What is the cheapest AI voice agent platform per minute?

Among the bundled managed platforms, Retell AI is the lowest published rate at about $0.07/min all-in, followed by Deepgram Voice Agent (~$0.075) and Vapi (~$0.08 with a standard model stack). If you can self-host, LiveKit Agents has no platform fee — you pay only your own STT/LLM/TTS (~$0.03/min) plus infrastructure — but it takes the most engineering.

Vapi vs Retell AI — which one should I use?

Retell bundles the models into one ~$0.07/min rate and includes phone numbers, so it is cheaper and faster to launch for a standard support or receptionist bot. Vapi charges a ~$0.05/min platform fee and lets you pick your own STT, LLM and TTS at cost, which is better when you want a specific model or already have provider credits. This tool prices both side by side for your minutes.

How much does an AI voice agent cost per month for 2,000 minutes?

At 2,000 minutes, Retell is about $140/month (2,000 × $0.07), Vapi and Synthflow about $160 (× $0.08), Bland about $180 (× $0.09), and ElevenLabs or Play.ai about $200 (× $0.10). Self-hosting on LiveKit is roughly $60/month in model costs plus your server bill. Enter your own minutes above for exact figures.

Can I self-host an AI voice agent with open source?

Yes — LiveKit Agents is an open-source framework you can run on your own servers with a $0 platform fee. You still pay for the STT, LLM and TTS you plug in (about $0.03/min here) plus hosting. The managed platforms (Vapi, Retell, Bland, ElevenLabs, Synthflow, Play.ai, Deepgram) are cloud-only. Toggle "Self-host required" above to rank only the self-hostable option.

Which platforms include a phone number and which need Twilio?

Retell, Bland, Vapi and Synthflow can provision phone numbers for you (built-in telephony). Deepgram Voice Agent, ElevenLabs Agents, Play.ai and LiveKit expect you to bring your own Twilio or SIP number. Set the telephony toggle to "Built-in number" to keep only the platforms that handle the number for you.

Do these platforms support Sinhala or Tamil?

Tamil is documented on the broadly multilingual stacks (ElevenLabs, Deepgram, Retell, Vapi, LiveKit). Sinhala is far narrower: it is only reachable on the pass-through platforms — Vapi and LiveKit — by plugging in OpenAI Whisper for speech-to-text, since Whisper is one of the few engines that transcribes Sinhala. Bundled platforms do not document Sinhala. Select a language above to see who is de-ranked.

How are the per-minute costs normalised so they're comparable?

Two pricing shapes exist. Bundled platforms publish one all-in per-minute rate that already includes their default STT + LLM + TTS. Platform-fee platforms (Vapi, LiveKit) charge a thin fee plus your own models at cost, so we add a single labelled model stack (~$0.03/min: Deepgram STT + GPT-4o-mini + an economical TTS) to make the totals line up. The assumption is shown next to every result.

Are these prices live, and how often are they checked?

No — vendors publish no pricing API, so this is a hand-verified static dataset last checked on 2026-07-17. Each figure links to the vendor's own pricing page. Provider prices move often and enterprise deals differ, so treat the numbers as a starting estimate and open the source before you quote a client.

Does this replace the AI Voice Agent Cost Calculator?

No — they answer different questions. This comparison ranks whole platforms by their headline per-minute rate for your volume. The AI Voice Agent Cost Calculator models one STT→LLM→TTS pipeline turn-by-turn (tokens, characters, telephony) when you want to see exactly where the money goes. Use this to pick a vendor, then that tool to tune the stack.

AI · Voice agents

AI Voice Agent Platform Comparison — cost, telephony & self-host

Compare eight AI voice-agent platforms — Vapi, Retell, Bland, ElevenLabs, LiveKit, Synthflow, Play.ai and Deepgram — on estimated monthly cost, built-in vs. bring-your-own telephony, self-host support and language coverage. Enter your minutes and constraints; the tool ranks them and names a best fit. No signup, sources cited.

By Induwara Ashinsana— Executive Director, Ryzera TechnologiesUpdated Jul 17, 2026

Compare voice agent platforms

Use case

Inbound help desk / FAQ answering with call handoff.

Expected monthly minutes

Quick set

Total conversation minutes across all calls per month.

Telephony

Should the platform provide the phone number, or will you connect a Twilio/SIP number yourself?

Self-host / open source

Require an open-source stack you can run on your own servers?

Priority languages

Platforms that don't document a selected language are de-ranked and badged. Sinhala is only reachable on the pass-through stacks (Vapi, LiveKit) via a Whisper STT model.

Best fit for your inputs

Retell AI

Managed voice engine with built-in numbers; dashboard + API.

$140.00

per month · $0.070/min

≈ Rs 42,000 at Rs 300/$

Cheapest platform that meets every hard constraint (telephony, self-host, language). Bundled voice engine at ~$0.07/min. A premium LLM or per-country telephony can add a small pass-through on top.

All platforms, ranked

Retell AIBundled
$140.00/mo
$0.070/min
VapiPlatform fee
$160.00/mo
$0.080/min
SynthflowBundled
$160.00/mo
$0.080/min
Bland AIBundled
$180.00/mo
$0.090/min
5
LiveKit AgentsPlatform feeSelf-host
- Bring your own number (Twilio/SIP)
$60.00/mo
$0.030/min
6
Deepgram Voice AgentBundled
- Bring your own number (Twilio/SIP)
$150.00/mo
$0.075/min
7
ElevenLabs AgentsBundled
- Bring your own number (Twilio/SIP)
$200.00/mo
$0.100/min
8
Play.aiBundled
- Bring your own number (Twilio/SIP)
$200.00/mo
$0.100/min

Feature matrix

Platform	Effective $/min	Builder	Concurrency	Latency (published)	Languages
Retell AI	$0.070	No-code + SDK	20 concurrent (scales on request)	Low (~800ms class)	English, Spanish, Hindi, Tamil, Other
Deepgram Voice Agent	$0.075	SDK-first	High (API concurrency limits)	Ultra-low (sub-500ms class)	English, Spanish, Hindi, Tamil, Other
Vapi	$0.080	No-code + SDK	10 concurrent default (scales)	Low (~800ms class)	English, Spanish, Hindi, Tamil, Sinhala, Other
Synthflow	$0.080	No-code	Standard (plan-tiered)	Standard	English, Spanish, Hindi
Bland AI	$0.090	No-code + SDK	High (enterprise concurrency)	Low	English, Spanish, Hindi
ElevenLabs Agents	$0.100	No-code + SDK	Tier-dependent	Low (Flash TTS)	English, Spanish, Hindi, Tamil, Other
Play.ai	$0.100	No-code + SDK	Standard	Standard	English, Spanish, Hindi
LiveKit Agents	$0.030	SDK-first	Scales (your own infra)	Low (you tune it)	English, Spanish, Hindi, Tamil, Sinhala, Other

"Effective $/min" normalises platform-fee stacks (Vapi, LiveKit) by adding a labelled ~$0.03/min model stack so every row is comparable.

Bundled = STT + LLM + TTS included in the per-minute ratePlatform fee = thin fee + your own models at costBest fit = cheapest option meeting every hard constraint

Sources cited: rates come from each vendor's published pricing page (linked in the feature matrix and the Sources section below), transcribed on 2026-07-17. Costs are estimates for the platform's default stack; premium models, per-country telephony, and enterprise discounts can move the real figure. This is a decision aid, not a quote.

How it works

Every platform is reduced to one comparable number — an effective dollars-per-minute rate — and your monthly cost is simply that rate times the minutes you enter. The hard part is that vendors price in two different shapes, so the tool normalises them before ranking:

Bundled platforms (Retell, Bland, Synthflow, ElevenLabs Agents, Play.ai, Deepgram Voice Agent) publish a single per-minute rate that already includes their default speech-to-text, language model and text-to-speech. The effective rate is that published number.
Platform-fee platforms (Vapi, LiveKit Agents) charge a thin per-minute fee and pass your chosen STT/LLM/TTS through at cost. To compare fairly, the tool adds one labelled model stack — Deepgram speech-to-text + GPT-4o-mini + an economical voice, about $0.03/min. So Vapi is priced as its ~$0.05 platform fee plus ~$0.03 = ~$0.08/min. The assumption is shown next to the result, and you can model the stack precisely in the linked cost calculator.
Monthly cost = effective $/min × monthly minutes. Deterministic — no hidden weighting.
Hard constraintsdecide which platforms "qualify". If you need a built-in phone number, platforms that only support bring-your-own telephony are de-ranked and badged. If you require self-hosting, only the open-source stack (LiveKit) qualifies. If a selected language is not documented, that platform drops down with a badge.
Best fitis the cheapest platform that satisfies every hard constraint — a transparent rule, not an opinion score. Ties in cost are broken by the vendor's published concurrency headroom.

No latency or accuracy "rankings" are invented. The feature matrix shows only vendor-published figures — latency and concurrency classes are the vendors' own claims, each row links to its pricing page, and every rate carries the 2026-07-17 verification date. Because vendors change prices without notice and offer enterprise discounts, treat the output as a well-sourced estimate to narrow your shortlist, then confirm at the source.

Worked examples

Restaurant reservation bot — 1,500 min/mo, built-in number, English

Constraint: built-in telephony required → keep Retell, Vapi, Synthflow, Bland.
Retell (bundled): 1,500 × $0.07 = $105/mo
Vapi (0.05 fee + 0.03 stack = 0.08): 1,500 × $0.08 = $120/mo
Synthflow (bundled): 1,500 × $0.08 = $120/mo
Bland (bundled): 1,500 × $0.09 = $135/mo
Best fit = Retell at $105/mo (cheapest with a built-in number).

Developer prototype — 500 min/mo, BYO number, self-host required

Constraint: self-host required → only LiveKit Agents qualifies.
LiveKit: $0 platform fee + ~$0.03/min your own models.
500 × $0.03 = $15/mo in model cost, plus your server bill.
Turn self-host off and the cheapest managed option is Retell:
500 × $0.07 = $35/mo (Vapi 500 × $0.08 = $40/mo).

High-volume support line — 20,000 min/mo, built-in number, English + Tamil

Constraint: built-in telephony + Tamil support.
Retell documents Tamil and has built-in numbers → qualifies.
Retell: 20,000 × $0.07 = $1,400/mo
Vapi: 20,000 × $0.08 = $1,600/mo (also Tamil-capable via its model stack)
Bland: 20,000 × $0.09 = $1,800/mo, but no documented Tamil → de-ranked.
Best fit = Retell at $1,400/mo.

Platforms compared

Eight platforms are in the dataset, each linked to its official pricing page. Rates are the vendors' published headline figures for their default stack, verified 2026-07-17.

Retell AI
Managed voice engine with built-in numbers; dashboard + API.
Deepgram Voice Agent
Single low-latency API; SDK-first, bring your own telephony.
Vapi
Orchestrator with number provisioning; pick your own models.
Synthflow
No-code visual builder with built-in numbers and integrations.
Bland AI
Vertically-integrated all-in-one with built-in numbers.
ElevenLabs Agents
Most natural voices; agent runtime with 30+ languages.
Play.ai
Play.ht-based voices with an agent layer and API.
LiveKit Agents
Open-source, self-hostable, SIP telephony; most engineering.

Frequently asked questions

Sources & references

Rates were transcribed from these pages on 2026-07-17. Vendor pricing changes without notice and enterprise/custom quotes differ — re-check the source before committing. This tool makes no live pricing calls and stores no personal data.

Related tools

LiveAI

AI Voice Agent Cost

Estimate the real per-minute and monthly cost of a cascaded AI voice agent — chaining speech-to-text, an LLM, text-to-speech, optional Twilio telephony and a Vapi/Retell platform fee. Breaks the blended cost down by layer so you see which dominates, in USD and LKR. Runs in your browser, rates cited.

Open tool

LiveAI

AI Reasoning Model Compare

Side-by-side comparison of the major AI reasoning models — OpenAI o3 & GPT-5 thinking, Claude extended thinking, Gemini thinking, DeepSeek-R1, Grok 4 and Qwen3 — by reasoning-token pricing, context window, knowledge cutoff and AIME / GPQA / SWE-bench scores. Sort, filter, pick by use case, and estimate a task's true cost once hidden thinking tokens are counted. Sources cited, no signup.

Open tool

LiveAI

Text-to-Speech Compare

Side-by-side comparison of the major hosted text-to-speech (TTS) APIs — ElevenLabs, OpenAI, Google Cloud, Azure AI Speech, Amazon Polly, PlayHT, Murf, Cartesia and Deepgram Aura — by price per character, voice cloning, streaming latency, language coverage, output formats, commercial-use terms and published naturalness. Pick your providers, enter your monthly volume in characters, words or minutes, and rank them by projected cost. Every figure cites the vendor source.

Open tool

Rate this tool

Be the first to rate

Comments & feedback

Spotted a bug or want an improvement? Tell us — our team reviews every comment, and good ideas get built. Comments are public and anonymous.

Spotted a stale rate, a missing platform, or an edge case?

Email me at [email protected] — most fixes ship within 24 hours.