Cartesia Sonic 4 does not exist: Sonic 3.6 is latest
There is no Cartesia Sonic 4 as of 8 October 2026. The current model is Sonic 3.6: 44 languages, sub-90 ms first audio, 4 credits per 1,000 characters.
There is no Cartesia Sonic 4. As of 8 October 2026 the current Cartesia text-to-speech model is Sonic 3.6 (sonic-3.6), generally available since 27 August 2026 (Cartesia docs). Cartesia's documentation index lists Sonic 3.6 as its newest model and files Sonic 3.5, Sonic 3, Sonic 2 and Sonic Turbo under older models. No Sonic 4 appears in the index or on its model pages.
An earlier version of this page, published on 21 August 2026, reviewed a "Sonic 4" with a 40 ms time to first audio and a price of $0.05 per 1,000 credits. That model does not exist, so the review has been replaced with what Cartesia and ElevenLabs actually publish, and this page is kept out of search results.
The short answer
- Current model: Sonic 3.6, released 27 August 2026. It is backwards compatible with Sonic 3.5 and supports 44 languages. Cartesia says it reads emotion from the transcript without SSML tags (Cartesia docs).
- Speed: Cartesia's Sonic page claims first audio in under 90 ms. Its Sonic 3.5 docs gave the same sub-90 ms figure. Cartesia's pages for the current model do not claim "40 ms".
- Price: Cartesia's Pro plan is $5 a month for 100,000 credits, and one credit is one character, so $0.05 per 1,000 credits. Its Scale plan is $299 for 8 million credits, about $0.037 per 1,000 (Cartesia pricing, read 8 October 2026). That is 37.5% below ElevenLabs v3 at $0.08 per 1,000 characters, not "3-4x cheaper", and above ElevenLabs Flash / Turbo at $0.04 (ElevenLabs API pricing, read 8 October 2026).
- Where it loses: language breadth. ElevenLabs lists 70+ languages for v3 and 90+ for its newer v4, against Sonic 3.6's 44.
- Inside Versely: Cartesia Sonic 3.6 is in the catalog at 4 credits per 1,000 characters, the same price as Sonic 3.5. Voices cloned on earlier Sonic versions carry over.
What a fast first byte unlocks
Two use cases genuinely change shape when audio starts in well under a tenth of a second, and neither is "faster video voiceover":
Real-time voice agents. Above a certain latency a voice agent stops feeling like it is listening and starts feeling like it is processing. Users register a small hitch as "the AI thinking", even when the language model was the fast part and the voice was the bottleneck. A model that starts speaking in under 90 ms, by Cartesia's own figure, takes the voice almost entirely out of that equation and leaves the reasoning latency as the thing left to optimise.
Live dubbing. Translating and re-voicing a live stream, call or broadcast requires the synthesized voice to keep pace with a speaker who is not waiting. A dubbed voice that lags by a few hundred milliseconds compounds across a conversation. Low first-audio latency keeps that gap small enough that the audience does not notice it.
Neither of those is a pre-rendered voiceover for a video export. If your job is "write a script, generate a file, attach it to a video", you are not in a turn-taking loop, and the viewer never perceives generation-time latency: the file already exists by the time anyone watches. Where speed still helps there is iteration speed during production, which is real but smaller than the headline number implies.
The pricing gap vs ElevenLabs
| Model | List price per 1,000 characters | Source |
|---|---|---|
| Cartesia Sonic 3.6, Pro plan | $0.05 | Cartesia pricing |
| Cartesia Sonic 3.6, Scale plan | about $0.037 | Cartesia pricing ($299 for 8M credits) |
| ElevenLabs v3 | $0.08 | ElevenLabs API pricing |
| ElevenLabs Flash / Turbo | $0.04 | ElevenLabs API pricing |
| ElevenLabs v4 | $0.08 (promotional $0.022 until 12 October 2026) | ElevenLabs API pricing |
The gap is real against v3, small against ElevenLabs' faster models, and reversed against Flash / Turbo at Cartesia's Pro rate. Cartesia also says a minute of speech takes 750-800 credits, so the Pro plan works out to roughly $0.04 per minute of audio. None of this is a Versely price. On Versely, text-to-speech is billed in credits per 1,000 characters: Sonic 3.6 is 4, Gemini 3.8 Flash TTS and Inworld TTS 2 are 2, and ElevenLabs Multilingual is 8.
Where Sonic 3.6 honestly loses
Being fast and cheap is not the same as being the most expressive or the broadest. ElevenLabs lists 70+ languages on v3 and 90+ on v4, and calls v4 its "most emotive model yet". If the project is a character-driven audiobook, a brand voice that needs a very specific emotional register, or a campaign across a dozen languages, ElevenLabs may still be the stronger pick whatever the price gap.
Hume Octave 2 is not an alternative for new work. Hume is sunsetting its TTS and EVI APIs, and access ends on 13 November 2026 (Hume changelog, 2 October 2026). If you want delivery steered in plain language, Gemini 3.8 Flash TTS and Inworld TTS 2 both take natural-language style direction on Versely at 2 credits per 1,000 characters.
Inside Versely, if your workflow is "generate a voiceover and attach it to a video" rather than "build a real-time voice agent", Versely's AI text-to-speech tool already gets you most of the practical benefit of Sonic 3.6: fast iteration and clean reads of numbers and codes. For a live-dubbing workflow see Versely's AI dubbing tool, and for a reusable cloned voice across scripts see AI voice cloning.
FAQ
Is there a Cartesia Sonic 4? No. As of 8 October 2026 Cartesia's latest model is Sonic 3.6. If you saw "Sonic 4" quoted anywhere, including an earlier version of this page, the model and its figures were not published by Cartesia.
How fast is Sonic 3.6? Cartesia's Sonic page claims first audio in under 90 ms, in 44 languages from one model. That is Cartesia's own figure for its stack, not a Versely benchmark.
Is Cartesia cheaper than ElevenLabs? Against ElevenLabs v3, yes: $0.05 per 1,000 credits on Cartesia's Pro plan against $0.08 per 1,000 characters. Against ElevenLabs Flash / Turbo at $0.04 it is not, at that plan. Compare on the plan you would actually buy.
Is Sonic 3.6 available inside Versely? Yes. Cartesia Sonic 3.6 is 4 credits per 1,000 characters.
Does low latency matter for a pre-recorded voiceover? Less than the headline suggests. A viewer never experiences the render-time gap in a finished export. The speed mainly helps you iterate through more takes during production.
When should I pick ElevenLabs over Sonic? When expressiveness, language breadth, or voice library depth matter more than speed or cost: character voice work, a very specific emotional register, or a campaign that needs the widest per-language voice choice.
The takeaway
Sonic 3.6 is a fast, inexpensive model for the narrow set of jobs that live inside a latency-sensitive loop, such as voice agents and live dubbing, and a fair price option for anyone generating speech at volume. It is not a Sonic 4, and it is not automatically the more expressive choice. If your content ships as a finished video rather than a live conversation, Versely's text-to-speech tool covers most of the practical benefit today.