decide who builds it. If nobody can build and tune an agent, favor dashboards or a vendor-led rollout. If you would rather not build at all, a ready-made option may fit better. See our guide to the best AI receptionists for small business.
- List the systems it must reach. Calendar, CRM, help desk, payments. Check each provider’s integration docs for yours.
- Check the rules that apply to you. Healthcare needs a BAA. Recording consent laws vary by state. Ask each vendor what is included and at which plan.
- Narrow to two finalists. Use the comparison table and the cost section. Keep the two that fit your builder, your systems and your rules. Drop the rest.
- Run a five-call test. Use the same script on both finalists. Compare the results side by side.
| Test call | What to check |
|---|---|
| 1. A simple question | Does it answer correctly and briefly? |
| 2. A booking | Does it write to your calendar or system? |
| 3. A caller who interrupts | Does it stop talking and recover? |
| 4. A request it cannot handle | Does it hand off to a person cleanly? |
| 5. A noisy line or accent | Does it still understand? |
Decision checklist
Before you sign, confirm each of these in writing:
Other options we looked at but didn’t rank
More products came up in our research. We did not score them, so they carry no rating and no rank. Each group was left out for a different reason.
| Group | Products | What kept them unranked |
|---|---|---|
| Business phone systems with a built-in AI agent | RingCentral AIR, Quo Sona, Nextiva XBert, Allo, Emitrr, Dialpad AI Agents, Aircall AI Voice Agent, Zoom Virtual Agent, GoTo Connect AI Receptionist, Ooma AI Receptionist, JustCall AI Voice Agent, CloudTalk AI Voice Agents, Vonage AI Studio | Mostly ready-to-use receptionists inside a phone plan, a different product from the platforms ranked here. We could only partly check some of them. |
| Other build-your-own platforms | Drips, 1mind, Voiceflow, PolyAI, Parloa, Cognigy, Twilio ConversationRelay, Deepgram Voice Agent API, xAI Grok Voice Agent Builder, OpenAI Realtime, Ultravox, Pipecat Cloud, OmniDimension, Trillet, Autocalls.ai, Lindy, Vocode | We could not verify enough to score them. Several are building blocks, not finished platforms. |
| Failed our entry test | Air AI, Vapify, CallSphere | Air AI is no longer sold as a voice agent. Vapify is a management dashboard. CallSphere is a managed service. |
Want no build at all? A ready-made receptionist is the closer fit. See our guide to the best AI receptionists for small business.
Have a developer? The eight ranked platforms above are the ones we could check most fully.
Frequently asked questions
What is the best AI voice agent for a small business?
It depends on who builds it. Vapi ranked first overall at 63.0 / 100. ElevenAgents (61.0) suits dashboard-first teams. Check the “Which one fits” table above.
How much do AI voice agents cost?
Published rates here run from $0.05 a minute for a voice engine to $0.20 for pay as you go. Totals rise once language model and phone costs are added. See the cost section.
Do I need a developer?
For Vapi, Retell AI, LiveKit and Telnyx, it helps. ElevenAgents and echowin offer dashboards. Synthflow sells a no-code build with vendor support.
Can an AI voice agent answer my business line?
Yes, on platforms that connect a phone number or a SIP line. Check each provider’s telephony options and fees.
Can an AI voice agent make outbound calls?
Often yes, but the rules are stricter. In 2024 the FCC ruled that AI-generated voices count as “artificial” under the Telephone Consumer Protection Act. Outbound AI calls generally need the called person’s prior consent. See the FCC announcement. This is not legal advice.
Does the TCPA apply to inbound calls?
The FCC says the TCPA’s artificial-voice rule covers outbound calls. It does not extend to technology that answers inbound calls. See FCC 24-84. Ask a lawyer about your own use.
Is there a free AI voice agent?
Several have free entry points. ElevenAgents has a free plan. LiveKit has a $0 Build plan. Retell AI, Telnyx and Bland AI start at $0 on usage. Free tiers are small, so use them to test.
Can I use one for HIPAA-covered calls?
Only with the right agreement. Vapi lists a HIPAA add-on at $2,000 a month. Bland AI lists its BAA as Enterprise only. Confirm terms with any vendor in writing.
Will an AI voice agent replace my receptionist?
Not always. It fits repeatable calls well. Complex or sensitive calls usually still need a person. Most buyers start with a handoff to staff.
How did you rank these?
We rated published pricing, features, integrations, security documentation and reviews. We counted unverified items as zero. See How we rank.
Methodology
This guide ranks AI voice agent platforms that businesses build on. It is research-based. We read and checked published sources. We did not run test calls.
What this guide covers
We split AI phone products into three groups.
- Custom AI voice agent platforms. You build the agent. These are scored and ranked here.
- Business phone systems with a built-in AI agent. These are listed under Other options with no scores.
- Ready-made AI receptionists. These are not scored here. They are covered in a separate guide.
Scores from this guide do not compare with scores from other guides. Each guide uses its own weights.
Which platforms we considered
A platform had to pass two checks. Its official pages had to state how it is deployed, such as self-serve, developer-built or sales-led. And its AI voice agent had to handle business phone calls.
Three products failed the gate. One is no longer sold as a voice agent, one is a management dashboard, and one is a managed service. Eight platforms were scored in this pass. Another 17 build-your-own platforms are listed without scores under Other options.
We did not verify US availability for every platform. Confirm it before you sign up.
What we checked
The score is out of 100. It has 23 checks in seven areas.
| Area | Points | What we looked at |
|---|---|---|
| Core capabilities | 20 | Conversation design, inbound and outbound call functions, knowledge and tool calling, handoff to a person, monitoring and debugging tools |
| Pricing and value | 20 | Normalized monthly cost (12), how clearly pricing is published (4), contract flexibility (4) |
| Integrations and workflows | 20 | API and SDK support (6), carrier and SIP options (5), business system integrations (5), deployment controls (4) |
| Industry and audience fit | 10 | Fit for technical buyers (5), scale and use-case suitability (5) |
| Setup and support | 5 | Developer documentation (3), support access (2) |
| Security and transparency | 10 | Data handling, retention and deletion, recording consent controls, access protection, security and contract documents (2 each) |
| Customer reviews and reputation | 15 | Review quantity (10), average star rating (5) |
How the score works
Each check is marked Verified, Confirmed absent, Not Verified or Conflicting. A platform earns points only for checks that are Verified or Confirmed absent.
A check marked Not Verified earns zero points in the total. That is a gap in our evidence. It does not mean the platform failed.
For six platforms, 27 points could not be checked. Those are cost (12), review quantity (10) and star rating (5). So their highest possible score on verified evidence was 73.
Synthflow had two more points unchecked (recording consent controls). echowin had four more (recording consent controls, and security and contract documents). Across the eight, we could verify between 69% and 73% of the points.
As we confirm more evidence, scores can rise. The order can change.
How we handle reviews
Review points follow one rule for every platform. We count only independent, product-specific reviews from one review source per product. We do not mix reviews from different products or from a parent company.
More reviews earn more points, with the gain flattening out for large pools. A higher average star rating earns more points too.
We checked G2, Trustpilot, Gartner Peer Insights and Capterra on November 13, 2026, and show each product’s own ratings in its review. We leave out company-wide pages and pages for other products. Only Retell AI and Synthflow have large pools.
We did not add review points to the scores. All 15 review points stay unscored until one review source is chosen for all eight platforms. A review count shows how much evidence exists. It does not show call quality.
How we handle pricing
We recorded each platform’s published plans and per-minute rates from its pricing page on November 13, 2026. Taxes are excluded.
The 12 points for normalized monthly cost need a total price under one shared scenario. That includes the AI model, voice, telephony and hosting. Research has not scored that for any platform, so those points are unresolved.
Our own published-rate estimates sit in the cost section and do not change any score. The two other pricing checks, price transparency and contract flexibility, were scored.
A per-minute rate is not a total cost. Most platforms bill the language model, speech and phone carrier separately. Check what the per-minute price leaves out.
How we rank
Platforms are ordered by verified points, highest first. We do not recalculate or adjust the order by hand. Differences of one to three points are small. They could change once the missing evidence is confirmed.
What we did not do
We did not place test calls. We did not measure speed, voice quality, accuracy or uptime. When a provider says its system can do something, we report it as the provider’s claim.
We score documentation, not performance. Our five-call test script shows how to test a platform yourself.
Limits to keep in mind
- Cost and review points are not yet scored for any platform. Ratings from several review sites are shown for reference only.
- Some documentation was read from page summaries, not every help article.
- Pricing pages change. A few of them list the same fee two ways. Where we found that, we say so in the review.
- The eight scored platforms are not the whole market. Others appear under Other options without scores.
Prices checked November 13, 2026.
The bottom line
Vapi ranked first at 63.0 / 100 for teams that want control. ElevenAgents (61.0) and Retell AI (58.0) are the easiest self-serve starts. LiveKit (55.0) is the choice if you want to host the stack.
Synthflow (52.0) suits a vendor-led enterprise rollout. Telnyx (51.0), Bland AI (50.0) and echowin (43.0) each fit a narrower need.
The gaps are small, and cost and review evidence are missing for most. Treat the order as a starting point. Test two or three with the same five calls before you commit.
An agent only helps if callers reach you in the first place. Our local SEO guide and our guide to AI search optimization cover how businesses get found.
Sources
- Retell AI pricing
- Vapi pricing
- Bland AI pricing
- Synthflow pricing
- ElevenAgents pricing
- Telnyx Voice AI Agents pricing
- echowin pricing
- LiveKit pricing
- G2: Vapi
- G2: Retell AI
- G2: Synthflow
- Trustpilot: Retell AI
- Gartner Peer Insights: Vapi
- BLS: Receptionists
- FCC: AI-generated voices in robocalls
- FCC 24-84
