CPaaS stands for Communications Platform as a Service: a cloud platform that lets developers add calls, texts, video, and verification to their own applications through APIs, without building or running any carrier infrastructure. Twilio created the category in 2008 and still defines it (its own one-page guide is the reference definition), with Vonage, Sinch, Plivo, Bandwidth, Infobip, and Bird (formerly MessageBird) working the same model.
The classic CPaaS buyer is a product team that needs a notification, a one-time passcode, or a masked phone call inside an app. The ride-share text that your driver has arrived, the reminder from your dentist, the 2FA code: all CPaaS traffic. The newer buyer is a team putting a voice agent on the phone, and that buyer should read to the end, because the platform question has changed shape.
What a CPaaS covers
The exact catalog varies by provider, but the core is the same set of channels, each reachable from an API:
- Messaging. SMS and MMS, toll-free messaging, and OTT channels like WhatsApp and Messenger.
- Voice. Programmable calls over VoIP and the public phone network, plus SIP trunking to connect your existing carrier.
- Video. Peer-to-peer calls, group rooms, and livestreaming.
- Email. Transactional and marketing delivery.
- Verification. One-time passcodes, two-factor authentication, and identity checks.
- Low-code tools. Visual builders for teams that want to assemble flows without writing every integration by hand.
You compose these into your product. A healthcare app might use verification for logins, SMS for reminders, and voice for the urgent cases. The CPaaS owns the carrier relationships, the delivery routing, and the compliance plumbing; you own the application.
CPaaS vs UCaaS vs CCaaS
The three acronyms get confused constantly, and vendors selling all three don’t help. The distinction is who the software is for:
| CPaaS | UCaaS | CCaaS | |
|---|---|---|---|
| What it is | Communications APIs you embed in your own product | Internal collaboration suite (calls, meetings, chat) | Hosted contact center for support teams |
| Who uses it | Developers | Employees | Contact-center agents |
| Examples | Twilio, Vonage, Sinch, Plivo | Zoom, RingCentral, Microsoft Teams | Genesys Cloud, Five9, NICE |
| Pricing | Per minute, per message, per call | Per seat per month | Per agent seat per month |
If you’re buying phones for your staff, that’s UCaaS. If you’re running a support queue, that’s CCaaS. If you’re building communication into something you sell, that’s CPaaS. The letters rhyme; the buyers are three different teams.
How CPaaS pricing works
CPaaS is usage-based, with no platform fee: per minute for voice, per segment for SMS, per successful verification for auth. The model is why the category spread: a startup can send its first text for less than a cent and scale the same account to millions of messages.
Twilio’s US list prices, as of October 2026, are the usual worked example: $0.014 per minute for an outbound call, $0.0085 per minute inbound to a local number, $1.15 per month for the number itself, and $0.0083 per SMS segment. Two billing rules matter more than the rates: calls bill per leg (an outbound call you forward pays for both legs), and most providers round up to the next minute. The Twilio pricing guide walks through the full rate card.
One honest warning: the per-minute price is for the pipe. Anything that listens, thinks, or talks on the call (transcription, an LLM, text to speech) is another metered layer, from the CPaaS or from whoever you buy models from. Twilio’s own agent product, Conversation Relay, adds $0.07 per minute of active session. Add the layers before you compare the totals.
The providers
The names you’ll meet, in rough order of how often they come up:
- Twilio: the category-maker and the default evaluation. Broadest API catalog, usage pricing, and the docs everyone else’s docs are compared against.
- Vonage: the former Nexmo; strong in voice and video APIs, owned by Ericsson.
- Sinch: large in messaging and voice, especially internationally, after buying up several regional providers.
- Plivo: voice and SMS APIs with a smaller, simpler catalog.
- Bandwidth: a CPaaS that owns its carrier network, which matters when you need number porting or 911 routing done in-house.
- Infobip: international messaging and omnichannel, strong outside North America.
- Bird: the former MessageBird, messaging-led with a European base.
The right shortlist depends on your channels and your countries. For a US voice workload, any of them can carry a call; the differences are price, reliability in your regions, and how much of the stack you want from one vendor.
Where voice AI changes the math
A CPaaS call is a pipe. It connects two ends and stays out of the way, which is exactly right when a human is on each end. But a growing share of business calls in 2026 have software on one end: an agent that qualifies a lead, takes an order, or reschedules a delivery. A pipe alone doesn’t give you that. A voice agent needs five things on top of the pipe:
- Ears. Streaming speech to text that hears the caller the moment they start talking.
- A brain. An LLM deciding what to say, with your tools and your data.
- A voice. Text to speech that sounds like someone worth listening to.
- Turn-taking. The model that decides when the caller finished, when to interrupt, and when to wait. Get this wrong and the call feels like a walkie-talkie.
- Telephony. Numbers, routing, recording, and the regulatory bits.
You can assemble all of that on a CPaaS: Twilio carries the call, you stream audio to your speech-to-text, run your model, synthesize the reply, and own the glue, the latency budget, and the turn-taking yourself. Teams with strong real-time engineering do this and get exactly the stack they want. The alternative is a voice agent platform that bundles the five pieces and treats the phone network as one input, the same move that made CPaaS itself attractive, one layer up.
The question that separates the two: does the call need to think? If the answer is no (appointment reminders, delivery notifications, OTP codes), buy the pipe. If yes, you’re shopping for the layers above the pipe, and that’s a different evaluation.
Where Cartesia fits
Cartesia is not a CPaaS. We build the conversation layer: Sonic for text to speech, Ink for speech to text, and Managed Agents, the hosted voice-agent builder. Telephony comes three ways, per the phone numbers docs: Cartesia-provisioned US numbers (free with your plan, from one number on Free to custom on Enterprise, plus a $0.014 per minute telephony fee), numbers imported from your Twilio account at your Twilio rates with international outbound, or your own carrier over a SIP trunk. Agent calls cost $0.06 per minute, your LLM’s tokens bill at the model’s listed prices, and the free plan includes $1 of agent calls a month, enough to call your own agent from your phone and see how it handles you.
If what you need is messaging APIs, a verification flow, or raw call minutes, a CPaaS like Twilio is the right tool, and our agents will happily run on numbers you import from it. If what you need is a voice that holds a conversation, start in the playground and read how voice agents work.