Hume Octave 2 vs Cartesia Sonic 3.5

Hume Octave 2 is not in the Versely catalog. Cartesia Sonic 3.5 is: 4 credits per 1,000 characters, text-to-audio and voice-clone, multilingual, emotion control, speed tuning. If a search brought you here on “Hume Octave 2,” this page is the substitution job — what Octave 2 is on Hume’s own blog, and which TTS rows we actually serve. It is not a second emotional-fidelity essay.

Hume is also closing the API that serves Octave 2. On 2 October 2026 it announced that it is sunsetting its TTS and EVI APIs, with access ending on 13 November 2026 (Hume changelog). Do not start a new project on Octave 2. The hosted alternatives below are the ones to pick from.

Hume’s launch note (Hume Blog, 1 October 2025) says Octave 2 extends TTS to 11 languages, is 40% faster and more efficient than Octave 1 (audio in under 200ms on their stack), and is half the price of Octave 1. Languages listed: Arabic, English, French, German, Hindi, Italian, Japanese, Korean, Portuguese, Russian, and Spanish. New controls they name: voice conversion and direct phoneme editing; more reliable pronunciation of uncommon words, numbers, and symbols. Those are Hume’s claims on Hume’s API. They are not Versely stickers. We do not publish an /models/hume-octave-2 slug. Do not invent one.

This is not Hume Octave 2: When Emotional TTS Beats a Flat Voice. That page is when emotional direction is the deliverable. This page is: Octave 2 is not here; Cartesia Sonic 3.5 is 4 credits per 1,000 characters; Inworld and Gemini Flash TTS sit next to it. If you need Octave 2 itself, you leave this catalog for Hume’s stack, and only until 13 November 2026. If you need a voice on this bill, you pick a row we host.

What Octave 2 is, on Hume’s blog

Treat the launch post as a vendor card, not a catalog card.

• Languages. 11 at launch, with a note that more were coming. Do not write “100+” onto Octave 2 from a different vendor’s row.
• Latency. Hume says under 200ms, 40% faster than Octave 1. That number is theirs. It is not a Versely benchmark.
• Price. Half of Octave 1, on Hume. Third-party USD stays on Hume. Versely prices in credits. Cartesia Sonic 3.5 is 4 credits per 1,000 characters here. That is the comparable object.
• Controls. Hume’s plain-English description field for acting instructions is documented for Octave 1 only; its docs list Octave 2 support as “coming soon” (Hume docs). Voice conversion and phoneme editing are named as new. We do not host those controls.

Octave 2 was the right engine for a project built around emotional performance — audiobook performance, a character voice, a support line where the feeling is the product. Those jobs still exist. They run on Hume’s API until 13 November 2026, not on a Versely model slug, and Hume’s API is closing, so there will not be one. Do not wait for a row we have not published.

GSC will still query “Hume Octave 2.” An honest page says the substitution. A dishonest page pretends we ship it.

What to run here: Cartesia Sonic 3.5 at 4 credits per 1,000 characters

The Cartesia card: cartesia-sonic-3-5, text-to-audio and voice-clone, 4 credits per 1,000 characters. Description: multilingual, highly expressive, emotion control, speed tuning, voice cloning. Cartesia’s newer Sonic 3.6 is the same 4 credits. That is the default substitution when the brief was “a clean, fast, expressive read” and the search term was Octave 2.

Use Sonic 3.5 when:

• The script is an ad read, a product line, an IVR-adjacent VO, a UGC wrap, a pack caption that needs a voice.
• You want cloning on the same family. AI voice cloning costs no credits to create, with up to 3 saved clones per voice engine. Clone, then read on 4 credits per 1,000 characters.
• You need emotion and speed as Cartesia exposes them, not Hume’s acting instructions.

Do not use Sonic 3.5 as a fake student, a fake pharmacist, or a fake CEO. TTS is a voice. It is not a licence, a patient, or an officer. Label a synthetic voice when the platform requires it. Article 50 is a video-and-deepfake label regime; a disembodied VO is still an ad claim. The FTC’s endorsement rules still apply if the voice implies a person who does not exist.

Four credits per 1,000 characters is Versely’s catalog price. It is not comparable to Hume’s “half of Octave 1”, because they are different meters.

After the WAV: lay it on a mute clip, or on a native-audio take you will replace. Wan 2.6 Image to Video Flash starts at 20 credits, silent I2V, if the picture is a still. Kling 3 Turbo Text to Video starts at 22 credits and is silent if the picture is a sentence. Native-audio video rows can speak; stacking Sonic on a talking take without a plan is how you get two voices.

Captions still matter. A Sonic read that autoplays mute is a burned-in line. The AI caption generator is the typesetting. Spell SKUs as the pack spells them. Sonic will not save a homophone in the subtitle file.

Inworld and Gemini Flash TTS on the same bill

If Cartesia is the wrong flavour, stay in the catalog. Do not bounce to Hume: its TTS API closes on 13 November 2026.

Inworld TTS 2 Flash is 2 credits per 1,000 characters, text-to-audio. Description: the low-latency tier of TTS-2, with voice cloning, best for high-volume and conversational use. Two credits is half of Sonic’s four.

Inworld TTS 2 is 2 credits per 1,000 characters, text-to-audio. Description: research preview, 100+ languages, natural-language style steering. That last clause is the closest hosted cousin to Hume’s acting instructions. Half the sticker of Sonic. Different vendor. If the search for Octave 2 was really a search for natural-language style steering on this bill, Inworld TTS 2 is the row to try.

The older Inworld TTS and TTS 1.5 Max rows are not in the catalog now; Flash is the cheap Inworld door.

Gemini 3.8 Flash TTS is 2 credits per 1,000 characters, text-to-audio. Description: 30 voices, natural-language style direction, inline audio tags ([sigh], [laughs]), multilingual, two-speaker dialogue. If the brief was tags in the script rather than a prose direction, Flash TTS is the tag row. The older Gemini 3.1 Flash TTS is 12.

Pick by control surface, then by credits:

• Tags in the script. Gemini 3.8 Flash TTS, 2 credits.
• Natural-language style steering, hosted. Inworld TTS 2, 2 credits.
• Expressive multilingual plus clone path. Cartesia Sonic 3.5 at 4, clone free to create.
• Cheapest hosted Inworld. Inworld TTS 2 Flash at 2.
• Hume Octave 2’s specific stack (11 langs, phoneme edit, voice conversion, their latency). Hume’s API, until 13 November 2026. Not a Versely slug.

Eleven Labs Voice Change starts at 20 credits (4 per second of output), audio-to-audio. That is conversion of an existing take, not TTS from a script. Do not use it as an Octave 2 stand-in. It is a different job.

Do not invent a Sonic 4 slug; there is no Cartesia Sonic 4. The newest Cartesia row here is Sonic 3.6, and Sonic 3.5 is 4 credits per 1,000 characters.

Substitution, not a second review

A team that already built Octave 2 into a pipeline has until 13 November 2026 to move it. Save the audio and voice samples you need before then, and add Cartesia or Inworld where the rest of the slate already runs on this bill: captions, mute I2V, a shared credit pool. Dual-running a voice API and a video catalog is normal. Pretending they are the same row is not.

A team that searched Octave 2 because they needed a voice for UGC wraps should not open a second vendor. Four credits per 1,000 characters of Sonic, or two of Gemini 3.8 Flash TTS, plus a silent picture row, is a complete wrap. The UGC video generator is the assembly. The voice is a catalog pick.

YouTube’s inauthentic-content policy, as reported by Mashable on 21 July 2026, will not monetize synthetic personas presenting as human authorities on healthcare, finance, law, and politics. A TTS voice reading a “doctor” script under a generated coat is that bucket. Sonic does not create a licence. Octave 2 would not either.

New York’s synthetic-performer law (N.Y. GBL § 396-b, in force 9 June 2026) is aimed at generated human performers in ads, not at audio-only ads; the statute exempts audio-only. A talking head with a cloned voice is not audio-only. Stay honest about the picture track.

If the job is Octave 2, Hume’s API is open only until 13 November 2026. If the job is a read on this catalog, start on Cartesia Sonic 3.5 at 4 credits per 1,000 characters, or Inworld TTS 2 at 2 if you wanted natural-language steering without leaving.

Run the hosted read in Versely.

FAQ

Is Hume Octave 2 available on Versely?

No. There is no published Octave 2 slug. Hume’s own API is the only route for that model, and it closes on 13 November 2026. Here the default substitution is Cartesia Sonic 3.5 at 4 credits per 1,000 characters.

Which hosted row is closest to Octave’s emotional direction?

Inworld TTS 2 and Gemini 3.8 Flash TTS, both 2 credits per 1,000 characters, take natural-language style direction; Gemini also uses inline tags. Neither is Octave 2.

Should I clone on Cartesia or stay on Hume for a character?

If the character is already an Octave 2 voice on Hume, you have until 13 November 2026 to move it. If the character needs to live next to captions and mute I2V on one bill, clone on Cartesia (free to create, up to 3 saved clones per voice engine), then read on Sonic 3.5 at 4 credits per 1,000 characters.