Retell AI Alternatives: What Each One Actually Fixes

Most articles about Retell AI alternatives are answering a question nobody asked. They line up five voice platforms, tabulate the per-minute rates, and declare a winner. But if you are searching for an alternative to something you are already using, you do not have a blank slate. You have a complaint. Something specific about Retell is blocking you, and the only useful answer is which platform removes that specific block.

So this piece is organised by the complaint, not by the vendor. Five real reasons teams leave Retell, and what actually fixes each one. Every rate below was read off the vendor's own pricing page on 13 August 2026.

There is also a correction to make first, because the most commonly repeated reason to leave Retell is no longer true.

The restriction that quietly disappeared

For most of the past two years, the standard advice ran like this: Retell is excellent until you need to handle protected health information, at which point compliance is locked behind an Enterprise contract and you should look elsewhere. That advice is now wrong.

Retell's security and compliance documentation states that the Business Associate Agreement and the Data Processing Addendum, including EU Standard Contractual Clauses, are available for self-signing at click-agreements.retellai.com, with no additional fee. Not a sales call. Not a minimum commitment. You sign it yourself, on the pay-as-you-go plan that costs nothing per month, and then configure per-agent data retention anywhere from one day to two years and choose whether each agent stores everything, excludes PII, or keeps basic attributes only.

Compare that with what the rest of the category charges for the same signature. Vapi lists HIPAA as a $2,000 per month add-on and Zero Data Retention at $1,000 per month, on both its Build and Scale plans. Bland puts BAA, SSO and data residency exclusively in the Enterprise column. Deepgram signs BAAs for Enterprise customers only. LiveKit includes a signed HIPAA BAA from its Scale tier at $500 per month.

At 5,000 minutes a month, Vapi's HIPAA add-on works out at forty cents per minute of pure compliance overhead. That is roughly four times what the entire Retell stack costs to run. If compliance is the reason you were told to leave Retell, check the current terms before you migrate anything, because you are about to pay a great deal for something you already have.

The same goes for the other half of the old advice. Retell includes twenty concurrent calls at no charge and bills $8 per additional concurrent line per month. Vapi includes ten and charges $10 per line. Retell charges nothing for telephony when you bring your own SIP trunk, and there is no plan tier that withholds product features. That combination is genuinely hard to beat, and a fair number of people who switch end up paying more for less.

With that out of the way, here are the reasons that hold up.

1. You need processing inside the EU

This is the hardest limitation Retell has, and it is the one they state plainly themselves. From the same compliance page: Retell complies with GDPR via AWS and its Data Processing Addendum, but does not currently operate services within the European Union. Signing the SCCs makes the transfer lawful. It does not make the processing local.

If your customer, your regulator or your procurement team requires that audio and transcripts stay inside the EU, no amount of money moves Retell. There is no tier that offers it. This is the one complaint where switching is not optional.

What fixes it. Deepgram publishes a dedicated EU endpoint at api.eu.deepgram.com specifically so that processing stays within the Union, and it is available on the ordinary pay-as-you-go plan rather than gated behind a contract. LiveKit offers region pinning from its Scale tier at $500 per month, alongside SOC 2 Type II and the signed BAA. Bland offers data residency, but only on Enterprise, which means a sales cycle and a committed volume.

There is a fourth answer that nobody markets, which is to stop using a hosted platform and run the pipeline yourself in an EU region. That is section four.

2. You want to pay model providers at cost

Retell resells the models. Its published pricing sets a fixed per-minute rate for every LLM on the menu, from $0.006 per minute for Gemini 2.5 Flash Lite up to $0.32 per minute for GPT 5.5 on the fast tier, and a fixed rate for each of five text-to-speech providers. You choose from the menu at Retell's price. There is a Custom LLM option, so you are not entirely locked in on the model, but the voice side is fixed and the $0.055 per minute Retell Voice Infra charge is unavoidable.

What fixes it. Vapi's whole commercial position is the opposite arrangement. It charges $0.05 per minute for hosting and passes model provider costs through at cost, dropping to zero if you bring your own API key for speech-to-text, the language model and text-to-speech alike. You pay OpenAI, Deepgram and ElevenLabs directly at whatever rate you have negotiated with them.

Whether that saves you anything depends entirely on your stack, and it is worth doing the arithmetic rather than assuming. Retell's markup is not obviously punitive. Its platform voices, Minimax, Fish, Cartesia and OpenAI voices all bill at $0.015 per minute, and its ElevenLabs voices at $0.040 per minute. LiveKit's pass-through table, on the same day, showed ElevenLabs Flash at $0.0900 per minute and Multilingual at $0.1800 on its lower tiers. ElevenLabs' own Agents pricing works out at $0.080 per minute on every tier. Retell's resold ElevenLabs rate is less than half what you would pay going direct through some routes.

Where bringing your own key genuinely pays is at volume, on a stack where you already have committed spend or negotiated rates with the providers. If you are running a few thousand minutes a month on standard rates, the $0.005 per minute difference in platform fee is worth twenty-five dollars and the model side is close to a wash. We walked through the full five-layer arithmetic in what an AI voice call actually costs per minute, and the short version is that people consistently misjudge which layer their money is going to.

3. You want one predictable number

Retell's bill has moving parts. Voice infrastructure, text-to-speech, the language model, telephony, and then the add-ons: knowledge base at $0.005 per minute, advanced denoising at $0.005, safety guardrails at $0.005, PII removal at $0.01, AI quality assurance at $0.10 per minute after the first hundred free, branded calling at $0.10 per outbound call, batch calling at $0.005 per dial. Each is small. Together they are the difference between a forecast and a guess, and if somebody on your team switches the agent to a more expensive model the bill moves without anyone approving it.

What fixes it. Bland bundles the language model, speech-to-text and text-to-speech into a single per-minute rate with no token charges and no provider pass-throughs. Start is $0.14 per minute with no platform fee, Build is $0.12 per minute on a $299 monthly platform fee, Scale is $0.11 per minute on $499. Telephony is separate, either on your own carrier or Bland's Twilio at pass-through cost. Transfers bill at a separate lower rate, $0.05, $0.04 and $0.03 per minute respectively.

Two things to check before you commit. First, run the crossover maths, because the platform fees only pay for themselves at real volume: Build overtakes Start at 14,950 minutes a month, and Scale does not overtake Build until 20,000. Below fifteen thousand minutes, Start is the cheapest Bland plan and the higher tiers are buying you rate limits and features, not a discount.

Second, look at the caps before the rate. Start allows 10 concurrent calls and 100 calls per day. That ceiling, not the price, is what will stop a growing receptionist deployment. And Bland gates considerably more behind Enterprise than Retell does, including warm and live transfers, SMS and web chat, the appointment scheduling node, guardrails, SSO, data residency and the BAA.

One caution on Bland's own comparison. Its pricing FAQ describes Retell's voice infrastructure as "about $0.07/min" and estimates production Retell stacks at $0.11 to $0.25 per minute. Retell's published card says $0.055, and the mid-range stack costed below comes to $0.097 including telephony. Never take a competitor's rate from a vendor's comparison page.

4. You want to own the stack

Retell is a closed platform on shared infrastructure. A dedicated server is an Enterprise line item. You cannot read the orchestration code, you cannot change how turn detection or interruption handling works, and you cannot run it on your own hardware or inside your own VPC. For most businesses that is a feature. For a team with an unusual conversational flow, a latency budget it needs to tune, or a regulator that wants the audio never to leave a network it controls, it is a hard stop.

What fixes it. Two genuinely open-source options, both with a hosted tier if you want to start quickly and self-host later.

LiveKit publishes both its Agents framework and its media server as open source. The hosted tiers run from a free Build plan with 1,000 agent session minutes, 5 concurrent sessions, one free number and $2.50 of inference credits, through Ship at $50 per month for 5,000 minutes and 20 concurrent sessions, to Scale at $500 for 50,000 minutes, up to 600 concurrent sessions, region pinning, RBAC, SOC 2 Type II and the signed BAA. On the published rates the metered side runs at $0.0100 per minute for the agent session, $0.0100 for observability and $0.0100 for US local inbound, plus $1.00 a month per local number, with models billed separately. LiveKit's own calculator publishes a default assembled total of $0.0672 per minute, which sits neatly below the range we independently arrived at, and is useful corroboration rather than a new claim.

Pipecat is the other. The framework, Pipecat Flows, the Smart Turn model and SmallWebRTC are all open source, and Daily runs a hosted version whose Pipecat Cloud pricing is unusually legible: agent hosting at $0.01 per minute for a half-vCPU voice agent, $0.02 for a full vCPU, $0.03 for one and a half, with reserved instances at a twentieth of those rates. WebRTC transport is free for one-to-one voice sessions. PSTN dial-in and dial-out is $0.018 per minute, or $0.003 to $0.02 on SIP. Krisp background-noise suppression is free to 10,000 minutes a month and $0.0015 after. Models are billed direct to whichever of the eighty-odd supported providers you pick.

Both give you something no hosted platform will, which is the ability to read exactly what happens between the caller finishing a sentence and the agent starting one. If your problem with Retell is that the conversation feels wrong and you cannot find out why, this is the category you want.

5. You need concurrency far past twenty

Twenty free concurrent calls is generous for inbound. It is nowhere near enough for a serious outbound campaign, where concurrency, not minutes, is the constraint that determines how fast you can work a list. Retell sells extra lines at $8 per concurrent call per month, which is reasonable, but it is a monthly subscription line item rather than something that flexes with a burst.

What fixes it. Pipecat Cloud states unlimited concurrency outright. Deepgram's Voice Agent API allows up to 45 concurrent websocket connections on pay-as-you-go and 60 on Growth. LiveKit Scale goes to 600. Bland scales by tier, 10 then 50 then 100, but layers daily and hourly call caps on top, which is a second ceiling most comparisons miss.

If outbound volume is your reason for looking, read how outbound AI calling actually works at scale before choosing, because concurrency interacts with carrier reputation and answer rates in ways that make the cheapest per-minute rate irrelevant.

The one that bundles speech in

Worth its own note, because it does not fit the five complaints neatly. Deepgram's Voice Agent API is not a platform in the Retell sense, it is an API that folds speech-to-text, orchestration and optionally text-to-speech into one per-minute rate. Standard is $0.056 per minute through 12 September 2026 and $0.075 after, Standard with your own TTS is $0.065, Custom with your own LLM is $0.050 through 12 September then $0.065, and Custom with your own LLM and TTS is $0.050 flat. The Advanced tier is $0.122 rising to $0.163.

Two things matter here. The first is that those promotional rates expire on a published date, so any budget you build past mid-September should use the standard column. The second is that this is not a cost win against Retell. Add US telephony at $0.015 and Standard lands at $0.071 per minute today and $0.090 after the promotion ends, against Retell's $0.097 for a comparable stack. You are switching for the EU endpoint, for one vendor across speech and orchestration, and for the ability to bring your own model, not to save money.

If you are assembling the layers yourself, Deepgram's component rates are the reference point: Nova-3 monolingual streaming at $0.0048 per minute promotional and $0.0077 standard, Flux English at $0.0065 and $0.0077, Aura-2 text-to-speech at $0.030 per thousand characters and Aura-1 at $0.0150. Text-to-speech bills per character, so converting to minutes requires an assumption. At 450 synthesised characters per call minute, Aura-2 is $0.0135 per minute and Aura-1 is $0.00675.

What it costs at 5,000 minutes a month

A single scenario, run consistently: a US inbound receptionist, mid-tier model, standard voice, 5,000 minutes a month, using the platform's own telephony where it offers it.

Read that table carefully, because the striking thing is not that anything beats Retell. It is how little difference the platform layer makes once you hold the models constant. The spread between the cheapest and the dearest platform-and-telephony line here is a few hundred dollars a month, and a single careless model choice moves the bill by more than that. Retell publishes GPT 5.5 fast tier at $0.32 per minute and Gemini 2.5 Flash Lite at $0.006. That is a factor of over fifty, on one dropdown, inside one platform. You will save more by choosing the model well than by choosing the platform well.

If you want to run your own numbers rather than trust a scenario, the voice AI cost per minute calculator breaks the stack into its layers, and the AI receptionist cost calculator does the same for an inbound answering deployment. For the telephony layer on its own, which is the layer that surprises people most, the Twilio cost calculator includes the Media Streams charge that most estimates omit. That charge is a real line on Twilio's own US voice rate card at $0.0044 per minute, and it is why a genuine outbound US call bills $0.0184 rather than the $0.0140 everyone quotes.

Where Retell still wins

Being honest about this is the difference between a useful comparison and an affiliate list.

The free self-signed BAA and DPA, described above, is the single strongest thing in Retell's commercial position and almost nobody has updated their advice to reflect it. Twenty included concurrent lines is double Vapi's ten. Bringing your own SIP trunk costs nothing, where several competitors either charge for it or reserve it for a higher tier. There is no plan that withholds product features, so a small deployment gets the same platform a large one does. Billing is per second with no per-call rounding. Failed connections are not billed at all, and voicemail is billed only for the duration the agent is actually talking.

Against that, two real irritants. You are charged during silence and hold, because the speech-to-text engine stays live throughout, which matters if your flow involves the caller waiting. And after a transfer the AI fee stops but the telephony fee continues for the remainder of the call, which is easy to miss when forecasting a transfer-heavy support line.

Our fuller assessment of the platform on its own terms is in the Retell AI review, and the head-to-head against its closest competitor is in Vapi vs Retell AI. If you are still choosing rather than switching, the current state of the voice agent platforms covers the field, and how to use Retell AI walks through the build itself.

The switching costs nobody puts in the table

Every comparison stops at the per-minute rate. The actual cost of moving is somewhere else entirely.

Prompts do not port. A prompt tuned against one platform's turn detection and interruption behaviour will behave differently on another, and you will spend real hours retuning it before the new agent is as good as the old one. Phone numbers either port, which takes days and paperwork, or get replaced, which means updating every listing, signature and printed card that carries the old one. Integrations get rebuilt, since webhook payload shapes and function-calling conventions differ between platforms. Call history usually does not come with you, and on Retell's pay-as-you-go plan it is only retained fourteen days anyway, so export anything you need before you close the account. And whoever knows the platform has to learn a new one.

For a small deployment that is a week of work. Set against a saving of a hundred dollars a month, it takes most of a year to break even, by which point the pricing has changed again. Which is the honest test: if the reason to switch is a number, it is probably not worth it. If the reason is something Retell structurally cannot do, like EU processing or self-hosting, the number is irrelevant and you should have switched already.

Deciding in one pass

Work down this list and stop at the first line that describes you.

If processing must happen inside the EU, you are leaving. Deepgram's EU endpoint or LiveKit Scale with region pinning.

If you need to read or change the orchestration code, or run it on your own infrastructure, you are leaving. LiveKit Agents or Pipecat, both open source.

If you need concurrency well past twenty with burst behaviour, look at Pipecat Cloud's unlimited concurrency or LiveKit Scale.

If a predictable single number matters more than the lowest number, Bland Start below fifteen thousand minutes a month.

If you have negotiated model rates or committed provider spend, Vapi with your own API keys.

If you got this far, you do not have a Retell problem, you have a configuration problem. Change the model, check whether you are paying for add-ons you are not using, sign the BAA you did not know was free, and save yourself the migration. Before you decide, test the agent design itself rather than the platform, and if the issue is call quality rather than cost, how to build an AI receptionist that actually works and setting up Twilio voice for AI agents cover the two places most deployments actually go wrong.

Frequently asked questions

Is Retell AI still gated behind Enterprise for HIPAA?

No, and this is the most out-of-date claim in the category. Retell's compliance documentation states the BAA and DPA are self-signing at no additional fee, available on the pay-as-you-go plan. You also get per-agent data retention from one day to two years and per-agent PII controls without a contract. Verify the current terms yourself before acting on any article that says otherwise, including this one.

Which Retell alternative is cheapest?

At 5,000 minutes a month with models held constant, Pipecat Cloud has the lowest platform and telephony line at roughly $0.028 per minute, followed by LiveKit Ship. But the platform layer is the smallest variable in the bill. Retell's own model menu spans $0.006 to $0.32 per minute, so your model choice moves the total far more than your platform choice does.

Can I self-host a Retell alternative?

Yes. LiveKit publishes both its Agents framework and its media server as open source, and Pipecat, Pipecat Flows, the Smart Turn model and SmallWebRTC are all open source too. Both have hosted tiers, so you can start managed and move to your own infrastructure later without rewriting the agent. Retell itself cannot be self-hosted at any tier.

Does any voice platform process calls inside the EU?

Deepgram publishes a dedicated EU endpoint on its ordinary pay-as-you-go plan. LiveKit offers region pinning from its Scale tier at $500 per month. Bland offers data residency only on Enterprise. Retell states in its own documentation that it does not currently operate services within the European Union, so signing the SCCs makes the transfer lawful but does not keep processing local.

How much does it cost to switch off Retell?

The per-minute difference is the small part. Budget for retuning prompts against different turn detection, porting or replacing phone numbers, rebuilding webhook and function-calling integrations, exporting call history before the account closes, and retraining whoever runs the agent. For a small deployment that is roughly a week of work, which takes most of a year to recover from a saving of a hundred dollars a month.

Are these prices going to change?

Yes, and some of them have a date on them. Deepgram's promotional streaming and Voice Agent rates revert on 13 September 2026, and Flux TTS stops being free on 12 September. Every figure here was read from the vendor's own pricing page on 13 August 2026. Check the source pages before committing to a budget, because voice AI pricing moves faster than almost any other software category.

If you would rather not run this comparison yourself, I build and deploy AI voice agents on whichever of these platforms actually fits the constraint you are working against. You can hire me on Fiverr or work with me on Upwork, and I publish build walkthroughs on YouTube.