For most teams it is a hosted agent platform: Retell AI at $0.07 to $0.31 a minute before telephony, Vapi at $0.05 a minute plus the speech, model and voice providers you choose, or Bland AI at $0.12 to $0.14 a minute with models included, on prices read November 13, 2026. Teams that want lower unit cost and more control build on LiveKit Agents or Pipecat, both open source, and pay a cent a minute or less for hosting plus their providers. Telnyx and Twilio suit teams that start from the phone network.
Key takeaways
- Eleven platforms in three layers: hosted agent platforms, speech and carrier APIs, and open-source frameworks.
- Lowest listed hosting: Pipecat Cloud at $0.01 a minute and LiveKit Cloud at $0.01 a minute after plan minutes, with speech, model and voice billed by your providers.
- Lowest hosted rates: Cartesia at $0.03 a minute plus telephony and model, Telnyx at about $0.06 with its own model and US inbound minutes, Retell from $0.07, Deepgram at $0.075.
- Concurrency differs widely: a limit of 4 on Vapi's free plan and 10 on its $29 plan, 20 on Retell before paying, 10 on Bland's Start plan, and 500 on Telnyx pay as you go.
- We build Botcadence, which is a finished agent for businesses and not a developer platform; it is described at the end and left out of the ranking.
How we chose and ordered them
We listed the platforms developers compare most, read each one's pricing page or repository on November 13, 2026, and ordered them by layer and then by the lowest listed cost per minute. We did not benchmark latency or accuracy for this version; AI voice agent latency explains how to measure it yourself.
The eleven, with what each publishes
| Platform | Layer | Price | You still supply | Concurrency and compliance from its pages |
|---|
| Cartesia | Hosted agent platform | $0.03 a minute; plans from free to $299 a month | Language model; telephony at $0.014 a minute or your own carrier | Agent concurrency of 8 to 60 by plan; SOC 2 Type II report; agreements on Enterprise |
| Vapi | Hosted agent platform | $0.05 a minute hosting; $0 and $29 plans; Pro from $999 a month | Speech-to-text, model and voice at provider rates | Limit of 4, 10 or 30 by plan; health-data agreement $2,000 a month |
| Retell AI | Hosted agent platform | $0.07 to $0.31 a minute | Telephony at $0.015 a minute or your own SIP; numbers $2 a month | Limit of 20, then $8 a month each; SOC 2 report; health-data agreement on Enterprise |
| ElevenLabs Agents | Hosted agent platform | $0.08 a minute after plan minutes; plans $6 to $990 | Language model; Twilio or SIP | Limit of 4 to 40 by plan, with paid bursts |
| Bland AI | Hosted agent platform | $0.14 a minute; $0.12 on the $299 plan | Telephony | Limit of 10 or 50 by plan; SOC 2 Type I and II reports |
| Telnyx Voice AI | Carrier API | $0.05 a minute engine; about $0.06 in production | Model tokens; SIP minutes from $0.0032 inbound | Limit of 500 on pay as you go |
| Twilio ConversationRelay | Carrier API | $0.07 a minute plus voice minutes | Your own server and language model | Not stated on the pages read |
| Deepgram Voice Agent API | Speech API | $0.075 a minute; $0.05 with your own model and voice | Telephony and agent code | Up to 60 in North America on Growth; SOC 2 reports |
| OpenAI Realtime API | Speech model API | By token: $32 per million audio input tokens and $64 output on the larger model | Telephony, orchestration and hosting | Not stated on the pricing page |
| LiveKit Agents | Open source, Apache-2.0 | Free to self-host; LiveKit Cloud from $0, then $50 a month for 5,000 agent minutes | Speech, model and voice providers | About 15,200 GitHub stars; cloud limit of 5 to 600 by plan |
| Pipecat | Open source, BSD-2-Clause | Free to self-host; Pipecat Cloud $0.01 a minute | Speech, model and voice providers; telephony at $0.018 a minute | About 17,200 GitHub stars |
Who should pick what
Retell AI: best for shipping a phone agent quickly with one itemized bill
Retell charges $0.015 a minute for telephony on top of its rate, bills per second and does not bill failed calls. See Retell AI pricing.
Vapi: best for choosing every provider
Vapi hosts the call and passes through the providers you select; its own estimate is $82 to $129 per 1,000 minutes. See Retell vs Vapi.
Bland AI: best for outbound campaigns at a flat rate
Model, speech and voice are inside one rate, and daily call limits are published. See Vapi vs Bland AI.
ElevenLabs Agents: best for voice quality with plan minutes
Plans include minutes, so small projects get a fixed bill. See ElevenLabs Agents vs Vapi.
Cartesia: best for a low hosted rate with your own model
Cartesia's managed agents run your code or its builder at $0.03 a minute and you connect your own language model.
Telnyx Voice AI: best when the phone network is the hard part
Telnyx owns its carrier network, so numbers, SIP and the agent are one account, billed in 60-second increments.
Twilio ConversationRelay: best for teams already on Twilio
Twilio handles speech in and out over a WebSocket; you write the conversation logic. See Twilio and AI voice agents.
Deepgram Voice Agent API: best for a speech-first stack
One API joins speech-to-text, your model and a voice. Deepgram does not sell telephony.
OpenAI Realtime API: best for speech-to-speech experiments
It is priced by token and is a model, not a phone product; you add telephony and hosting.
LiveKit Agents and Pipecat: best for control and unit cost
Both are open-source frameworks with paid hosting. You choose and pay each provider, and you own the reliability work. See open-source AI voice agents and build vs buy.
Where a finished agent fits instead
A platform sells parts. If nobody on the team will write prompts, connect calendars, test calls and watch quality every week, a finished agent costs less in total. Botcadence is one: $299 a month for 1,500 minutes with booking, transfers, on-call alerts and chat channels, set up for service businesses such as HVAC and restoration companies. It is not a developer platform: there is no pipeline to program, it offers no health-data agreement and has no SOC 2 report. Products of that kind are compared in best AI voice agents, and prices side by side in the AI voice agent pricing comparison. The plan is on the AI voice agent page.