AIDealsPro is reader-supported. When you buy through links on this page, we may earn an affiliate commission at no extra cost to you. See How we review for how we evaluate AI tools.
ElevenLabs currently offers five main TTS models: Eleven v4 (most emotive, 90+ languages, 10,000 chars), Eleven v4 Turbo (real-time, ~100ms), Eleven v3 (expressive, 70+ languages, 5,000 chars), Multilingual v2 (29 languages, 10,000 chars), and Flash v2.5 (fastest, 32 languages, 40,000 chars) (checked 4 Nov 2026). Older Turbo models are deprecated in favor of Flash.
Best for: anyone choosing a voice engine — v4 for emotion, Flash v2.5 for speed and volume, Multilingual v2 for stability.
Not for: users who want a single "best" model; each engine trades quality, speed, languages, and cost differently.
Starting price: Free $0 (10,000 credits/mo) (checked 4 Nov 2026); all models are available on every plan including Free.
The five TTS models at a glance
ElevenLabs’ official docs (last updated October 16, 2026) describe five current text-to-speech models. Here is how they compare on the dimensions that matter:
| Model | Best for | Languages | Chars/request | Latency |
|---|---|---|---|---|
| Eleven v4 | Most emotive, highest quality | 90+ | 10,000 | Standard |
| Eleven v4 Turbo | Real-time voice agents | 90+ | — | ~100ms |
| Eleven v3 | Expressive, audio tags | 70+ | 5,000 | Standard |
| Multilingual v2 | Stable long-form | 29 | 10,000 | Standard |
| Flash v2.5 | Fast, affordable volume | 32 | 40,000 | ~100ms |
All figures checked 4 Nov 2026. Model IDs for API users: eleven_v4, eleven_v4_turbo, eleven_v3, eleven_multilingual_v2, eleven_flash_v2_5.
Eleven v4: the flagship
Eleven v4 is ElevenLabs’ most emotive, high-quality model, launched in September 2026. The docs credit it with exceptional voice cloning, support for 90+ languages, and natural multi-speaker dialogue — the ability to render conversation between distinct voices in one request. Its documented sweet spots are character voiceovers, emotional dialogue, audiobook production, and multilingual projects. The per-request cap is 10,000 characters (~10 minutes).
Choose v4 when the read itself is the product: audiobooks, dramatic narration, character work, or any project where emotional range sells the result. For prompt craft on the expressive models, see our ElevenLabs v3 prompts guide — the audio-tag techniques transfer to v4’s expressive engine.
Eleven v4 Turbo: real-time voice agents
Eleven v4 Turbo is the real-time sibling of the flagship: the docs call it the most emotive real-time model, with median first-audio latency around 100ms across 90+ languages and support for audio tags. Its documented use cases are real-time support agents, AI assistants, and interactive characters — anywhere a voice must respond conversationally with minimal delay.
Because v4 Turbo pairs the flagship’s emotional range with real-time speed, it is the natural default for conversational AI on the ElevenLabs platform. If your agent needs to react mid-sentence or hold a natural back-and-forth, the ~100ms first-audio latency is what makes the exchange feel conversational rather than turn-based.
Eleven v3: expressive with audio tags
Eleven v3 (launched June 2025) is the emotionally rich, expressive model known for dramatic delivery and audio tags — bracketed directions like [laughs] or [whispers] that steer performance mid-script. It supports 70+ languages with a 5,000-character per-request cap, the smallest of the lineup. A conversational variant, v3 Conversational, targets real-time use at around 165ms latency.
v3 remains the pick when you want maximum directorial control over delivery via tags and do not mind splitting longer scripts into 5,000-character chunks.
Multilingual v2: the stable workhorse
Eleven Multilingual v2 is documented as the lifelike, consistent model and the most stable choice for long-form content. It covers 29 languages with a 10,000-character per-request cap. When you need hours of narration that sounds the same at minute 200 as at minute 1 — courses, documentaries, corporate training — Multilingual v2 is the conservative, proven pick.
Flash v2.5: fast and cheap
Eleven Flash v2.5 is the speed and volume play: ultra-low latency (~100ms median), 32 languages (the v2 set plus Hungarian, Norwegian, and Vietnamese), and a 40,000-character per-request cap — the largest in the lineup. For API users, the docs note a 50% lower price per character than larger models. It is the right engine for high-volume, latency-sensitive work: live applications, bulk content pipelines, and real-time features where cost per character dominates.
Deprecated models to avoid
The docs flag several models as deprecated, meaning new projects should not be built on them:
- Eleven Turbo v2.5 and Turbo v2: deprecated — the docs direct users to Flash models instead.
- Music v1 (music_v1): scheduled for removal on January 31, 2027.
- Text to Sound v2 (eleven_text_to_sound_v2): scheduled for removal on March 1, 2027.
If your integration still references a Turbo model ID, migrate to the corresponding Flash model before the removals take effect.
Which model should you pick?
- Audiobooks, character voices, emotional narration: Eleven v4.
- Real-time voice agents and assistants: Eleven v4 Turbo (~100ms).
- Maximum directorial control via audio tags: Eleven v3 (accept the 5,000-char splits).
- Long stable narration (courses, docs): Multilingual v2.
- High volume, low latency, lowest cost: Flash v2.5 (40,000 chars/request).
Plans, each price tagged as checked on 4 Nov 2026: Free $0 (10,000 credits/mo) (checked 4 Nov 2026); Starter $6/mo (30,000 credits) (checked 4 Nov 2026); Creator $22/mo (121,000 credits) (checked 4 Nov 2026); Pro $99/mo (600,000 credits) (checked 4 Nov 2026). All five models are available on every plan, so model choice never forces a plan upgrade — only your monthly character volume does.
How model choice affects your credit bill
Credit billing is refreshingly simple: text to speech costs one credit per character on every model, so a 10,000-character v4 job and a 10,000-character Flash v2.5 job both cost 10,000 credits. The difference is API usage pricing, where Flash v2.5’s price per character is documented as 50% lower — meaningful at bulk volumes. Either way, model choice never forces a plan upgrade: all five models are available on every plan, so you pick the engine for quality and speed, then size the plan for monthly volume.
Pros and cons
Pros
- Five distinct engines cover emotion, real-time, expressiveness, stability, and speed.
- Eleven v4 leads on emotional range with 90+ languages and multi-speaker dialogue.
- Flash v2.5 offers 40,000 characters per request at ~100ms latency and 50% lower API price/char.
- All models available on all plans, including the Free $0 tier (checked 4 Nov 2026).
- Clear deprecation notices with migration targets and removal dates.
Cons
- No single best model — choosing wrong means rework (e.g., v3’s 5,000-char cap for long scripts).
- Most expressive models (v3, v4) have the smallest per-request caps.
- Deprecated Turbo IDs still appear in old tutorials and will break eventually.
- Latency figures are medians for first audio; full-generation time varies with text length.
Who should NOT overthink model choice
If your scripts are short social clips or ad reads, any current model will sound good — pick v4 for quality or Flash v2.5 for speed and move on. Model choice only becomes critical at volume (tens of thousands of characters monthly), for real-time agents (where ~100ms latency matters), or for audiobooks (where emotional range and 90+ language coverage justify v4).
Frequently asked questions
What is the best ElevenLabs TTS model?
There is no single best model — it depends on the job. Eleven v4 is the most emotive and highest quality (90+ languages); Flash v2.5 is the fastest and cheapest per character (40,000 chars/request, ~100ms); Multilingual v2 is the most stable for long-form; v3 offers the most directorial control via audio tags (checked 4 Nov 2026).
What is the difference between Eleven v4 and v3?
Eleven v4 (September 2026) is the newer flagship: most emotive, 90+ languages, 10,000 characters per request, multi-speaker dialogue. Eleven v3 (June 2025) is the expressive model known for dramatic delivery and audio tags, with 70+ languages and a 5,000-character cap. Choose v4 for quality and language coverage; v3 for tag-driven direction.
Which ElevenLabs model is fastest?
Eleven Flash v2.5 and Eleven v4 Turbo both return first audio at a median of around 100ms, the fastest documented figures. Flash v2.5 is the pick for high-volume speed; v4 Turbo adds the flagship’s emotional range to real-time use cases like voice agents and AI assistants.
Are old ElevenLabs models being removed?
Yes, some. Turbo v2.5 and Turbo v2 are deprecated in favor of Flash models; Music v1 is scheduled for removal on January 31, 2027; Text to Sound v2 on March 1, 2027 (checked 4 Nov 2026). If your integration uses a deprecated model ID, migrate before the removal dates.
Do I need a paid plan to use Eleven v4?
No. All five TTS models are available on every plan, including the Free $0 tier with 10,000 credits per month (checked 4 Nov 2026). Plan tiers control your monthly credit allowance and concurrency limits — not which models you can use.
Can I switch models mid-project?
Yes. Model choice is per request, so nothing stops you mixing engines across a project. For a consistent sound, keep one model per continuous piece of audio — and test that the voices match before committing, since different engines can render the same voice ID with slight differences.
What model ID do I use for the ElevenLabs API?
The documented model IDs are eleven_v4, eleven_v4_turbo, eleven_v3, eleven_multilingual_v2, and eleven_flash_v2_5 (checked 4 Nov 2026). Pass the ID in your text-to-speech request; the docs list full specifications, language lists, and character limits per ID.
Sources checked
- ElevenLabs official docs — Models: https://elevenlabs.io/docs/overview/models — checked 4 Nov 2026
- ElevenLabs official docs — overview: https://elevenlabs.io/docs/overview/intro — checked 4 Nov 2026
- ElevenLabs official homepage (v4 launch, model timeline): https://elevenlabs.io/ — checked 4 Nov 2026
- ElevenLabs official pricing: https://elevenlabs.io/pricing — checked 4 Nov 2026
Last updated 4 Nov 2026. Prices checked on elevenlabs.io/pricing on that date; confirm at checkout.