Cartesia
Cartesia: Sonic AI voices for narration and live applications
Cartesia creates AI speech in 44 languages with Sonic 3.6. Use its Playground or API for narration, custom voices and live conversational apps.
Free plan availableAccount required

- Access
- Free and paid plans
- Example plan
- USD 5 / monthPro
- Available on
- Web, API
A speech platform for interactive products
Cartesia combines Sonic text-to-speech, Ink transcription and Managed Agents in one speech platform. Sonic 3.6 supports 44 languages, including Odia and Urdu. Its browser Playground provides voice selection and delivery controls; the API brings speech into applications. The Sonic 3.6 alias follows stable updates, while a dated snapshot keeps model behaviour fixed for as long as that version remains available.
Instant voice cloning starts on Pro with a short recording. Sonic 3.6 accepts up to 60 seconds, with a 16 MB file limit. Professional clones need at least 30 minutes of single-speaker audio and a Startup or higher plan. Startup includes two training slots per organisation and Scale includes four. Training is free, and professional-clone speech uses the standard character rate. Existing professional clones remain usable after a downgrade; the lower plan limits further training.
Speed and volume settings apply to each generation rather than changing a voice globally. Speed ranges from 0.6 to 1.5 and guides the delivery rather than guaranteeing an exact duration. Professional clones keep the pacing and loudness of their training recordings, so request-time speed and volume settings have no effect. Emotion guidance is currently an English-only beta feature. For file-based narration, the Bytes endpoint returns MP3 or WAV; SSE and WebSocket endpoints stream raw audio.
Speech generation and transcription draw from the same credit balance. Standard TTS costs approximately one credit per character, with small differences from transcript processing. Transcription is charged by audio duration, including silence; the rate varies by model. Paid credits roll over up to twice the monthly allowance. Managed Agents have a separate prepaid dollar balance: calls cost USD 0.06 per minute, with another USD 0.014 per minute for a Cartesia phone number. LLM charges can add to that bill; a current promotion waives them for agents created in the UI.
The Free plan is for noncommercial use; paid plans add commercial rights. Cloning requires your own voice or the speaker’s explicit consent. Standard terms allow model training, with a form-based opt-out for selected content categories that takes effect after processing and does not undo earlier use. Enterprise zero-data-retention is a separate option for TTS and STT inference across the organisation. It excludes voice creation and cloning, retains operational metadata, and can remove Playground history.
Best for
- Developers adding speech to interactive applications
- Teams building multilingual voice agents
- Creators generating narration with custom voices
Limitations
Professional clones have fixed pacing and volume; emotion controls are English-only beta. TTS and transcription share credits, while Managed Agents have separate dollar charges. Free output is noncommercial. Enterprise zero retention excludes cloning. Sonic 2, Sonic Turbo and the Sonic 3 October 2025 snapshot are scheduled to retire on 20 October 2026.
Cartesia speech and voice features
Sonic 3.6 multilingual speech
Create speech in 44 languages. Choose the Sonic 3.6 alias for stable updates or a dated snapshot for fixed behaviour while that version remains available.
Streaming and file output
Get MP3, WAV or raw audio from the Bytes endpoint, or stream raw audio over SSE and WebSocket. Output sample-rate options span 8–48 kHz.
Instant custom voices
Pro and higher support short-reference cloning. Sonic 3.6 accepts up to 60 seconds of a speaker’s recording within a 16 MB file limit.
Professional voice training
Train on at least 30 minutes of one speaker. Startup includes two slots per organisation; Scale includes four. Existing clones remain usable after a downgrade.
Delivery controls
Set speed and volume per generation, with English emotion guidance in beta. Professional clones retain their recorded pacing and loudness.
Enterprise retention options
Enterprise zero retention covers TTS/STT inference across the organisation. Cloning material and operational metadata remain outside that scope.
Cartesia pricing
Free includes 20,000 speech credits per month. Paid plans start at USD 5/month for commercial use and instant cloning; professional cloning starts at USD 49/month. TTS and transcription share credits, while Managed Agents use a separate prepaid dollar balance. Paid rollover is capped at twice the monthly speech allowance.
Example plans.
Pro
USD 5 / month
Cartesia speech platform
Plan details
100,000 shared speech credits/month; commercial use and instant cloning. TTS is approximately 1 credit/character.
Startup
USD 49 / month
Cartesia speech platform
Plan details
1.25 million shared speech credits/month; two professional-clone slots per organisation.
Scale
USD 299 / month
Cartesia speech platform
Plan details
8 million shared speech credits/month; four professional-clone slots and 15 concurrent TTS requests.
Technical specifications
Voice generation
| Feature | Cartesia |
|---|---|
| Speech languages | Sonic 3.6 supports 44 languages, including Odia and Urdu. A cloned voice starts in its recorded language; additional native accents use the voice-localisation capability.Language coverage is model-specific |
| Voice and pronunciation controls | Per-generation speed (0.6–1.5) and volume (0.5–2.0) guidance; English emotion controls are beta. Professional clones ignore request-time speed and volume changes.Controls vary by voice type; emotion tags English only |
| Voice cloning | Instant cloning starts on Pro; Sonic 3.6 accepts up to 60 seconds within a 16 MB file limit. Professional cloning needs 30+ minutes and Startup or above, with two Startup or four Scale slots per organisation. Existing professional clones remain usable after a downgrade; further training follows the new slot limit.Paid tiers; use your own voice or obtain explicit consent |
| Audio exports and streaming | The Bytes API returns MP3, WAV or raw audio; SSE/WebSocket streaming returns raw audio. Output sample-rate options span 8–48 kHz.Container support differs by endpoint |
| Commercial-use terms | Paid plans include commercial use subject to input rights and acceptable-use rules. Standard terms allow model training; category-specific form opt-out applies after processing, without reversing prior use. Enterprise zero retention covers inference, excluding cloning.Paid commercial rights; Enterprise retention scope |
| Output sample rate (Hz) | 48000Maximum selectable output rate; not a voice-quality score |
| API access | SupportedBytes, SSE and WebSocket endpoints; account/API key required |
Cartesia alternatives
Other speech products offer browser production tools or a self-hosted model family.
ElevenLabs
Hosted speech and custom voices alongside a studio, dubbing and other audio tools.
Resemble AI (Chatterbox)
Open-source voice models for local deployment and self-managed infrastructure.
Cartesia FAQs
Is Cartesia free?
Free includes 20,000 monthly speech credits for noncommercial use. Pro starts at USD 5/month and adds commercial use and instant cloning. Standard TTS uses approximately one credit per character; transcription uses the same credit balance.
What is the difference between instant and professional cloning?
Instant cloning starts on Pro with a short recording; Sonic 3.6 accepts up to 60 seconds within a 16 MB file limit. Professional cloning needs at least 30 minutes and starts on Startup, with two training slots per organisation or four on Scale. Professional clones have fixed pacing and volume. Existing clones remain usable after a downgrade, but new training is restricted by the lower plan’s slot allowance.
Can Cartesia generate MP3 or WAV audio?
Yes, through the TTS Bytes endpoint, which also supports raw audio. SSE and WebSocket endpoints stream raw audio instead. Selectable output sample rates range from 8 to 48 kHz.
Do Cartesia credits roll over?
Unused paid speech credits roll over up to twice the monthly allocation. Tier changes preserve the existing balance, with future rollover subject to the new plan’s cap. If speech overages are disabled, requests stop when credits run out; enabling them adds usage charges. Managed Agents use a separate dollar balance.
Can I change the speed and emotion of a Cartesia voice?
Speed and volume settings apply per generation. Speed values from 0.6 to 1.5 guide pacing rather than guaranteeing an exact duration. Instant clones support speed changes; professional clones ignore request-time speed and volume settings. Emotion guidance is an English-only beta feature.
Which Sonic model should a new project use?
Sonic 3.6 is the current stable model. Its general alias follows stable updates; sonic-3.6-2026-08-27 pins the current dated snapshot. Sonic Preview is a changing beta. Cartesia schedules all Sonic 2 and Sonic Turbo aliases and snapshots, plus sonic-3-2025-10-27, to retire on 20 October 2026. A dated snapshot fixes behaviour while available; it does not guarantee permanent availability.
Are Cartesia voice agents included in speech credits?
Managed Agents use a separate prepaid dollar balance. Calls cost USD 0.06 per minute, plus USD 0.014 per minute when using a Cartesia phone number. LLM usage can add charges. Cartesia currently waives LLM costs for agents created in the UI as a limited-time promotion; that is not a permanent inclusion in speech credits.
Does Cartesia offer zero-data-retention?
Enterprise can enable zero retention across the organisation for TTS and STT inference content. It excludes voice creation and cloning, retains operational metadata and can remove Playground history. Standard terms otherwise allow model training; a form-based opt-out for selected content categories applies after processing and does not reverse earlier use.
Is this your product? Claim this page to update your listing.