Best text-to-speech APIs for voice agents
Real-time synthesis APIs for building a voice agent's speaking half, compared on the credit and character pricing each vendor actually publishes.
Cartesia wins on price for a team starting small: its $5/month Pro tier and 20,000 free monthly credits undercut ElevenLabs' equivalent entry tier. ElevenLabs wins instead once you need the widest published tier ladder, from a free plan through a $990/month Business plan, giving a team more granular room to grow without an early jump to custom pricing.
The short answer
- ElevenLabs' Free tier gives 10,000 credits/month (about 10 minutes); its Starter tier is $6/month for 30,000 credits, scaling to $299/month (Scale, 1.8M credits) and $990/month (Business, 6M credits), per elevenlabs.io/pricing.
- Cartesia's Free tier gives 20,000 credits/month plus $1 of prepaid agent minutes; its Pro tier is $5/month for 100,000 credits, scaling to $49/month (Startup, 1.25M credits) and $299/month (Scale, 8M credits), per cartesia.ai/pricing.
- Azure AI Speech bills text-to-speech per character rather than a flat monthly tier, and Microsoft's own pricing page renders exact per-character rates through a region-dependent calculator rather than a static number.
Text-to-speech APIs compared
| Provider | Entry price | Free tier | Top published tier | Best for | Not for |
|---|---|---|---|---|---|
| ElevenLabs | $6/mo (Starter, 30,000 credits) | 10,000 credits/mo (~10 min) | $990/mo (Business, 6M credits) | A team wanting a wide tier ladder before jumping to custom pricing | A team wanting the absolute lowest paid entry price |
| Cartesia | $5/mo (Pro, 100,000 credits) | 20,000 credits/mo + $1 agent minutes | $299/mo (Scale, 8M credits) | A team wanting the cheapest paid tier with cloning included early | A team wanting a published tier above $299/mo instead of custom Enterprise |
| Azure AI Speech | Per-character, rate shown via calculator | Standard Azure free-tier credit (general, not TTS-specific) | None published as a flat monthly tier | A team already billing everything through Azure | A team that wants a quotable flat rate without opening a calculator |
Figures checked on elevenlabs.io/pricing and cartesia.ai/pricing, September 2026. Azure's exact per-character rate could not be read as a static figure from azure.microsoft.com/en-us/pricing/details/cognitive-services/speech-services/, which renders pricing through an interactive, region-dependent calculator; PlayHT's pricing page was unreachable at the time of writing and is not included in this table.
How the two priced-and-published options differ in practice
ElevenLabs
The widest published tier ladder of any TTS vendor checked, from a free plan through six paid tiers up to $990/month, which lets a team scale spend gradually before hitting a custom Enterprise conversation. Voice cloning access and commercial-use rights are gated by plan tier rather than included flat across every paid plan. Not for: a team that wants the cheapest possible paid entry point.
Cartesia
Positions its entry paid tier at $5/month, undercutting ElevenLabs' $6/month Starter, and bundles instant voice cloning into that same Pro tier rather than gating it to a higher plan. Its pricing page states unlimited workspace seats on every plan, so a team's headcount doesn't change the bill the way per-seat software does. Not for: a team that wants a published tier between $299/month and full custom Enterprise pricing.
When none of these is the answer
If you're recording a one-off voiceover or narration rather than building a live agent that responds to a caller, a batch TTS tool optimized for narration quality is a better fit than a real-time streaming API billed for low-latency response, which is a different product shape even when the underlying model sounds similar.
SalesCrew is not included in the ranking above because it does not offer a text-to-speech API; it runs outbound calls through Vapi as one channel, which selects its own voice provider underneath. Inbound voice and an AI receptionist are on the roadmap, not shipped, so SalesCrew is not a candidate in a text-to-speech comparison today.
Questions
- What does text-to-speech cost for a voice agent, not only a voiceover?
- A voice agent needs low-latency streaming synthesis, not only a rendered audio file, so the relevant number is the per-character or per-credit rate on a real-time tier, not a batch narration price. ElevenLabs and Cartesia both publish real-time tiers; the rate on the page is transcription-adjacent cost only, not the full per-minute cost of the call.
- Is a free tier enough to prototype a voice agent?
- It's enough to hear the voice, not to run a real call flow. ElevenLabs' free tier gives 10,000 credits a month, about 10 minutes of audio, per elevenlabs.io/pricing. Cartesia's free tier gives 20,000 credits a month plus $1 of prepaid agent minutes, per cartesia.ai/pricing.
- Do these platforms charge extra for voice cloning?
- It depends on the tier. Cartesia includes instant voice cloning starting on its $5/month Pro tier, with professional cloning added at $49/month Startup. ElevenLabs gates commercial-use voice cloning by plan as well; check the current tier breakdown on elevenlabs.io/pricing before committing to a voice.
- What does published TTS pricing leave out?
- The telephony carrying the call, the language model generating what the agent says, and the speech-to-text transcribing the caller. A per-character or per-credit TTS rate is one of four cost components in a live voice agent, not the total.