Best AI Voice Agent Platforms in 2026: Vapi vs Retell vs Bland vs ElevenLabs Agents

Phone calls are the last customer channel that software never fully absorbed, and an AI agent that answers them has to chain speech-to-text, an LLM, text-to-speech and telephony inside roughly a second of latency, then survive interruptions, hold music, and a caller spelling a policy number. Somebody has to host that pipeline and bill for it, and the way each vendor bills is the real difference between the platforms.

As of September 5, 2026, Vapi charges $0.05 per minute for hosting and passes model costs through at cost, Retell's own estimator lands at $0.115 per minute, Bland charges $0.14 per minute all-in with no platform fee, and ElevenLabs lists its agent speech engine at $0.08 per minute before the LLM. The money says this is where 2026 venture budgets went: Vapi hit a $500 million valuation in May after Amazon Ring evaluated more than 40 voice vendors and routed 100% of its inbound calls through Vapi, ElevenLabs raised $500 million at an $11 billion valuation in February on $330 million of ARR, and Deepgram closed $130 million at $1.3 billion in January. Between July 15 and July 31 alone, TechCrunch covered four more voice raises: Rime ($24 million), Fish Audio ($52 million), Encore AI ($30 million) and Smallest.ai ($13 million).

Quick Comparison

The billing unit is the decision that compounds. Vapi and Pipecat sell compute and let you bring model keys, Retell and LiveKit meter the model minutes for you, and Bland folds LLM, STT and TTS into one rate.

Vapi

Vapi is the developer-first orchestrator: define an assistant (model, voice, transcriber, tools), attach a phone number, and place or receive calls through an API. The Build plan is usage-based with no contract. Hosting is $0.05 per minute for calls and $0.005 per message for SMS and chat. Speech-to-text, LLM and text-to-speech are billed "at cost," and drop to $0 on Vapi's side if you bring your own provider API keys. Ten concurrent calls are included and each additional line is $10 per month. Call history is kept 14 days on Build.

The verified server SDK shape (@vapi-ai/server-sdk 1.2.0):

TypeScript
Copy
import { VapiClient } from "@vapi-ai/server-sdk";

const vapi = new VapiClient({ token: process.env.VAPI_API_KEY! });

const assistant = await vapi.assistants.create({
  name: "Appointment reminders",
  firstMessage: "Hi, this is the clinic calling to confirm your appointment.",
  model: {
    provider: "openai",
    model: "gpt-4o-mini",
    messages: [{ role: "system", content: "Confirm, reschedule, or cancel. Keep answers short." }],
  },
  voice: { provider: "11labs", voiceId: "burt" },
  transcriber: { provider: "deepgram", model: "nova-3" },
});

const call = await vapi.calls.create({
  assistantId: assistant.id,
  phoneNumberId: process.env.VAPI_PHONE_NUMBER_ID!,
  customer: { number: "+14155550123" },
});

What it does well. The traction is real: over 1 billion calls processed, between 1 million and 5 million calls a day, more than 1 million developers, and enterprise names including Amazon Ring, New York Life and Intuit, per TechCrunch's May 12, 2026 report on the $50 million Series B led by Peak XV.

What to watch. Compliance is priced as add-ons: HIPAA is $2,000 per month and Zero Data Retention is $1,000 per month on both plans. SOC 2, SSO and RBAC only arrive on the annual Scale contract. A hobby project pays $0.05 a minute; a healthcare deployment pays $2,000 a month for the HIPAA add-on before the first call, $3,000 if it also wants Zero Data Retention.

Pricing: Build usage-based, $0.05/min hosting · models at cost or $0 BYOK · 10 concurrent + $10/line/mo · Scale annual contract

Retell AI

Retell AI sells the opposite promise from Vapi: no keys to manage, every component metered by Retell. The published range for voice agents is $0.07 to $0.31 per minute, and the detailed component table explains the spread. Voice infrastructure is $0.055 per minute. Text-to-speech is $0.015 per minute for Retell's own voices and for Minimax, Fish, Cartesia and OpenAI voices, or $0.040 per minute for ElevenLabs voices. The LLM line is where the range comes from: GPT-5 nano is $0.003 per minute, GPT-5 mini $0.012, GPT-4.1 $0.045, Claude 5 Sonnet $0.08, and GPT-5.5 $0.16, or $0.32 on the fast tier. The estimator's default stack (GPT-4.1, Retell voice, no telephony) prints $0.115 per minute.

New accounts get $10 in credits and 20 concurrent calls, the most generous included concurrency of the fully metered platforms. The SDK is conventional (retell-sdk 5.64.0): client.call.createPhoneCall({ from_number, to_number, override_agent_id }) returns a call_id and call_status.

What to watch. SSO, role-based access control, a dedicated stable server and custom MSA, DPA and BAA terms are Enterprise-only, and Enterprise is custom pricing. The pay-as-you-go tier gets a shared server and community support.

Pricing: $0.07 to $0.31/min all-in · $0.002+/msg for chat agents · $10 free credits · 20 concurrent calls · Enterprise custom

Bland AI

Bland AI prices like a phone company: one talk-time rate. "No token charges. No model-provider pass-throughs. No surprise bills." Start is $0.14 per minute with no platform fee, 10 concurrent calls and 100 calls a day, plus two credits and an inbound number on signup. Build drops the rate to $0.12 per minute for a $299 monthly platform fee with 50 concurrent calls and 2,000 calls a day. Scale is $0.11 per minute for $499 a month, 100 concurrent calls and 5,000 calls a day. Transfer time (the minutes after a handoff to a human) is billed separately at $0.05, $0.04 or $0.03 per minute.

Bland's own FAQ notes that Vapi's $0.05 and Retell's roughly $0.07 are platform fees with model costs on top. The compliance story is the strongest of the self-serve platforms: SOC 2 Type I and Type II, HIPAA eligibility, GDPR, PCI DSS, and a FedRAMP certification banner on the pricing page as of this writing, though the signed BAA itself is an Enterprise-plan item. Enterprise adds on-prem or VPC deployment and forward-deployed engineers.

What to watch. The arithmetic. Build's $299 fee only beats Start above 14,950 minutes a month, and Bland's all-in rate is roughly three times Vapi's hosting fee, so a team running cheap models on its own keys pays for simplicity it may not want.

Pricing: Start $0.14/min, $0 fee · Build $0.12/min + $299/mo · Scale $0.11/min + $499/mo · Enterprise custom

ElevenLabs Agents

ElevenLabs Agents is the voice model company selling the whole agent. The self-serve plans bundle minutes: Free includes 15 minutes and 4 concurrent calls, Starter is $6 a month for 75 minutes, Creator is $22 a month (first month $11) for 275 minutes, Pro is $99 a month for 1,238 minutes and 20 concurrent calls, with Scale at $299 and Business at $990 above that. On the API price list, the Agents speech engine is $0.08 per minute, covering speech-to-text and text-to-speech "in one low-latency agent pipeline" across 70+ languages, and the platform is LLM-agnostic, so you bring your own model.

What it does well. If voice quality is the product, this is the shortest path to the voices customers already recognize, and the company behind it is the best capitalized in the category, with $781 million raised in total.

What to watch. The per-minute API price excludes the LLM, and minute bundles run out faster than they look on a real call center: 1,238 minutes on Pro is about 20 hours of talk time.

Pricing: Free 15 min · Starter $6/mo · Creator $22/mo · Pro $99/mo · API $0.08/min speech engine, LLM extra

LiveKit Agents

LiveKit Agents is the open-source route that still comes with a hosted bill if you want one. The framework is Apache 2.0 with 14,000 GitHub stars, ships for Python and Node, and includes turn detection, telephony integration and native MCP tool support. You can run it against the open-source LiveKit media server on your own hardware, or on LiveKit Cloud.

Cloud pricing is itemized per minute. Build is $0 a month with a free phone number and inference credits, Ship starts at $50, Scale starts at $500 and unlocks HIPAA and security reports. The pricing calculator's default phone-call stack reads: agent session $0.0100, US local number $0.0100, Gemma 4 31B $0.0014, Deepgram Nova-3 $0.0058, Cartesia Sonic 3 $0.0300, observability $0.0100, for a total of $0.0672 per minute. Swapping to GPT-5 mini drops the LLM line to $0.0011.

What to watch. You are assembling a pipeline, not configuring a product. Voice cloning, canary rollouts and analytics dashboards are yours to build or buy.

Pricing: Open source free · Cloud Build $0 · Ship from $50/mo · Scale from $500/mo · example stack $0.0672/min

Pipecat

Pipecat is the other open-source framework, BSD-2 licensed with 15,200 stars and maintained by Daily. Frames flow through Python processors for transport, STT, LLM and TTS, with OpenAI, Anthropic, Gemini, Deepgram, ElevenLabs, Cartesia and dozens more supported.

Pipecat Cloud hosts the agents at a compute rate that matches LiveKit's $0.01 session minute: the agent-1x profile (0.5 vCPU, 1 GB) is $0.01 per minute active and $0.0005 per minute reserved, with unlimited concurrency. WebRTC transport is free for one-to-one voice sessions, PSTN dial-in and dial-out is $0.018 per minute, SIP is $0.003 to $0.02 per minute, and Krisp noise cancellation is included up to 10,000 session minutes a month. Every model provider is billed to you directly.

What to watch. Same tradeoff as LiveKit, and telephony is one of several transports rather than the default.

Pricing: Open source free · Cloud agent-1x $0.01/min active · PSTN $0.018/min · providers billed to you

For teams that do not want to build at all, Synthflow is the enterprise-only option; its pricing page states that contracts start at $30,000 annually.

Do the Math on Your Own Minutes

Every platform quotes a per-minute number that hides a different bundle, so price your actual volume. This script uses the rates above; run it with your monthly talk minutes.

JavaScript
Copy
// voice-agent-cost.mjs - what a month of AI phone calls costs on each platform.
// Usage: node voice-agent-cost.mjs 20000   (minutes of talk time per month)
const minutes = Number(process.argv[2] ?? 20_000)

// Published rates, read from each vendor's pricing page on September 5, 2026.
const PLATFORMS = [
  { name: "Vapi (BYO model keys)", fixed: 0, perMin: 0.05, note: "models at cost, $0 if you bring your own API keys" },
  { name: "Retell (GPT-5 mini)", fixed: 0, perMin: 0.055 + 0.015 + 0.012, note: "infra + platform voice + LLM, all Retell-metered" },
  { name: "Bland Start", fixed: 0, perMin: 0.14, note: "all-in, 10 concurrent calls" },
  { name: "Bland Build", fixed: 299, perMin: 0.12, note: "all-in, 50 concurrent calls" },
  { name: "Bland Scale", fixed: 499, perMin: 0.11, note: "all-in, 100 concurrent calls" },
  { name: "ElevenLabs Agents", fixed: 0, perMin: 0.08, note: "speech engine only, LLM extra" },
  { name: "LiveKit Cloud (example stack)", fixed: 0, perMin: 0.0672, note: "calculator default: Gemma 4 31B, Deepgram Nova-3, Cartesia Sonic 3" },
  { name: "Pipecat Cloud (agent-1x + PSTN)", fixed: 0, perMin: 0.01 + 0.018, note: "STT/LLM/TTS billed by your providers" },
]

const usd = (n) => n.toLocaleString("en-US", { style: "currency", currency: "USD", maximumFractionDigits: 0 })
console.log(`\n${minutes.toLocaleString("en-US")} talk minutes per month\n`)
for (const p of PLATFORMS) {
  const total = p.fixed + p.perMin * minutes
  console.log(`  ${p.name.padEnd(32)} ${usd(total).padStart(9)}   ${p.note}`)
}
const breakEven = Math.ceil(299 / (0.14 - 0.12))
console.log(`\n  Bland Build beats Bland Start above ${breakEven.toLocaleString("en-US")} minutes/month\n`)

At 20,000 minutes a month it prints:

text
Copy
20,000 talk minutes per month

  Vapi (BYO model keys)               $1,000   models at cost, $0 if you bring your own API keys
  Retell (GPT-5 mini)                 $1,640   infra + platform voice + LLM, all Retell-metered
  Bland Start                         $2,800   all-in, 10 concurrent calls
  Bland Build                         $2,699   all-in, 50 concurrent calls
  Bland Scale                         $2,699   all-in, 100 concurrent calls
  ElevenLabs Agents                   $1,600   speech engine only, LLM extra
  LiveKit Cloud (example stack)       $1,344   calculator default: Gemma 4 31B, Deepgram Nova-3, Cartesia Sonic 3
  Pipecat Cloud (agent-1x + PSTN)       $560   STT/LLM/TTS billed by your providers

  Bland Build beats Bland Start above 14,950 minutes/month

Read the notes before the totals. Vapi's $1,000 and Pipecat's $560 exclude the models you still pay OpenAI and Deepgram for; Retell's and LiveKit's include them; Bland's include everything plus a compliance posture. At 100,000 minutes the spread runs from $2,800 (Pipecat compute plus PSTN) to $14,000 (Bland Start).

How to Pick

Shipping a developer product where the agent is one feature among many and you already hold model keys: Vapi, for the itemized bill and the SDK. Wanting one vendor to meter everything with no key management, and a generous free concurrency to test with: Retell. A regulated deployment where HIPAA, PCI or FedRAMP paperwork matters more than $0.05 a minute: Bland, or Vapi Scale if you need its model flexibility and can carry $2,000 a month for the HIPAA add-on. Voice quality as the product and a small call volume: ElevenLabs Agents. Engineers who want to own the pipeline and the cost curve: LiveKit Agents or Pipecat, hosted on their clouds until self-hosting pays for itself.

Conclusion

The voice agent market priced itself three ways in 2026: compute plus pass-through (Vapi, Pipecat), fully metered (Retell, LiveKit Cloud) and flat all-in (Bland). Vendors are raising fast enough that the rates will keep moving, so treat the September 5, 2026 numbers as a snapshot and re-run the script before signing anything annual. The shape of the decision will not change: whoever meters the LLM minute owns your margin.

Related DevToolLab Tools

• Text to Speech - prototype your agent's opening line in a browser voice before you pay a TTS provider per character.
• WAV File Generator - generate a test tone, silence or white-noise clip to feed a speech-to-text endpoint and measure how it handles dead air.
• Number to Words Converter - check how a dollar amount or confirmation code should be spelled out so your TTS layer reads it the way a human would.
• WAV Compressor - shrink recorded call audio before archiving it for QA or compliance review.

Related Guides

• Best Speech-to-Text and AI Voice APIs - the STT and TTS models these platforms wrap, priced on their own
• Best LLM Gateways and API Routers - routing and fallback for the model layer inside a voice pipeline
• Prompt Caching - cut the per-turn LLM cost that dominates a long call
• Best AI Agent Frameworks - the tool-calling and memory patterns behind an agent that does more than talk