AI Stacks · Pricing checked November 21, 2026
Voice AI APIs Compared: ElevenLabs vs Deepgram vs Cartesia vs Vapi
Text-to-speech, speech-to-text and voice-agent platforms priced from their own pages on November 21, 2026, with the voice build that lagged on our show and how a voice-agent founder assembles the stack.
By Jordan Metzner & Sam Nadler · Prices from each vendor's own pricing page · Updated November 2026
The short answer
For a natural-sounding voice, ElevenLabs is the one we use on the show, and the voice-agent founder we hosted called it "one of the best, very fast" (at 04:18). For speech-to-text, Deepgram (from $0.0043 a minute) and ElevenLabs Scribe ($0.20 an hour) both come in well under a cent a minute. For a full phone agent, platforms like Vapi charge a small per-minute fee on top of whichever speech and LLM providers you plug in. Cartesia is the budget text-to-speech option, at $5 a month for about 133 minutes.
Pricing and limits (checked November 21, 2026)
Voice pricing comes in three units: credits or characters (text-to-speech), minutes (speech-to-text and agent hosting) and tokens (OpenAI Realtime). The tables keep each vendor's own unit rather than converting, because the conversions depend on how fast you talk and how much the agent says.
ElevenLabs
| Plan / model | Price | Included | Overage / pay-as-you-go |
|---|
| Free | $0 | 10,000 credits/mo (about 10 min of TTS) | — |
| Starter | $6/mo | 30,000 credits/mo (about 30 min) | About $0.20 per extra TTS minute |
| Creator | $22/mo | 121,000 credits/mo (about 121 min) | About $0.18 per extra minute |
| Pro | $99/mo | 600,000 credits/mo (about 600 min) | About $0.17 per extra minute |
| Scale | $299/mo | 1,800,000 credits/mo | About $0.17 per extra minute |
| API (pay as you go) | TTS from $0.04 per 1K characters; Agents $0.08/min + LLM | Speech-to-text (Scribe) $0.20/hour | — |
Source: ElevenLabs pricing, checked November 21, 2026. API rates from elevenlabs.io/pricing/api, same date. Standard monthly prices shown; a limited Black Friday discount was running.
Deepgram
| Plan / model | Price | Included | Documented limits | Overage / pay-as-you-go |
|---|
| Pay as you go | $200 free credit, no card, no expiry | Nova-3 streaming $0.0048/min; pre-recorded $0.0043/min | 150 concurrent STT streams | Aura-2 TTS $0.030 per 1K chars; Voice Agent $0.075/min |
| Growth | Commitment (amount not stated) | Nova-3 streaming $0.0042/min | 225 concurrent STT streams | Voice Agent $0.068/min |
| Enterprise | Custom | — | Set by peak traffic | — |
Source: Deepgram pricing, checked November 21, 2026. Deepgram labels its current speech-to-text rates as promotional.
Cartesia
| Plan / model | Price | Included | Documented limits | Overage / pay-as-you-go |
|---|
| Free | $0 | 20K credits/mo (about 27 min TTS) | 2 TTS concurrency; overages not available | — |
| Pro | $5/mo | 100K credits (about 133 min) | 3 TTS concurrency | $65 per 1M credits |
| Startup | $49/mo | 1.25M credits (about 1,667 min) | 5 TTS concurrency | $45 per 1M credits |
| Scale | $299/mo | 8M credits (about 10,667 min) | 15 TTS concurrency | $38 per 1M credits |
| Voice agents | $0.03/min + $0.014/min telephony | LLM billed separately | — | — |
Source: Cartesia pricing, checked November 21, 2026.
Vapi
| Plan / model | Price | Included | Documented limits | Overage / pay-as-you-go |
|---|
| Pay as you go | $0.05/min hosting | $5 free credits | 4 concurrent calls, 1 phone number | STT, LLM and voice passed through at cost |
| Core | $29/mo + $0.05/min | — | 10 concurrent calls, 5 numbers | Pass-through |
| Pro | 10% of hosting fee, $999/mo minimum | 99% uptime SLA | 30 concurrent calls | Pass-through |
| Premier | Custom | 99.9% SLA | Custom | — |
Source: Vapi pricing, checked November 21, 2026.
OpenAI Realtime API
| Plan / model | Price | Included | Documented limits |
|---|
| gpt-realtime-2.1 (audio) | $32 in / $64 out per 1M tokens | Cached audio input $0.40 per 1M | API rate limits by tier |
| gpt-realtime-2.1-mini (audio) | $10 in / $20 out per 1M tokens | Cached audio input $0.30 per 1M | API rate limits by tier |
Source: OpenAI Realtime API pricing, checked November 21, 2026. OpenAI publishes no per-minute equivalent for Realtime.
What broke when we built with it
We haven't built a phone agent on the show ourselves, so this section is short. Our one live voice build was a Halloween "mirror mirror" app made in Google AI Studio. It ran real-time speech-to-text into Gemini and read the answer back with text-to-speech. It answered the wrong company, and it lagged:
It said it was Microsoft here. It looks like it says that it's Microsoft instead of Nvidia. But, you know, and it's a little slow and laggy.
— Jordan Metzner, Episode 19, at 04:08
Jordan uses ElevenLabs text-to-speech to narrate code-generated videos, for example turning "a 65 page slide deck" into a video with Remotion (at 11:43). He didn't report a problem with it. We have no build of our own with Deepgram, Cartesia, Vapi or the OpenAI Realtime API, so those appear in the price table only.
On dictation, which is speech-to-text for your own typing, Jordan dropped Apple's built-in dictation because "it's really clunky" and moved to Wispr Flow (at 29:57). His one failure mode: "Sometimes I'll not click in the right box" and the words go nowhere, which a newer re-paste feature fixes (at 33:00).
What our guests reported
Karim Malhas, CEO of the voice-agent platform Breeze, built an agent live as "ears, brains, and voice", picking Deepgram for speech-to-text and ElevenLabs for the voice (at 03:59). He also pointed out a gap in do-it-yourself platforms: for smart phone-number routing, "in [Vapi] and the rest, well, it's on you to build it" (at 12:46).
How to choose
- Narration and voiceover: ElevenLabs Creator ($22 a month, about 121 minutes) or Cartesia Startup ($49 a month, about 1,667 minutes).
- Transcribing calls or meetings: Deepgram Nova-3 pay as you go ($0.0043 a minute pre-recorded) or ElevenLabs Scribe ($0.20 an hour).
- Phone agents: add up three meters: the platform (Vapi $0.05 a minute, Cartesia agents $0.03 a minute plus telephony, ElevenLabs agents $0.08 a minute), speech-to-text, and the LLM (see LLM API pricing).
- Speech-to-speech in one model: OpenAI Realtime is billed per audio token. Run a test call and read your usage page before estimating a per-minute cost.
FAQ
How much does ElevenLabs cost per minute?
ElevenLabs' pricing page lists Starter at $6 a month for 30,000 credits (about 30 minutes of text-to-speech), Creator at $22 for about 121 minutes and Pro at $99 for about 600 minutes, with extra minutes at roughly $0.17 to $0.20. On its API page, text-to-speech starts at $0.04 per 1,000 characters and agents cost $0.08 a minute plus LLM costs (November 21, 2026).
Is Deepgram cheaper than ElevenLabs for speech-to-text?
Not on list price. On November 21, 2026, Deepgram listed Nova-3 streaming speech-to-text at $0.0048 a minute pay as you go (labelled promotional), about $0.29 an hour, while ElevenLabs listed its Scribe speech-to-text at $0.20 an hour. Deepgram does give $200 of free credit with no card, which covers a lot of testing.
What does Vapi charge?
Vapi charges a $0.05 per minute hosting fee on pay as you go, with speech-to-text, LLM and voice providers passed through at cost. Core is $29 a month plus $0.05 a minute, and Pro is 10% of the hosting fee with a $999 monthly minimum.
What does OpenAI's Realtime API cost?
OpenAI prices Realtime by token, not by minute: gpt-realtime-2.1 audio is $32 per million input tokens and $64 per million output tokens, and gpt-realtime-2.1-mini audio is $10 and $20. OpenAI does not publish a per-minute equivalent.
What stack do voice agents use?
A voice agent is usually three parts: speech-to-text (for example Deepgram), an LLM, and text-to-speech (for example ElevenLabs or Cartesia), with a platform such as Vapi or Breeze handling calls, phone numbers and routing. That is the setup Breeze's CEO built live on Built This Week.