Retell vs. Vapi vs. Bland vs. ElevenLabs: a voice AI platform roundup

Teams shortlisting a voice AI platform for phone calls keep landing on the same four names: Retell AI, Vapi, Bland AI, and ElevenLabs Agents. All four will answer and place calls with an AI agent; they differ sharply in how they price a minute, how many languages they cover, how they connect to your phone system, what compliance posture they publish, and what testing tooling they ship.

One thing up front, in the interest of transparency: this roundup is published by ThunderPhone, which sells a competing voice AI platform. The four platforms above are profiled first, on their own published facts, each with a source link and accessed date. ThunderPhone appears as a clearly labeled fifth option at the end, described with the same discipline. Nothing here is a quote, and every price is a point-in-time reading of a public page.

Retell AI

Retell prices components separately on a pay-as-you-go basis: voice infrastructure at $0.055/minute, language models from $0.003 to $0.16/minute by model, text-to-speech at $0.015–$0.040/minute by provider, standard telephony at $0.015/minute (varies by country), and phone numbers at $2/month. The advertised all-in range is $0.07–$0.31/minute depending on the model and add-ons; optional per-minute add-ons stack on top (knowledge base, denoising, and guardrails at $0.005 each, PII removal at $0.01, AI quality assurance at $0.10 after the first 100 free minutes). There is no mandatory subscription; 20 concurrent calls are included, with extra concurrency at $8 per concurrent call per month. Retell pricing, accessed 2026-08-18.

Languages are provider-dependent rather than a single published total — per-provider counts run roughly 30–80, and Retell's docs defer to the dashboard as the source of truth; multilingual agents need one voice and one speech-recognition provider covering the whole combination. Retell language docs, accessed 2026-08-18. Telephony is flexible: "no charge for SIP trunking/custom telephony," with native Twilio, Vonage, and Telnyx integrations and support for existing numbers. Retell homepage, accessed 2026-08-18.

On compliance, Retell states SOC 2 Type 1 and Type 2, HIPAA with a signed BAA required before transmitting PHI, and GDPR via AWS infrastructure with a DPA and SCCs — while noting it does "not currently operate services within the European Union." Retell compliance docs, accessed 2026-08-18. Testing includes LLM simulation testing in the docs plus the paid AI quality assurance add-on noted above. Retell simulation testing docs, accessed 2026-08-18.

Vapi

Vapi's Build plan lists call hosting at $0.05/minute. Speech recognition, language models, and text-to-speech bill at provider cost on top — or show as $0 on Vapi's bill when you bring your own API keys, in which case the providers bill you directly. Telephony is separate. Ten concurrent calls are included, with extra lines at $10 each per month; the Scale plan is enterprise pricing with a fixed platform fee and committed volume. Compliance is an add-on: HIPAA at $2,000/month and zero data retention at $1,000/month (Vapi's docs note HIPAA mode and zero data retention are mutually exclusive). The key point for any comparison: $0.05/minute is the hosting portion of a call price, not the call price. Vapi pricing, accessed 2026-08-18.

Vapi publishes no platform-wide language total; its docs list per-provider ceilings (Deepgram "100+" for transcription, Azure "140+" for text-to-speech, and so on), with automatic language detection limited to specific transcription providers. Vapi multilingual docs, accessed 2026-08-18. Telephony supports bring-your-own SIP trunks (documented for Plivo, Telnyx, and Zadarma plus a generic any-provider path) and Twilio number import; free Vapi numbers are US-only, up to 5 per account. Vapi SIP docs and phone calling docs, accessed 2026-08-18.

Vapi's homepage states SOC 2, HIPAA, and PCI compliance; GDPR is not claimed on the pages reviewed. Vapi homepage, accessed 2026-08-18. Testing is a strength: Test Suites place real phone calls (or faster chat tests) against your agent following scripted scenarios, with an LLM judging transcripts against your criteria — up to 50 cases per suite, 5 runs each, 15 minutes per test, and test calls bill like regular calls. Vapi Test Suites docs, accessed 2026-08-18.

Bland AI

Bland prices an all-inclusive talk minute — the rate covers the language model, speech recognition, and text-to-speech, with no token charges or model-provider pass-throughs — in tiers with platform fees and operational limits. Start is $0.14/minute with no platform fee, 100 calls/day, and 10 concurrent calls; Build is $0.12/minute plus $299/month (2,000 calls/day, 50 concurrent); Scale is $0.11/minute plus $499/month (5,000 calls/day, 100 concurrent); Enterprise is custom. Transfer minutes bill separately at $0.05/$0.04/$0.03 by tier, waived for bring-your-own-telephony customers, and built-in telephony is Twilio at pass-through cost. Bland pricing, accessed 2026-08-18.

Bland's homepage states "40+ languages natively, with real-time translation available in 23 of them." Bland homepage, accessed 2026-08-18. Telephony is bring-your-own via Twilio or SIP trunks, or Bland's built-in Twilio at pass-through. Bland pricing, accessed 2026-08-18.

Compliance is Bland's chosen ground: SOC 2 Type II and PCI DSS v4.0 are third-party audited, while HIPAA (with BAA and real-time PII redaction) and GDPR (with DPA and configurable EU data residency) are self-attested on its trust page; data residency is listed as Enterprise-only. Bland also emphasizes proprietary in-house models — "zero AI sub-processors in the call path" — with self-hosted, VPC, on-premise, and air-gapped deployment options. Bland trust page, accessed 2026-08-18. Testing appears on the homepage as real-time call watching for QA and scenario testing with pass/fail tracking; detailed public docs on test billing or CI hooks were not established from the reviewed public sources. Bland homepage, accessed 2026-08-18.

ElevenLabs Agents

ElevenLabs is first a voice-technology vendor, and its Agents platform is priced as subscription tiers with included call minutes: Free (15 minutes), Starter $6/month (75), Creator $22 (275), Pro $99 (1,238), Scale $299 (3,738), and Business $990 (12,375), with concurrency rising by tier from 4 to 40. Overage minutes cost $0.08 on every self-serve tier — the included allotments price out at almost exactly $0.08/minute, so tier choice mainly buys concurrency and features, not a cheaper minute. Crucially, the pricing page states the LLM and any telephony "are billed separately on top, based on usage," with external telephony "at cost." ElevenLabs Agents pricing, accessed 2026-08-19.

Agents support 31 languages when "All" additional languages are enabled, drawing on 5k+ voices; language selection is fixed for the duration of a call by default, but an opt-in language-detection system tool lets the agent switch mid-call. ElevenLabs language docs, accessed 2026-08-19. Telephony includes SIP trunk integration (TLS transport, digest or ACL authentication) and native Twilio integration, plus batch outbound calling. ElevenLabs SIP docs and batch calling docs, accessed 2026-08-19.

The enterprise page states "certified SOC2 and GDPR compliant," an optional Zero Retention Mode, BAAs "with HIPAA compliant configurations for qualifying enterprises," and regional data residency options including the US, EU, and India; note Zero Retention Mode cannot be enabled for batch calls. ElevenLabs enterprise page and batch calling docs, accessed 2026-08-19. Testing is well developed: simulation testing with a simulated user, next-reply and tool-call tests, LLM evaluators, CI via CLI and API, and probabilistic pass rates, plus live-traffic A/B experimentation. ElevenLabs agent testing docs, accessed 2026-08-19.

At a glance

The 10,000-minute row uses one workload: 2,000 calls × 5 connected minutes, peaking at no more than 10 simultaneous calls, one primary language, no add-ons. Illustrative arithmetic from public pages, not quotes.

Sources: the per-platform profile sections above; Retell/Vapi/Bland accessed 2026-08-18, ElevenLabs accessed 2026-08-19.

How to choose

You want to pick every component yourself. Retell and Vapi both expose the model and voice layers as choices. Retell publishes per-component prices that sum toward its advertised range; Vapi goes further and lets you bring your own provider API keys, so the platform bills hosting only and the providers bill you directly. The trade-off on both is that your true per-minute cost is configuration-dependent, and on Vapi it is not computable from the public page at all.

You want one predictable number per minute. Bland's all-inclusive talk minute is the most legible headline rate in this group — but price the tier system honestly: at 10,000 minutes, Build and Scale cost more than Start ($1,499 and $1,599 versus $1,400) because their platform fees outweigh their lower rates; what they buy is daily-call and concurrency headroom.

Voice quality and voice variety lead your requirements. ElevenLabs Agents sits on the vendor's own speech models with 5k+ voices, and its testing and A/B tooling is unusually complete. Budget for the split bill: roughly $0.08/minute to ElevenLabs, plus LLM usage and telephony billed separately at cost.

Compliance drives the purchase. Compare postures precisely, not badge counts. Bland has audited SOC 2 Type II and PCI plus self-hosting options; Retell documents HIPAA BAAs and a GDPR DPA but notes no EU operations; Vapi prices HIPAA and zero data retention as monthly add-ons; ElevenLabs claims SOC2 and GDPR with enterprise BAAs and EU residency. Which of these satisfies your obligations is a question for your counsel, not a spec sheet.

High or spiky concurrency. Included concurrency ranges from 4 (ElevenLabs Free) to 20 (Retell), with overage priced per line ($8 Retell, $10 Vapi) or gated by tier (Bland, ElevenLabs). A traffic spike can cost more in tier upgrades than in minutes.

Where ThunderPhone fits

ThunderPhone — the publisher of this page — is a fifth option in the same category, and here is its published shape, stated the same way the other four were.

Pricing is pay as you go with no subscription or seat fee, as an all-in engine rate: Spark 2¢/minute, Bolt 5¢/minute, Storm 9¢/minute, covering speech recognition, the language model, and the voice in one number. Premium voice or language paths can add up to 3¢/minute, and optional add-ons carry published per-minute surcharges. On the same 10,000-minute scenario, the arithmetic is $200, $500, or $900 all-in by engine — below or inside the ranges above, though telephony is setup-dependent and different configurations can change any ordering. The catalog spans 47 languages and 123 voices, with automatic mid-call switching to configured additional languages. Telephony supports SIP trunks and bring-your-own numbers through direct VoIP connections or manual SIP configuration. Testing includes AI-caller simulations, reusable graded scenarios, regression suites with minimum pass-rate gates, CI execution, and live-traffic A/B experiments; simulations are billable real calls, with the charge shown before a run. The public compliance posture is GDPR and HIPAA compliant, and customers remain responsible for configuring their workflows, data handling, recordings, integrations, and notices for their own obligations.

That is the extent of the claim: a simpler bill and a broad language catalog, verifiable the same way you should verify everything else on this page — against the live product.

FAQ

Which is cheapest for 10,000 minutes a month: Retell, Vapi, Bland, or ElevenLabs?

No single answer is computable from public pages alone. The knowable figures for the defined scenario: Retell $700–$3,100 advertised all-in; Vapi $500 hosting only, before pass-through provider and telephony costs; Bland $1,400 on Start within its limits; ElevenLabs ≈$800–$990 for the platform layer, before separately billed LLM and telephony. ThunderPhone's published arithmetic is $200/$500/$900 all-in by engine. Rerun with your own call length and configuration.

Is Vapi really $0.05 per minute?

That is the hosting fee, not the call price. Speech recognition, the language model, and text-to-speech bill at provider cost on top — or bill you directly under your own API keys — and telephony is separate. Vapi pricing, accessed 2026-08-18.

Which platform supports the most languages?

No clean winner, because the counts measure different things. Bland states 40+ native languages; ElevenLabs Agents supports 31; Retell and Vapi publish per-provider counts rather than platform totals, so their effective coverage depends on the provider combination you configure. ThunderPhone's catalog lists 47 languages and 123 voices.

Can these agents switch languages mid-call?

Where documented, differently. ElevenLabs fixes language per call by default but offers an opt-in language-detection tool for mid-call switching; Vapi's automatic language detection depends on the transcription provider; ThunderPhone agents can switch automatically to configured additional languages. Mid-call switching behavior for Retell and Bland was not established from the reviewed public sources.

Which platforms are HIPAA compliant?

All five publish a HIPAA posture, with different mechanics. Retell requires a signed BAA before PHI; Vapi prices HIPAA mode at $2,000/month; Bland self-attests HIPAA with a BAA and PII redaction; ElevenLabs signs BAAs for qualifying enterprises; ThunderPhone's public posture is GDPR and HIPAA compliant with customer configuration responsibility. Have counsel review the specific mechanism against your obligations.

Do test and simulation calls cost money?

Usually, yes. Vapi's docs state test-suite calls bill like regular calls, and ThunderPhone's simulations are billable real calls with the charge shown before a run. Retell's AI quality assurance add-on bills $0.10/minute after 100 free minutes. Test-call billing for Bland and ElevenLabs was not established from the reviewed public sources — ask before running a large regression suite.

Sources and freshness

All third-party pricing and feature claims on this page were checked against the linked vendor pages on the accessed dates shown beside each claim. ThunderPhone pricing and product details reflect its published pricing and documentation as of August 19, 2026. This comparison is documentation-based — no vendor account, invoice, or live call was tested — and vendors change pricing often, so confirm against the live vendor pages before relying on them.