Ranked · Voice agents
The 11 best AI voice agents (2026)
The best AI voice agents in October 2026 are OpenAI Realtime API, Deepgram Voice Agent API and Cartesia.
11 agents compared · 5 with a free plan · prices checked 2 September–24 October 2026
Reviewed by The Agents Index editorial team.
Ranked: 11 of the 18 voice agents we list. The 8 not yet ranked follow the ranking. All 18 in voice agents →
All picks at a glance
| # | Agent | Best for | Starts at |
|---|---|---|---|
| 1 | OpenAI Realtime APIThe raw model itself, no platform at all | Engineering teams building a voice agent from scratch | From $10/1m audio input tokens |
| 2 | Deepgram Voice Agent APIThe speech-infra company's own stack, deployable anywhere | Engineering teams that want a managed speech pipeline with real LLM-provider choice | From $0.05/minute |
| 3 | CartesiaThe other speech-infra company's stack — open-source SDK, 100+ LLMs | Engineering teams that want an open-source voice-agent SDK with real LLM choice | Free, then $5/month |
| 4 | VapiMaximum flexibility, bring-your-ownvs Retell AI | Engineering teams that want maximum control to build custom voice agents on their own choice of model, voice and telephony. | From $0.05/minute |
| 5 | Retell AIVoice-quality and low-latency leadervs Vapi | Product and engineering teams that want a low-latency, human-sounding phone agent | From $0.07/minute |
Checked between 2 September 2026 and 23 October 2026.
About this list
AI voice-agent products are not interchangeable. The first three entries are the underlying building blocks: OpenAI Realtime API is the raw model, Deepgram Voice Agent API is speech infrastructure with managed, VPC-hosted and self-hosted deployment, and Cartesia combines its Sonic and Ink models with an Apache 2.0 Line SDK. They suit teams that want to assemble the system themselves. Vapi, Retell AI and Bland AI are the next tier of developer-facing platforms: Vapi maximizes provider control, Retell emphasizes turn-taking and voice quality, and Bland AI publishes the clearest all-in hosted rate.
The bill is a second ordering principle because “per minute” does not mean the same thing across this set. OpenAI is token-metered and Deepgram is connection-time metered, with additional options depending on the components supplied. Cartesia separates agent minutes from monthly speech-model credits and telephony. Vapi and Retell separate orchestration from model, voice and telephony costs; ElevenLabs uses a subscription minute pool plus overage and can bill an external LLM and telephony separately. Synthflow AI currently has no public pricing anchor, while LiveKit Agents is free up to 1,000 agent-session minutes but still leaves inference and provider spending outside that allowance. PolyAI remains sales-led with no public rate card, and SkipCalls is a flat subscription whose own site and App Store listings currently show different prices. Bland AI is the clearest published all-in entry rate, although its higher tiers add platform fees.
The useful choice is therefore fit, not one universal winner. Start with OpenAI for lowest-level model control, Deepgram for speech infrastructure and deployment flexibility, or Cartesia for an open SDK. Choose Vapi or Retell for a custom build, Bland for a predictable hosted bill, ElevenLabs for voice quality and variety, Synthflow for a no-code ops path, LiveKit for self-hosting, PolyAI for established enterprise deployments, or SkipCalls when a finished receptionist is more useful than infrastructure. The list is intentionally ordered from the rawest building block to the most packaged product; it is not claiming that a model, an API, a no-code platform and a receptionist are direct substitutes. Every entry is a Published listing, and the notes below use the current pricing and capability anchors rather than treating a headline rate as a complete bill.
11 agents, ranked. How we rank
Featured agents
This spot is open
Featured agents show here and at the top of their category.
- API access: Not stated
- Free tier: Not stated
- MCP server: Not stated
- Own API key: Not stated
- Open source: Not stated
- Self-hosted: Not stated
The ranking
The raw model itself, no platform at all
Editor’s pick
Best forEngineering teams building a voice agent from scratch who want to go directly to OpenAI's own speech-to-speech model (no orchestration platform, no per-minute markup) and are comfortable being single-vendor on OpenAI.
VerdictRanked first as the raw model, not a platform: gpt-realtime-2.1 bills $32/$64 per 1M audio tokens in/out (the mini is about one-third), with no platform fee or bundled telephony. WebRTC, WebSocket and native SIP are available, but session management, tool calling and observability remain the builder’s job; the listing estimates $0.06–$0.11/min with cache-friendly usage.
- Price
- From $10/1m audio input tokens · source
- Watch for
- 60-minute maximum session duration.
The speech-infra company's own stack, deployable anywhere
Best forEngineering teams that want a managed speech pipeline with real LLM-provider choice, and regulated enterprises that need a self-hosted or VPC deployment option a typical closed-source voice-agent vendor does not offer.
VerdictSecond for teams that need speech infrastructure and deployment control in one API. The current Standard tier is $4.50/hour, while pay-as-you-go spans $0.050–$0.163/min depending on which STT, TTS and LLM you bring or buy; managed, VPC and self-hosted deployment, SOC 2 Type II, HIPAA BAAs and a one-time $200 credit distinguish it. It still lacks the dashboard, CRM integrations and permanent free tier of a full platform.
- Price
- From $0.05/minute · source
- Watch for
- No dedicated free tier. A one-time $200 credit only, then usage billing from the first paid session.
The other speech-infra company's stack — open-source SDK, 100+ LLMs
Best forEngineering teams that want an open-source voice-agent SDK with real LLM choice, built on speech models several rival platforms in this category already license as a component.
VerdictThird for speech-model quality plus an open agent SDK. Sonic and Ink are the product foundation, while Line connects to 100+ LLMs; pricing is deliberately two-meter: $0.06/min agent time plus a separate $0.014/min Cartesia phone number and monthly Sonic/Ink credits. The SDK is Apache 2.0, but the deployed Line runtime is not self-hostable, and the current free tier is $0/month with 20K credits and a $1 agent balance.
- Price
- Free, then $5/month · source
- Free to try
- Free plan
- Watch for
- Line's voice-agent runtime cannot be self-hosted — only the underlying Sonic/Ink models can, on an Enterprise contract.
Maximum flexibility, bring-your-own
Best forEngineering teams that want maximum control to build custom voice agents on their own choice of model, voice and telephony.
VerdictFourth for developer control: choose the speech-to-text, LLM and text-to-speech providers, then use Vapi to orchestrate calls. The current pricing anchor starts at $0.05/min for the Build orchestration fee, with model, voice and telephony pass-throughs kept separate. That flexibility is valuable for custom stacks, but it demands engineering and a complete budget for every component; there is no turnkey no-code route.
- Price
- From $0.05/minute · source
- Watch for
- True per-minute cost is the sum of several passed-through services, so the headline platform fee understates spend.
Voice-quality and low-latency leader
Best forProduct and engineering teams that want a low-latency, human-sounding phone agent with flexible model and telephony choice.
VerdictFifth for teams prioritizing turn-taking, call flow and QA around a developer-built stack. Pay-as-you-go starts at $0.07/min, and the listing’s basic-build example is about $0.11/min after its voice infrastructure, TTS, telephony and LLM components. It is easier to operate than a bare API, but the bill still depends on the selected stack and engineering remains useful for production quality.
- Price
- From $0.07/minute · source
- Watch for
- True per-minute cost is the sum of stacked components, so budgeting requires modelling model, voice, and telephony together.
Ranked 6 to 11
| # | Agent | Price | Actions |
|---|---|---|---|
| All-in pricing, owned infrastructure | |||
| 6 |  BlandSecurity-conscious enterprises in regulated industries that want a compliant Free, then $0.14/minute | Free, then $0.14/minute | |
| The voice-quality specialist's own agent layer | |||
| 7 |  | Free, then $6/month | |
| No-code, ops-team-first | |||
| 8 |  SynthflowOperations and CX teams at mid-market or enterprise companies who want a visual Free, then $30,000/year | Free, then $30,000/year | |
| Open infrastructure, self-hosted | |||
| 9 |  LiveKit AgentsEngineering teams that want to own their voice-agent stack end to end Free, then $50/month | Free, then $50/month | |
| Enterprise-proven, sales-led | |||
| 10 |  PolyAITeams that want a proven, enterprise-scale, compliance-certified voice AI vendor Price not established | Price not established | |
| A packaged receptionist, nothing to assemble | |||
| 11 |  SkipCallsVerifiedAI receptionist that answers a small business's existing phone number 24/7 From $19.99/month | From $19.99/month | |
| In this category, not yet ranked | |||
 VoiceAgentsNigerian SMB and mid-market operators who need a no-code phone agent that handles calls in Yoruba | |||
 AudioCodes Live HubSelf-service cloud platform for building voice AI agents | |||
 OmniDimensionNo-code platform for building voice AI agents and running them on phone, web chat, WhatsApp, email | |||
 RinglyAI phone agent that answers Shopify brands' inbound customer calls | |||
 Telnyx Voice AI AgentsCarrier-owned platform for phone and web voice agents, with telephony, STT, TTS, LLM | |||
 Leaping AIVoice and SMS AI agents for call centres, with built-in telephony, CRM integrations | |||
 HeyRikBuild, clone and deploy AI phone agents that answer inbound calls and run outbound calling campaigns | |||
 TelEchoConversational AI platform for voice and messaging agents that run customer calls | |||
Side by side
Every field, side by side
Scroll sideways to see them all.
| Feature | OpenAI Realtime API | Deepgram Voice Agent API | Cartesia | Vapi | Retell AI | Bland | ElevenLabs Agents | Synthflow | LiveKit Agents | PolyAI | SkipCallsVerified | VoiceAgents | AudioCodes Live Hub | OmniDimension | Ringly | Telnyx Voice AI Agents | Leaping AI | HeyRik | TelEcho |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Description | Speech-to-speech API for voice agents, with tools, MCP and WebRTC or WebSocket transport. | One WebSocket API that runs speech-to-text, LLM orchestration and text-to-speech for real-time voice agents. | Real-time speech models and a managed runtime for production phone and web voice agents. | Developer platform for building, testing and deploying voice AI agents that make and receive phone calls. | Pay-as-you-go platform for building, testing and running AI phone, chat and SMS agents. | Enterprise voice AI platform for inbound and outbound phone agents, running on Bland's own self-hosted models. | Hosted platform for building voice and chat agents, with telephony, SDKs and a choice of LLM. | Enterprise voice AI platform with in-house telephony for inbound and outbound call automation. | Open-source Python and Node.js framework for voice AI agents, with LiveKit Cloud hosting, telephony and observability. | Enterprise platform for building voice and chat agents that resolve customer calls, priced per minute. | AI receptionist that answers a small business's existing phone number 24/7. | Voice AI platform for phone agents that speak Yoruba, Igbo, Pidgin, Hausa and Nigerian English. | Self-service cloud platform for building voice AI agents and connecting them to real telephony channels. | No-code platform for building voice AI agents and running them on phone, web chat, WhatsApp, email and SMS. | AI phone agent that answers Shopify brands' inbound customer calls and escalates what it cannot resolve. | Carrier-owned platform for phone and web voice agents, with telephony, STT, TTS, LLM and orchestration in one stack. | Voice and SMS AI agents for call centres, with built-in telephony, CRM integrations and a no-code agent builder. | Build, clone and deploy AI phone agents that answer inbound calls and run outbound calling campaigns. | Conversational AI platform for voice and messaging agents that run customer calls and update connected systems. |
| Category | Voice agents, Model APIs & inference, Voice AI APIs | Voice agents, Voice AI APIs | Voice agents, Voice AI APIs | Voice agents | Voice agents | Voice agents | Voice agents, Agent platforms | Voice agents | Voice agents, Agent frameworks, Voice AI APIs | Voice agents | Support agents | Voice agents | Voice agents | Voice agents | Voice agents | Voice agents | Voice agents | Voice agents | Voice agents |
| Pricing | From $10/1m audio input tokens | From $0.05/minute | From Free | From $0.05/minute | From $0.07/minute | From Free | From Free | From Free | From Free | - | From $19.99/month | From $15/month | From $0.05/minute | From $15/month | From $349/month | From $0.05/minute | From $1,000/month | From ₹999/month (about $11) | From $0.07/minute |
| API | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | No | Yes | Yes | Yes | Yes |
| Tags |
Questions buyers ask
What is the best AI voice agent platform?
There is no universal winner. OpenAI Realtime API is the raw speech-to-speech model for teams that want full control. Deepgram Voice Agent API combines speech infrastructure with managed, VPC and self-hosted deployment. Cartesia pairs Sonic/Ink speech models with an Apache 2.0 Line SDK. Vapi and Retell AI are developer-facing platforms, while Bland AI is the clearest published all-in rate. ElevenLabs prioritizes voice quality and variety, Synthflow AI is the visual no-code path, LiveKit Agents is the open self-hostable framework, PolyAI is the enterprise incumbent, and SkipCalls is a packaged receptionist.
Which AI voice agent platform has true all-inclusive pricing?
Bland AI is the clearest published all-in option: its entry rate is $0.14/min with no platform fee, while Build publishes $0.12/min plus a $299/month platform fee. OpenAI is token-metered rather than per-minute; Deepgram is connection-time metered; Cartesia separates agent minutes from speech-model credits and telephony. Vapi, Retell AI and ElevenLabs expose component costs, LiveKit leaves inference spending outside its free allowance, and PolyAI has no public rate card.
How much do AI voice agent platforms actually cost per minute?
OpenAI Realtime API is usage-based: the current listing shows $32/$64 per 1M audio tokens in/out for gpt-realtime-2.1 and about one-third of that on the mini, with an estimated $0.06–$0.11/min under cache-friendly production usage. Deepgram lists $4.50/hour Standard time and $0.050–$0.163/min pay-as-you-go depending on the components supplied. Cartesia lists $0.06/min for agent time, $0.014/min for a Cartesia phone number and separate monthly model credits. Vapi starts at a $0.05/min Build orchestration fee; Retell starts at $0.07/min and gives an approximately $0.11/min basic-build example; Bland publishes $0.11–$0.14/min before its tier platform fees; ElevenLabs starts with a $0/month plan and charges overage from $0.08/min; Synthflow currently has no public pricing anchor; LiveKit includes 1,000 agent-session minutes at $0/month but usage beyond that is separate; PolyAI is custom; and SkipCalls is a flat subscription with conflicting public prices.
Which of these eleven owns its own infrastructure end to end?
They divide into several models. OpenAI Realtime API is the model itself, with no platform layer around it. Deepgram Voice Agent API can run managed, VPC-hosted or fully self-hosted. Cartesia owns its Sonic/Ink speech models, but its Line agent runtime is managed. LiveKit Agents is open-source and fully self-hostable. Bland AI and PolyAI run their own hosted stacks, while Vapi, Retell AI, ElevenLabs and Synthflow AI are bring-your-own-provider to some degree. SkipCalls is a finished receptionist rather than an infrastructure stack.
Which is cheapest to start using?
Vapi has the lowest published orchestration number at $0.05/min, but it excludes model, voice and telephony. LiveKit Agents has the most usable long-term free allowance at 1,000 agent-session minutes/month, while OpenAI and Deepgram have no dedicated ongoing free tier and PolyAI is sales-led. ElevenLabs has a $0/month plan with 15 call minutes, and Cartesia has a $0/month plan with 20K model credits and a $1 agent balance. Headline price is not the same as the complete bill for most entries.
Is there an open-source alternative to these hosted platforms?
Yes, but the shape differs. LiveKit Agents is the only genuinely open-source, self-hostable framework in this set and removes platform lock-in at the cost of writing and deploying the code yourself. Cartesia’s Line SDK is Apache 2.0, but the Line runtime and the Sonic/Ink models are not self-hostable in the same way. Deepgram offers managed, VPC-hosted and self-hosted deployment of its closed-source engine. OpenAI is a closed model API, so its openness is provider choice and direct access rather than self-hosting.
Which of these eleven is built for regulated industries like healthcare or finance?
Bland AI, Deepgram, PolyAI, Cartesia and Synthflow AI are the most direct regulated-industry candidates. Bland publishes an all-in per-minute entry rate and enterprise deployment options; Deepgram offers managed, VPC and self-hosted deployment with SOC 2 Type II and HIPAA BAAs; Cartesia publishes enterprise compliance and BAA options; PolyAI has the longest enterprise track record but is sales-led; Synthflow targets enterprise phone agents with visual operations tooling. LiveKit is a strong choice when a team wants self-hosting and accepts responsibility for its own compliance program, while OpenAI is a lower-level API rather than a turnkey regulated-industry product.
Which of these eleven has the deepest enterprise track record?
PolyAI, by a clear margin. Founded in 2017, it is the oldest name in the set and has the deepest established enterprise deployment history, including a large multilingual customer base; its current listing remains sales-led and publishes no rate card. That track record comes with an access tradeoff: buyers must enter a custom enterprise engagement, and the current third-party contract estimate is around $150,000/year rather than a self-serve price.
Sources and method
Every price here was read on the maker's own pricing page.
- OpenAI Realtime API: developers.openai.com/api/docs/pricing, checked 23 October 2026
- Deepgram Voice Agent API: deepgram.com/pricing, checked 2 September 2026
- Cartesia: cartesia.ai/pricing, checked 2 September 2026
- Vapi: vapi.ai/pricing, checked 3 September 2026
- Retell AI: retellai.com/pricing, checked 23 September 2026
All 10 sources
- Bland: bland.ai/pricing, checked 18 October 2026
- ElevenLabs Agents: elevenlabs.io/pricing/agents, checked 2 September 2026
- Synthflow: synthflow.ai/pricing, checked 3 September 2026
- LiveKit Agents: livekit.com/pricing, checked 24 October 2026
- SkipCalls: skipcalls.com/pricing, checked 26 September 2026
The digestFree
Which agents actually ship.
New agents and articles, by email, at most once a week.
Last issue · No. 1 · 13 October 2026Devin vs Claude Code: Which Should You Pick?
One-click unsubscribe