What ElevenLabs publishes about Eleven v3, its credits and its licences.
ElevenLabs is a voice and audio platform, and Eleven v3 — model id `eleven_v3` — is the flagship speech model its own documentation calls “Our most emotionally rich, expressive speech synthesis model”. It is one of five speech models whose limits ElevenLabs publishes separately: Multilingual v2, Flash v2.5, Flash v2 and v3 Conversational each carry their own language count and characters-per-request ceiling, so no one figure covers the family.
Figures on this page come from ElevenLabs’ own pages: the model documentation and changelog for model names, ids, limits and dated entries; the pricing page and API pricing page for plans, credits and per-character rates; and the voice-cloning guides and the help-centre licensing page for cloning rules and commercial-use terms. ElevenLabs sets every one of these figures and changes them without notice.
How ElevenLabs describes Eleven v3 and the models around it
ElevenLabs publishes its speech models as a list with per-model limits rather than as one model with one specification, and Eleven v3 is the model the vendor’s own wording singles out. The commercial model is equally specific: every one of its products draws on a single shared monthly credit pool, and different model families consume credits at different rates.
Eleven v3 is the model the vendor leads with
The current speech line is `eleven_v3`, `eleven_v3_conversational`, `eleven_ttv_v3`, `eleven_multilingual_v2`, `eleven_flash_v2_5`, `eleven_flash_v2`, `eleven_multilingual_sts_v2`, `eleven_multilingual_ttv_v2` and `eleven_english_sts_v2`. ElevenLabs marks `eleven_turbo_v2_5`, `eleven_turbo_v2` and `scribe_v1` as deprecated, and describes `music_v1` as “Outclassed by `music_v2` and `music_v2_5`”.
Language counts belong to models, not to the family
Eleven v3 and v3 Conversational cover 70+ languages. Multilingual v2 covers 29. Flash v2.5 covers 32 — the 29 plus Hungarian, Norwegian and Vietnamese — and Flash v2 is `en` only. Reading any one of those numbers as an ElevenLabs-wide figure would misstate the others.
Character ceilings are per model as well
Eleven v3 accepts 5,000 characters per request and Multilingual v2 accepts 10,000. Flash v2.5 accepts 40,000 and Flash v2 accepts 30,000. ElevenLabs publishes no characters-per-request figure at all for `eleven_v3_conversational`, and states no latency for `eleven_v3` or `eleven_multilingual_v2`.
One credit pool, several per-character rates
Plans run from Free at $0 with 10,000 credits to Business at $990 with 6,000,000 credits, and “All products draw from one shared monthly credit pool.” A character costs 1 credit on Eleven v3; “For V2 Multilingual models, 1 text character equals 1 credit”; and V2 Flash/Turbo English and V2.5 Flash/Turbo Multilingual models consume “between 0.5 and 1 credit per character”.
The licence turns on whether you paid
ElevenLabs states that “The free plan does not include a commercial license and cannot be used for any commercial purpose,” while “All paid plans include a commercial license, provided you’re not using Beta Services.” It adds that content created outside a paid subscription, before or after, “cannot be used commercially and always requires attribution when shared non-commercially.”
What ElevenLabs documents
This section carries only what ElevenLabs’ own documentation states, including the places where it states nothing. Where a figure is missing — a character limit for v3 Conversational, a latency for Eleven v3 — the gap is shown rather than filled.
Speech synthesis sold as a set of models
Text to speech is published as a line rather than a single model: `eleven_v3`, `eleven_v3_conversational`, `eleven_ttv_v3`, `eleven_multilingual_v2`, `eleven_flash_v2_5` and `eleven_flash_v2`, alongside `eleven_multilingual_sts_v2` and `eleven_english_sts_v2` for speech to speech. Because the limits differ, the choice is made on the model table rather than on one headline specification.
Per-model languages and request limits
Eleven v3: 70+ languages, 5,000 characters per request. Multilingual v2: 29 languages, 10,000 characters. Flash v2.5: 32 languages, 40,000 characters. Flash v2: `en` only, 30,000 characters. Eleven v3 Conversational: 70+ languages, with no published character limit and a stated latency of roughly 280ms. Latency is the one column the documentation fills unevenly — v3 Conversational, Flash v2.5 and Flash v2 publish a figure (about 280ms, 75ms and 75ms respectively), while eleven_v3 and Multilingual v2 publish none. Every figure the docs give excludes application and network latency.
Model-specific behaviour the vendor calls exclusive
ElevenLabs lists multi-speaker dialogue and audio tags as Eleven v3 features. For Multilingual v2 it claims “Most stable on long-form generations” and the best number normalization. For Flash v2.5 it cites “50% lower price per character for API generations” and notes that normalization is off by default.
Voice cloning sits behind consent verification
Instant Voice Cloning asks you to “confirm that you have the right and consent to clone the voice”. Professional Voice Cloning requires a Creator plan or above and uses what ElevenLabs calls “voice-captcha technology”; for fine-tuning the vendor states that “we only allow you to clone your own voice” and that you “will be asked to go through a verification process before submitting your fine-tuning request”. The restriction is absolute in its wording: “Even with their consent, you cannot clone someone else’s voice.”
Speech to text, music and sound effects
`scribe_v2` is documented at 90+ languages, 1,000 keyterms and 32-speaker diarization, with `scribe_v2_medical` and `scribe_v2_realtime` alongside it; `scribe_v1` is deprecated. Music runs as `music_v2_5`, `music_v2` and `music_v1`, and sound effects as `eleven_text_to_sound_v2`. For these ElevenLabs publishes duration and format limits rather than character limits.
The wider product surface, in the vendor’s names
ElevenLabs names Dubbing (v1, v2), Sound Effects, Eleven Music, Speech to Text (Scribe), Text to Dialogue, Speech Engine, ElevenAgents, Voice Changer, Voice Isolator, Voice Design, Voice Remixing, Forced Alignment, Image & Video, Ads Engine, Productions, ElevenReader and Reception.ai. Those are the vendor’s own product names, and their limits sit outside the speech table above.
Documented specifications
Every row below is stated on ElevenLabs’ own documentation or pricing pages, and each limit is attached to the model id it belongs to rather than to the platform as a whole.
- Developer
- ElevenLabs
- Flagship speech model
- Eleven v3 (`eleven_v3`)
- `eleven_v3`
- 70+ languages · 5,000 characters per request · latency not published
- `eleven_multilingual_v2`
- 29 languages · 10,000 characters per request · latency not published
- `eleven_flash_v2_5`
- 32 languages · 40,000 characters per request · ~75ms excluding application & network latency
- `eleven_flash_v2`
- `en` only · 30,000 characters per request · ~75ms excluding application & network latency
- `eleven_v3_conversational`
- 70+ languages · characters per request not published · ~280ms excluding application & network latency
- Credits per character, `eleven_v3`
- 1 credit per character
- Credits per character, V2 families
- V2 Multilingual: 1 text character equals 1 credit · V2 Flash/Turbo English and V2.5 Flash/Turbo Multilingual: between 0.5 and 1 credit per character
- API price per 1K characters
- $0.10 for Eleven v3 and v2 Multilingual · $0.05 for v3 Conversational and Flash/Turbo
- Plans, monthly
- Free $0 (10,000 credits) · Starter $6 (30,000 credits) · Creator $22 (121,000 credits, $11 for the first month) · Pro $99 (600,000 credits) · Scale $299 (1,800,000 credits, 3 seats) · Business $990 (6,000,000 credits, 10 seats) · Enterprise is custom · all products draw from one shared monthly credit pool
- Instant Voice Cloning audio
- “Record at least 1 minute of audio”; approximately 1-2 minutes of clear audio is recommended
- Professional Voice Cloning audio
- “The bare minimum we recommend is 30 minutes of audio, but for the optimal result … closer to 2-3 hours of audio”, with 3-6 hours for fine-tuning · requires a Creator plan or above
- Commercial licence
- Free plan: “does not include a commercial license and cannot be used for any commercial purpose” · paid plans: “include a commercial license, provided you’re not using Beta Services”
Documented access channels
Eleven v3 is reached through ElevenLabs’ own platform and API. The pages below are the vendor’s own, and between them they carry every figure quoted on this page.
Model documentation
The list of current and deprecated models, with the languages and characters-per-request figure for each model id, including the ones ElevenLabs leaves blank.
OpenPlans and credit allowances
The seven tiers, their monthly credit allowances and the statement that all products draw on one shared pool.
OpenAPI pricing
The published per-character rate cards for Eleven v3, Multilingual v2, v3 Conversational and the Flash/Turbo models.
OpenCommercial-use policy
ElevenLabs’ own help-centre page on publishing what you generate, which carries the free-versus-paid licence split quoted on this page.
OpenElevenLabs and Eleven v3 FAQ
Which model leads the line, and what are its limits?
Eleven v3, id `eleven_v3`, which ElevenLabs’ model documentation calls “Our most emotionally rich, expressive speech synthesis model”. It is documented at 70+ languages and 5,000 characters per request, and ElevenLabs publishes no latency figure for it. It became available through the API on 20 August 2025 and came out of alpha on 2 February 2026; the original launch date is not published anywhere in the documentation.
How much does ElevenLabs cost?
Plans are Free at $0 with 10,000 credits, Starter $6 with 30,000, Creator $22 with 121,000 credits and $11 for the first month, Pro $99 with 600,000, Scale $299 with 1,800,000 credits and 3 seats, and Business $990 with 6,000,000 credits and 10 seats; Enterprise is custom. API generation is $0.10 per 1K characters on Eleven v3 and v2 Multilingual and $0.05 per 1K characters on v3 Conversational and Flash/Turbo. All products draw from one shared monthly credit pool, and a character costs 1 credit on Eleven v3.
Can I use ElevenLabs output commercially?
It depends on the plan the content was generated under, and ElevenLabs’ wording is plain: “The free plan does not include a commercial license and cannot be used for any commercial purpose,” while “All paid plans include a commercial license, provided you’re not using Beta Services.” The vendor also states that content created outside a paid subscription, before or after, “cannot be used commercially and always requires attribution when shared non-commercially,” and that content generated during a paid subscription “can be used commercially, and indefinitely, subject to our Service-Specific Terms.”
What does cloning a voice require?
Consent and verification. Instant Voice Cloning asks you to “confirm that you have the right and consent to clone the voice”; Professional Voice Cloning requires a Creator plan or above, uses what ElevenLabs calls “voice-captcha technology”, and for fine-tuning the vendor says that “we only allow you to clone your own voice” and that you “will be asked to go through a verification process before submitting your fine-tuning request”. It is explicit that permission from a third party is not enough: “Even with their consent, you cannot clone someone else’s voice.”
How much audio does each cloning type need?
Instant Voice Cloning requires “at least 1 minute of audio” as a floor, with approximately 1-2 minutes of clear audio recommended. Professional Voice Cloning is a different order of magnitude: “The bare minimum we recommend is 30 minutes of audio, but for the optimal result … closer to 2-3 hours of audio”, and fine-tuning takes 3-6 hours. ElevenLabs states the Instant Voice Cloning minimum inconsistently across its own pages, so the 1-minute figure is reproduced here as the documented floor.
Why are there no screenshots or example outputs on this page?
Because ElevenLabs publishes no static product figure to reproduce. Its brand page carries logos only — black-and-white PNG and SVG versions of the “11” symbol — and its press page adds two photographs of people, labelled “Co-founders” and “Team”. Everything else in its documentation is interface screenshots, which are not product output. A page with no gallery is better than one padded with images we have no right to reproduce.