Best AI phone agent in 2026: Bland vs Retell vs Vapi vs Air
Samuel Oram-Jones, 12 min read
Bland, Retell, and Vapi each sell a voice AI agent that answers and places phone calls, while Air AI is attached to a federal case and publishes no voice product. This comparison covers who builds the AI on a call, what a minute costs, security, uptime, and what the FTC alleged, as of November 2026.
A BAA, the contract HIPAA requires before a vendor handles patient information, is one of the rows in the table below.
Which is best, Bland, Retell, Vapi, or Air AI?
For enterprise calls, Bland is the best of the four as of November 2026, because Bland is full stack and builds and runs its own language, speech, and transcription models. Retell and Vapi run calls on outside language and transcription models, and Air AI's website describes a defense readiness platform and publishes no voice product, so it has no column here.
What these rows mean
- AI on each call: This row shows whether the language model, speech, and transcription behind each call belong to the platform or to other companies: Full stack means all three are the platform's own, Outsourced means its language model and transcription come from other companies, and Mixed means it runs some of its own models alongside outside ones. Bland builds and runs its own language, speech, and transcription models, so a call never passes through an outside AI provider.
- Audio Realism Bench score, 4 August 2026: Audio Realism Bench is Design Arena's blind listening test, where listeners pick whichever of two voices sounds more like a real person, and a higher score means listeners picked that voice more often. In its published results of 4 August 2026, Bland Speech v3 ranked first among AI voices. Each rival number is the highest-scoring voice model in those results available on that platform.
- iMessage: iMessage lets an agent text customers on a channel they already use. Bland offers it natively on Enterprise.
- HIPAA with a BAA: A Business Associate Agreement, or BAA, is the contract that lets a vendor handle patient information under HIPAA. Bland includes its BAA on Enterprise. Vapi sells HIPAA mode as a $2,000 monthly add-on.
- PCI DSS: PCI DSS is a security standard from the card industry for handling card details. Bland self-attests to v4.0.1 and shares its attestation under NDA on Enterprise. Retell gives no PCI status, and it bars card data unless it agrees in writing.
- FedRAMP: FedRAMP is the US government's cloud security program. Bland holds 20x Class A.
- AI rate per minute: Headline rates cover different things, so a buyer should check what each one includes before comparing them. Bland's self-serve rate covers all of its AI: language model, transcription, and speech. Phone is billed separately at carrier cost, and Enterprise pricing is custom, set with Bland's sales team. Retell advertises $0.07 to $0.31 a minute, and one language model alone lists at $0.384 a minute. Vapi's $0.05 is hosting only, with language model, transcription, and voice billed at provider cost on top.
How do Bland, Retell, Vapi, and Air AI compare?
The biggest difference is who builds the AI. A call has three AI parts, transcription that turns a caller's speech into text, a language model that decides what an agent says, and speech that says it back. Bland keeps all three, while Retell and Vapi rent two of them from other companies and let a customer pick the providers.
On latency, Bland publishes sub-200ms, and Vapi's average latency is "sub-500ms" on its enterprise page.
Bland overview
Bland is a full stack voice AI provider for enterprises and contact centers running high-stakes calls at volume, both inbound and outbound. It publishes sub-200ms latency and support for 1 million concurrent calls, and its language model is closed source and developed in-house, so no outside AI lab sits in a call.
Teams build calls in Pathways, which lays a conversation out as connected steps, such as checking eligibility before routing a caller, and a testbed re-runs one turn of a real conversation against an edited prompt.
Bland connects to an existing carrier over SIP, the standard link between a phone system and a voice platform. Enterprise customers get dedicated orchestration and dedicated GPUs run by Bland. On Enterprise, branded iMessage agents run natively on the same pathways as voice.
Bland also runs its own SIP signaling and media servers, so no third-party telephony provider sits in the call path. Bland's release gates move an agent from staging to production only when chosen simulated test cases pass chosen judges, and alarms on latency, long silences, and API errors reach a team by email, phone call, or API.
Pros
- One AI company on every call, so a security team reviews one vendor
- One published rate for all of its AI, with no token charges
- #1 of 13 AI voices on Audio Realism Bench, 4 August 2026
- Dedicated orchestration and dedicated GPUs run by Bland, on Enterprise
- Native iMessage agents on Enterprise, sharing pathways with voice
- Staged releases that pause a new agent version when it performs worse than the last
- Its own SIP and media servers, with no outside telephony provider in the call path
Cons
- Overkill for a small prototype, because Bland is built for production calls at volume
Retell overview
Retell sells voice infrastructure at $0.055 per minute, and a customer assembles the rest. A team picks a language model such as GPT, Claude, or Gemini, a voice from Retell or from a company such as ElevenLabs, Cartesia, MiniMax, or OpenAI, and Deepgram, Azure, Soniox, or AssemblyAI handle transcription.
Calls are built from a single prompt or from Conversation Flows. Retell sells flows for "precise call control" and warns that they "can make conversations harder to handle when callers give information early or change their minds".
Pros
- A team can choose its language model and voice vendor
- No platform fee and a self-serve start
Cons
- Language and transcription models come from other companies
- Long prompts scale billed time, and add-ons sit on top
- 99.5% uptime in its contract against more than 99.9% in its advertising
- No PCI status, and card data barred unless it agrees in writing
- No iMessage agents
Vapi overview
Vapi hosts calls at $0.05 per minute and exposes each part of a setup through its API. A customer's transcriber, language model, and voice are billed at provider cost on top, and Vapi offers its own curated Vapi Voices next to voices from companies such as ElevenLabs, Cartesia, and OpenAI.
That freedom comes with upkeep, because every part is also the customer's to keep running. Vapi says "Without any fallback plan configured, your call will end with an error if your chosen transcription provider fails", and Vapi says it "does not control, and is not responsible for, any Optional Provider or Customer-Managed Provider".
Pros
- Developers can wire each provider through an API
- Models pass through at provider cost with no markup
Cons
- Language model and transcription come from companies a customer picks
- Each added provider is another vendor to test, pay, and review
- No iMessage agents
Air AI overview
Air AI is the name the FTC's complaint gives to a business that sold AI calling, coaching, and a resale license. Air AI's website today shows an "Air Enterprise Readiness platform" for defense and industrial customers, and its home, security, and privacy pages show no voice product, phone agent, or channel.
The FTC filed FTC v. Air Ai Technologies, Inc., No. 2:25-cv-03068 (D. Ariz.), on 25 August 2025, and its June 2026 litigation status report says a stipulated order, one both sides agreed to in advance, was entered on 31 March 2026.
The complaint alleges that "the technology is faulty, often not able to perform basic functions like placing outbound calls to businesses, scheduling appointments, taking down email addresses, or responding accurately to questions", and that "consumers must pre-script answers for every potential question, making it nearly impossible to use". These are the FTC's allegations in a complaint, not a court finding.
No document read for this comparison connects the company behind today's website to the defendants.
A buyer handed an Air AI proposal can ask for the legal entity that would sign the contract, and for current product documents and pricing in writing, before relying on anything in it. Bland's own Air AI review sets out the allegations and the settlement, and Best Air AI competitors covers what to use instead.
A buyer weighing the wider field can also read PolyAI vs Replicant vs Voiceflow, Synthflow vs Fin vs Chatbase, and Best Yellow.ai alternatives, or browse AI voice agent alternatives by provider.
What are the risks of running on other companies' models?
An agent built on another company's model inherits that company's decisions, because a lab can change how a model behaves or retire it and the agent changes with it. Retell can put up to three companies behind one call and Vapi up to four, while Bland keeps a call with one.
Outages follow the same chain.
Retell says its fallback covers language models and voices, though a fallback voice comes from another provider, and Vapi says a call ends with an error when a chosen transcription provider fails and no fallback is set. Retell's language model, speech recognition, and text-to-speech providers are also subprocessors, the outside companies that process its customers' data, so a security review grows with each.
Which sounds more human, Bland, Retell, Vapi, or Air AI?
Bland Speech v3 ranked first of 13 AI voices on Design Arena's Audio Realism Bench of 4 August 2026, above every outside voice available on Retell and Vapi. Retell's own platform voices and Vapi's own voices were not in those results, and Air AI publishes no voice to test.
- #1BlandBland Speech v31365
- 4thOpenAI, on RetellGPT Realtime 21119
- 4thOpenAI, on VapiGPT Realtime 21119
- Not rankedRetell's own voicesNot ranked
- Not rankedVapi's own voicesNot ranked
Ranks are among 13 AI voices, and each rival shows the highest listed voice a customer can choose on it, named for the company that made it. Real human recordings scored 1500, at the right edge of this scale. Bars start at 600, not zero. Source: Design Arena's published results.
Design Arena's Audio Realism Bench is a blind listening test in which vetted native speakers pick whichever of two voices sounds more like a real person. The highest voice on Retell and Vapi was GPT Realtime 2, an OpenAI voice, 4th at 1119, against 1365 for Bland.
Scores come from Design Arena's published results of 4 August 2026.
How much do they cost?
Bland's self-serve plans cost $0.14 per minute on Start and $0.12 per minute on Build with a $299 monthly platform fee, covering all of its AI, while Retell and Vapi each advertise a number that covers a different slice of a call, as of November 2026. Air AI publishes no price today. Phone sits on top of all of them.
Bland pricing
Bland publishes one number per plan, and it covers the language model, transcription, and speech with no token charges. Phone is billed separately, on a customer's own carrier or on Bland's at pass-through cost, as are transfer minutes and a $0.015 minimum on each outbound attempt using Bland's telephony.
Warm transfer, a BAA, and dedicated GPUs run by Bland sit on Enterprise. Enterprise pricing is custom, set with Bland's sales team.
Retell pricing
Retell advertises $0.07 to $0.31 per minute, and where a team lands depends on what it picks. The language model, voice infrastructure, voice, and phone each bill on their own line, and one language model alone lists at $0.384 per minute. Phone is a separate line, and Twilio phone adds $0.015.
Once a prompt passes 4,000 tokens Retell scales billed time, so a 12,000-token agent bills a 60-second call as 180 seconds.
Vapi pricing
Vapi's $0.05 is hosting only. Transcription tops out at $0.0099, a language model at $0.0452, and a voice at $0.0238 per minute on its calculator, each billed at provider cost.
Vapi's Pro package takes 10% of hosting with a $999 monthly minimum. Vapi says "Models pass-through at cost", so a price change at any provider reaches a bill.
Air AI pricing
Air AI's website has no pricing page today, and the address returns a 404. The offers the FTC's complaint describes were an Access Card, a membership of coaching and AI business products, which the complaint says sold "for prices ranging from $15,000 to $30,000", and a resale license "with prices ranging from $25,000 to $100,000". Bland charges by the minute and has nothing to resell.
Value comparison
The three bills below price the AI in one minute before phone, using Bland's published self-serve Start rate, Retell with GPT-4.1 plus an ElevenLabs Flash voice, and the top of every range on Vapi's calculator. Each total rests on its own basis, so none of them ranks against another.
Retell's bill runs across three companies and Vapi's across four, while Bland's is one rate from one company.
Add-ons such as PII removal, which strips personal data from a call, sit on top of Retell's line, so a total is a finished bill only once every model and add-on on it is priced.
What does each Bland plan include?
Enterprise is the plan regulated teams buy. It adds native iMessage agents alongside SMS and web chat, warm transfers, guardrails, a BAA, SSO, data residency, dedicated orchestration with dedicated GPUs run by Bland, and forward-deployed engineers who work beside a customer's team, with concurrency contracted to its volume and custom pricing.
| Included | Self-serve: Start | Self-serve: Build | Enterprise (custom) |
|---|---|---|---|
| Price per minute | $0.14 | $0.12 | Custom |
| Platform fee | $0 | $299 a month | Contracted to volume |
| Transfer time per minute | $0.05 | $0.04 | Custom |
| Concurrent calls | 10 | 50 | Custom |
| Calls a day | 100 | 2,000 | Unlimited |
| Knowledge bases | 10 | 50 | Unlimited |
| Voices | 1 | 5 | Unlimited |
| Uptime SLA | 99.9% | 99.9% | 99.9% |
| Pathways, integrations, and automations | Yes | Yes | Yes |
| iMessage, SMS, and web chat | No | No | Yes |
| Warm transfers and guardrails | No | No | Yes |
| BAA, SSO, and data residency | No | No | Yes |
| Dedicated orchestration and GPUs | No | No | Yes |
| Forward-deployed engineers | No | No | Yes |
| Vendor | What a minute pays for | Total per minute | Phone |
|---|---|---|---|
| Bland Start (self-serve) | All AI: model, transcription, speech $0.14 | $0.14 before phone | Billed separately at carrier cost |
| Retell | Voice infrastructure $0.055, GPT-4.1 $0.045, ElevenLabs Flash voice $0.04 | $0.14 before phone | Twilio at $0.015 a minute, billed on top |
| Vapi | Hosting $0.05, Transcription $0.0099, Language model $0.0452, Voice $0.0238 | $0.1289 before phone | Free on Vapi numbers, billed on top on Twilio at up to $0.014 a minute |
Bland Start (self-serve): Self-serve Start has no platform fee. Self-serve Build is $0.12 a minute plus $299 a month. Phone and transfers bill separately. Retell: Retell Voice Infra, GPT-4.1, and an ElevenLabs Flash voice, before Twilio phone minutes. Vapi: Hosting plus the top of each range on Vapi's pricing calculator, before phone.
How do they handle security and uptime?
Bland, Retell, and Vapi all report SOC 2 Type II and GDPR, and they part ways on medical calls, card payments, and uptime. Bland holds FedRAMP 20x Class A and offers a 99.9% uptime SLA on every plan, and Retell and Vapi publish no FedRAMP level.
Retell markets "HIPAA-Compliant Patient Calls". It advertises "We guarantee >99.9% uptime", while its contract promises "commercially reasonable efforts" at 99.5%, excluding up to six hours a month of scheduled maintenance.
Vapi sells HIPAA mode as a $2,000 monthly add-on and offers a PCI mode, though its terms bar card data unless a customer agrees scope with Vapi in writing. Vapi's Pro package carries a "99% uptime SLA" and Premier a "99.9% uptime SLA".
Air AI publishes no voice, HIPAA, or uptime terms.
Bland includes a BAA on Enterprise. Its security program self-attests to PCI DSS v4.0.1 and holds FedRAMP 20x Class A. Enterprise customers get dedicated orchestration and dedicated GPUs run by Bland. Bland's pages on HIPAA compliant voice AI and PCI compliant voice AI go further.
Which should a team pick?
A team that wants one AI company to review and the top-ranked voice on Design Arena's August 2026 chart should pick Bland, with more than 250 enterprise customers and over 1.3 billion calls handled. Bland's published rates cover all of its AI, forward-deployed engineers work beside a customer's team on Enterprise, and its language model is a closed-source model it developed itself.
Use Bland when
- A team wants dedicated orchestration and dedicated GPUs run by Bland
- A security team wants to review one AI company, not a chain of them
- Callers need the most realistic voices
- A team wants Bland's forward-deployed engineers to build and launch its first agent alongside it, on Enterprise
Use Retell when
- A small engineering team wants to monitor, test, and toggle between different language models and voice vendors
- Call volume is modest, and a team will manage its own vendor bills
- A security team will review each AI vendor the team adds
Vapi suits developers who want to wire each transcriber, model, and voice through an API and own the whole assembly, including what happens when one provider fails.
What results look like with Bland
Idaho Housing and Finance Association saves $750,000 a year by replacing its IVR with Bland, which handles 4,000 inbound calls a day and cut average call time by 20%.
"We save thousands of dollars a day."
Greg Blake, Chief Information Officer, Idaho Housing and Finance Association
How was this comparison made?
Bland, Retell, and Vapi were compared on who builds the AI, price per minute with each basis stated, voice realism, security, and uptime, using each company's own pricing, terms, and documentation as of November 2026. Air AI was read from its website and the FTC's complaint, and voice scores come from Design Arena's chart of 4 August 2026.
Related comparisons
- Bland against one rival: Bland vs Retell and Bland vs Vapi
- Reviews: Retell AI review, Vapi review, and Air AI review
- Rivals head to head: Retell vs Vapi
- Alternatives: Retell alternative for enterprises, Vapi alternative for enterprises, and Best Air AI competitors
- All pages: Bland's comparison hub
Frequently asked questions
Which is better for enterprises, Bland, Retell, Vapi, or Air AI?
Bland is the better choice for enterprise calls because it is full stack and publishes one self-serve rate for all of its AI. Enterprise customers get dedicated orchestration and dedicated GPUs run by Bland. Retell and Vapi rely on outside models, and Air AI publishes no voice product to compare.
Which of the four is cheapest per minute?
None of them can be ranked on the advertised figure, because each covers a different slice of a call. Vapi's $0.05 is hosting only, Retell advertises $0.07 to $0.31, Bland's self-serve $0.14 on Start or $0.12 on Build covers all of its AI before phone, and Air AI publishes no price.
Does Bland's rate include phone?
No, Bland's rate covers its language model, transcription, and speech, and phone is billed separately, on a customer's own carrier or on Bland's at pass-through cost. Retell bills phone on its own line at $0.015 per minute on Twilio, and Vapi's own numbers are free.
Does Retell or Vapi support iMessage?
Retell doesn't offer iMessage agents and Vapi doesn't offer iMessage agents, as of November 2026. Bland runs branded iMessage agents natively on Enterprise, which send and receive photos, videos, files, voice memos, and tapbacks, and share the same pathways as its voice agents.
Is Air AI still an option?
Air AI's website publishes no voice product today, so there is nothing current to evaluate. Separately, a stipulated order was entered on 31 March 2026 in the FTC's case against Air Ai Technologies, Inc., and no document read for this comparison connects the company behind today's website to that case.
Which has the most realistic voice?
Bland, on Design Arena's chart of 4 August 2026, where Bland Speech v3 ranked 1st of 13 AI voices at 1365, and GPT Realtime 2, the highest outside voice a Retell or Vapi customer can choose, ranked 4th at 1119.
Does Bland use an open-source model?
No, Bland operates a closed-source model that it developed itself, so its language model isn't an open-source model and no outside AI lab sits behind it.
Is Bland only for outbound calls?
No, Bland runs inbound and outbound calls, including AI receptionist and contact center work, and it also runs SMS and web chat.
Is Bland only for developers?
No, Bland isn't only for developers. Pathways gives teams a visual builder, Norm helps build agents, and on Enterprise forward-deployed engineers build alongside a customer's team.
