Call-bot.ai

Comparisons14 min read

Voice AI Platforms Compared: Pricing, Languages and Compliance from Public Documentation

This comparison of five voice AI platforms rests on their public documentation and pricing pages, read on October 21, 2026; Call-bot.ai has not tested these products hands-on. It sets published prices, language support, telephony options and vendor compliance claims side by side, with no scores and no winner.

A hand typing on a laptop whose screen shows lines of code

Photo: Lukas Blazek · Pexels

Disclosure: Call-bot.ai is published by Nehos Groupe, which also publishes Alyas.ai, mentioned in this article. Who we are.

This is a documentation-based comparison, not a test. Call-bot.ai has not run hands-on tests of these products: everything below comes from the vendors' public pricing pages and documentation, read on October 21, 2026, and every claim about compliance or speed is flagged as the vendor's own. If we run hands-on tests later, the results will be added in a separate dated section.

Nehos Groupe, which publishes Call-bot.ai, also publishes Alyas.ai. We left it out of this comparison, which covers five platforms that publish developer documentation; Alyas.ai is set up online without a developer.

The five platforms are Vapi, Retell AI, Synthflow, Bland and ElevenLabs' ElevenAgents. All five publish developer documentation and were live when we read them. For background, see what an AI voice agent is; for cost layers no price page includes, hidden costs; for campaigns on these tools, AI outbound calling.

How we read these pages

For each platform we read the pricing page, the documentation on telephony, languages, compliance and limits, and the trust center when public, all on October 21, 2026, in U.S. dollars. To normalize prices we separate each platform's own charge from the components (speech recognition, language model, voice, telephony), which some vendors pass through at provider cost and others bill as their own lines, and we do not add components across vendors because the stacks differ. The "1,000 minutes" row applies each vendor's published rates to one month of 1,000 minutes and counts only each platform's base charge (hosting fee, voice infrastructure line, per-minute rate plus platform fee, or per-minute rate or plan fee); each priced cell names what stays extra, even where the vendor bills it itself. We do not rank by latency: the vendors define it differently. Not established from the pages we read: which languages an ElevenLabs agent covers, since its agent guide says 31 while its models page lists 90+ for the v4 Turbo and Realtime models, and the content of reports shared under NDA. Prices change; the linked pages are the reference. Our testing methodology describes how we evaluate products once we test them.

Vapi: hosting fee plus components

Usage-based. The pricing page lists a $0.05 per minute hosting fee, prepaid, with models passed through at cost, $5 in free credits, four concurrent calls and 14 days of raw data retention. Packages add support and capacity: Core at $29 a month (10 concurrent calls), Pro at 10% of the hosting fee with a $999 monthly minimum (30 calls, 99% uptime SLA), and a sales-only Premier. Components are billed separately: the page's calculator shows Deepgram speech recognition at $0.0095 to $0.0099 per minute, OpenAI models at $0.0077 to $0.0452 and ElevenLabs voices at $0.0146 to $0.0238. Vapi's own telephony and SIP transport are free; Twilio outbound is listed at $0.014 per minute and Telnyx at $0.0055. HIPAA handling is a $2,000 monthly add-on per organization and extra concurrency costs $10 per line per month. The page's own estimate for 1,000 minutes is $82 to $129 a month.

Retell AI: itemized per-minute lines

Pay as you go, with $10 in free credits and 20 concurrent calls included; each extra concurrent call is $8 a month. The headline reads "$0.07-$0.31 / min," but the detailed grid is itemized: voice infrastructure at $0.055 per minute; voices from $0.015 to $0.10 across 12 options; language models from $0.0016 to $0.64 across 36 models (the highest figure is a fast-tier price); and telephony at $0.015 per minute for the default U.S. selection when Retell supplies it (the country selector shows other rates, such as $0.06 for France and $0.10 for Germany, Spain or the United Kingdom), with no charge for your own SIP trunk. Add-ons include $0.005 per dial for batch calls, $0.10 per outbound call for branded caller ID, $0.005 per minute for the knowledge base and $0.10 per minute for AI quality assurance after 100 free minutes. Enterprise is custom.

Synthflow: sales-led contracts

Sales-led. The pricing page states that "Enterprise contracts start at $30,000 annually," with final pricing scoped by volume, concurrency, telephony and security needs. The billing documentation calls its model sales-led, and the concurrency page says the earlier Pro, Growth and Agency plans are no longer sold. No per-minute rate is published. The homepage FAQ, by contrast, says you can build an agent for free and then pay as you go, again without giving a rate.

Bland: per minute, models included

Per minute, with a platform fee on higher tiers. On the pricing page, Start is $0.14 per minute with no platform fee (10 concurrent calls, 100 calls a day), Build is $299 a month plus $0.12 per minute (50 concurrent calls, 2,000 calls a day), and Enterprise is custom. The rate includes the language model, speech recognition and voices, with no token charges; telephony is billed separately, on your own carrier or Bland's at pass-through cost. Transfers to a human on Bland numbers add $0.05 (Start) or $0.04 (Build) per minute, and the billing documentation adds a $0.015 minimum per outbound attempt on Bland's telephony. Bland's own pages disagree in several places. The documentation index says the per-minute price also covers telephony, while the pricing page says it is billed separately. The billing page lists a $499 Scale tier at $0.11 per minute (no billing period shown), in a pricing structure "effective December 5th, 2025", and a $29.99-a-month Agent Phone tier for a single number, neither of which the pricing page shows. The billing page also says SIP connectivity is for enterprise customers, while the pricing-page FAQ lists SIP trunks from any provider without that restriction.

ElevenLabs (ElevenAgents): plans with included minutes

Plans bundle call minutes; the figures are price per month (the page's monthly-billing setting), included minutes and concurrent calls: Free ($0, 15, 4), Starter ($6, 75, 6), Creator ($22, 275, 10), Pro ($99, 1,238, 20), Scale ($299, 3,738, 30) and Business ($990, 12,375, 40); the page's yearly-billing setting shows lower monthly equivalents, such as $82.50 for Pro. Extra minutes are $0.08 on every plan, including the Free / pay-as-you-go tier, which includes 15 minutes. Above the concurrency limit a burst rate of $0.16 applies, up to three times the limit. The language model is billed separately, on top. ElevenLabs charges no telephony fee: you connect a provider such as Twilio or a SIP trunk, and that provider bills you.

Side by side

Vapi Retell AI Synthflow Bland ElevenLabs
Billing hosting fee plus components at cost; optional packages itemized per-minute lines sales-led annual contract per minute; LLM, speech, voices included monthly or yearly plans with included minutes, or pay as you go at $0.08/min
Platform-owned charge, 1,000 min (our arithmetic) $50; components extra $55; components extra not published $140 (Start), $419 (Build); telephony extra $80 (pay as you go, $0.08/min, no commitment) or $99 (Pro on monthly billing, 1,238 minutes included); LLM and carrier charges extra
Languages, as documented by provider; auto-detect via Deepgram, Google, Gladia, AssemblyAI; Arabic on Azure, Deepgram, ElevenLabs pages 55-language table; Arabic, English, French, German, Spanish on all four speech-recognition providers 34 distinct entries (Hebrew is listed twice) incl. a multilingual mode; no Arabic "40+" claimed; API list has no Arabic code 31 per the agent guide, 90+ per the models page for v4 Turbo and Realtime models; Arabic in model lists
Telephony free U.S. inbound number; import Twilio, Vonage, Telnyx or SIP; no outbound on free numbers Retell U.S./Canada numbers; custom SIP buy U.S./Canada/Australia number; Twilio; SIP/PBX on Enterprise built-in Twilio; own Twilio; SIP (billing docs: for enterprise customers) bring your own: Twilio, SIP, others
Compliance (vendor claims) HIPAA, SOC 2 Type II, GDPR; trust center also lists PCI DSS v4.0.1, SOC 3, CCPA; EU region Premier only HIPAA, GDPR, SOC 2 Type 1 and 2; trust center also lists ISO 27001:2022, CCPA; says it runs no EU services trust center lists SOC 2 Type 2, ISO 27001, HIPAA, GDPR, PCI DSS v4.0.1; EU cluster SOC 2, HIPAA, GDPR, PCI DSS (self-attested), ISO 27001; documents under NDA SOC 2, HIPAA-eligible, GDPR-aligned; trust center also lists ISO 27001:2022, PCI DSS 4.0.1 Level 1; EU residency on Enterprise
Latency (vendor claims) "typically around 800 milliseconds for end-to-end voice processing" (FAQ) "as low as 600ms" (from the end of the user's speech to the start of the agent's reply) "sub-500ms" (homepage FAQ, scope not stated); telephony round-trip target under 100 ms "sub-400ms response latency" (documentation index) about 75 ms for the Flash v2.5 voice model only, excluding application and network latency (about 100 ms for v4 Turbo)
Integrations (examples from each documentation index) API, function, MCP, transfer, SMS tools; Make HubSpot, Salesforce, Zendesk, Cal.com, Calendly, MCP Cal.com, HubSpot, Salesforce, GoHighLevel, Zapier, MCP Cal.com, Salesforce, Slack, HubSpot, custom API, MCP Cal.com, Calendly, HubSpot, Salesforce, Zendesk, MCP
Published limits 4/10/30 concurrent calls by package; 10,000 contacts per campaign 20 concurrent calls; burst $0.10/min set by contract 10/50 concurrent calls; 100/2,000 calls a day 4 to 40 concurrent calls by plan

Compliance and latency cells are vendor statements as read on October 21, 2026: see Vapi's FAQ, EU region page and trust center, Retell's compliance and latency pages and trust center, Synthflow's trust center and homepage, Bland's pricing page, trust and security page and documentation index, and ElevenLabs' models and data residency pages and trust center. The ElevenLabs figure carries the vendor's footnote that it excludes application and network latency.

Which kind of team should look at which

This follows what each page says, not our judgment.

  • You want to choose each component and pay providers at cost. Vapi's FAQ says it is built for developers and modular at every level of the voice pipeline, and its pricing page passes models through at cost. Retell also itemizes lines but says it runs the telephony, speech and model layers so you configure the agent.
  • You want one per-minute rate. Bland's page puts the language model, speech recognition and voices in the rate, with telephony separate.
  • You want included minutes and your own carrier. ElevenLabs' plans bundle minutes; telephony comes from your provider.
  • You run a contact center on SIP or a PBX. Synthflow's documentation lists Five9, Genesys Cloud, Talkdesk and others, and places SIP/PBX on its Enterprise plan. Bland lists on-prem or VPC deployment, a BAA, SSO and data residency under Enterprise; Vapi's Premier adds SSO and dedicated SIP.
  • You need data in the EU. Vapi's EU region is Premier-only. Synthflow's EU cluster keeps storage and processing in the EU, though its Data Region page notes that if you connect your own Twilio account, only some Twilio API requests run in Twilio's Ireland region. ElevenLabs offers EU residency to Enterprise customers, with storage in the EU and processing possibly elsewhere unless Zero Retention Mode is used with the API; its page adds that optional integrations, such as custom LLMs or post-call webhooks that require out-of-region processing, may still result in processing outside the EU. Retell's page says it does not currently operate services in the EU. Bland lists data residency under Enterprise.
  • You need HIPAA. All five describe a path. Retell's plan table lists HIPAA/BAA on both plans; Vapi sells it as a $2,000 monthly add-on; Bland lists the BAA under Enterprise, ElevenLabs lists BAAs for HIPAA customers under Enterprise, and Synthflow's security page points HIPAA customers to its Business Associate Agreement.
  • You just want to try one. Free starting points: $5 at Vapi, $10 at Retell, two credits and a number at Bland, 15 minutes on ElevenLabs' Free plan. Synthflow's pricing page shows no self-serve price, while its homepage FAQ says you can build an agent for free and then pay as you go.

Frequently asked questions

What is a voice AI platform?

A service that connects speech recognition, a language model, a voice and telephony into agents that answer or place calls. The pages here split into platforms that bill components separately and ones that bundle them.

What are the alternatives to Vapi?

The other four platforms in this comparison. Retell itemizes components like Vapi, Bland bundles language model, speech and voices, ElevenLabs bundles minutes into plans, and Synthflow sells by contract.

How do Vapi and Retell prices differ?

Vapi lists a $0.05 per minute hosting fee plus components at cost; Retell lists $0.055 per minute for voice infrastructure plus its own voice, model and telephony lines. Vapi's usage-only tier includes four concurrent calls and Retell's pay-as-you-go includes 20; extra lines cost $10 and $8 a month.

What is the best AI voice agent platform?

We do not rank them: we have not tested these products, and the pages describe different billing structures. Start from your constraints, such as EU data, HIPAA, telephony or volume.

Are the compliance claims verified?

No. They are vendor statements, linked above. Ask for reports, data processing agreements and BAAs; Bland, Synthflow and Vapi say some documents are shared under NDA.

In short

The five platforms bill in different ways, so a headline per-minute price does not compare. Documentation shows concrete differences in telephony, EU data handling, HIPAA routes and published limits, while compliance and latency figures remain vendor claims we have not verified. Treat this as a reading list for a shortlist, not a verdict. To scope a build, talk to us.

Sources

Talk to us about your project