Sonic 3.5
Cartesia's current TTS model, documented at sub-90ms first-byte latency across 42 languages.
- Resolution
- 48kHz
- Weights
- Not released
- API
- Yes
- Commercial use
- Permitted subject to conditions
It is tuned for conversational delivery and for reading alphanumerics - order numbers, phone numbers, emails - naturally, plus context-aware English homograph pronunciation. Output supports raw, WAV and MP3 containers at sample rates from 8kHz to 48kHz.
A commercial use licence is included on Pro and higher tiers; the Free tier does not include it.
- Depends on your plan tier. A commercial use licence is included on Pro and higher tiers; the Free tier does not include it
Specification
- What it takes in
- text
- What it makes
- audio
- Maximum resolution
- 48kHz
- Controls
- streaming-low-latency, voice-cloning, emotion-control, speed-control
- Weights
- Not released
- API
- Yes
- Pricing
- Credit-based subscriptions: Free $0/mo (20K credits, ~27 TTS minutes), Pro $5/mo (100K credits), Startup $49/mo, Scale $299/mo (~10,667 TTS minutes), Enterprise custom. Voice agents $0.06/minute; Cartesia-provided telephony numbers $0.014/minute; professional voice clone 225 credits one-time. No published per-character rate.
- Released or updated
- 2026-05-04 (pinned sonic-3.5 version date); Sonic 3 reported October 2025
- Languages
- 42, not individually listed by the vendor
Strengths
- Sub-90ms documented model latency, among the lowest published
- 42 languages
- Correct handling of numbers, IDs and emails without preprocessing
- Pinned model versions available (e.g. sonic-3.5-2026-05-04)
- Commercial licence starts at $5/month
Weaknesses
- Independent measurements of real-world latency are materially higher than the sub-90ms claim
- No published per-character price; credits make cost comparison awkward
- Documentation requires login for parts of the API reference
- Max characters per request not documented on public pages
- Consent/verification policy for voice cloning not documented publicly
Best for
- realtime voice agents
- high-volume short-form VO
- product demos needing fast turnaround
Where the sources disagree
Recorded rather than resolved. Picking the tidier number would hide the fact that the vendor's own pages do not agree.
- Cartesia claims sub-90ms model latency; a third-party writeup cites ~166ms median including network (attributed to Vapi, June 2026) and ~190ms median from Cartesia's own changelog. Treat sub-90ms as model-only under ideal conditions.
- Cartesia's own site header mentions Sonic 3.5 as latest while much third-party coverage still refers to Sonic 3.
What to understand before using it
Explainers naming the mechanisms this model's own documentation mentions.
MCP servers that front Cartesia
Servers in the media MCP directory whose own description or homepage names this vendor. Absence of a server here is not evidence that none exists: most publish no tool list at all.
-
InVideo
Create professional marketing videos with AI-powered generation, templates, and automated voiceovers.
See it from the vendor
Shotcraft hosts no sample output. Judging a model means watching its own showreel with your own eyes, and a clip re-hosted here would be a clip we chose.
Nearby in the directory
Azure AI Speech text-to-speech (Neural, Neural HD, MAI voices)
Microsoft's enterprise speech synthesis service, with 200+ prebuilt neural voices across 80+ locales plus tiers for HD voices, multilingual voices, mu
Voice and speech
XTTS-v2
An open-weights multilingual TTS and voice cloning model covering 17 languages at 24kHz, still widely used for local voice cloning. Coqui Inc. shut do
Voice and speech
Deepdub (eTTS, Speech-to-Speech, Live)
Enterprise AI voice and dubbing stack covering expressive text-to-speech with emotional control, speech-to-speech voice conversion and translation, vo
Voice and speech