Eleven v4 is ElevenLabs’ most emotive text-to-speech model, launched on September 28, 2026 alongside a low-latency sibling, Eleven v4 Turbo. It is also the model to beat on independent votes: on the Artificial Analysis Speech Arena, Turbo sits at #1 and v4 at #2, ahead of every other vendor’s text-to-speech model and well ahead of its predecessor Eleven v3, now #18.
Two models, two jobs
| Eleven v4 | Eleven v4 Turbo | |
|---|---|---|
| Built for | Produced content — audiobooks, voice-over, dubbing, character performances | Agents and interactive apps |
| Speed | Not listed by ElevenLabs | ~100 ms median inference; ~150 ms to first speech over WebSocket (vendor-measured) |
| Request limit | 10,000 characters | Not listed |
| Languages | 90+ | 90+ |
| API price | $0.08 per 1,000 characters | $0.04 per 1,000 characters |
| Arena rank (Artificial Analysis) | #2, Elo 1326 | #1, Elo 1330 |
ElevenLabs says it optimized Turbo and its ElevenAgents platform together as one system, so teams building agents on that stack get a model tuned for the loop they will run it in.
What changed
- Direction by tag. Inline audio tags such as
[laughs],[excited, happy]or[said angrily in French accent]shape delivery, and tags such as[light rain]or[phone buzzing]add sound effects. You can also describe how a line should sound in plain language. IPA phoneme support is stronger, so custom pronunciations hold up better. - Multi-speaker scenes. v4 uses a new method for capturing speaker identity and generates dialogue where speakers respond to what was just said, instead of stitching isolated lines together. It is served through the Text to Dialogue API.
- One voice, many languages. A voice recorded in one language speaks any of the 90+ supported languages with a native accent while keeping its identity. ElevenLabs says accent drift over a long generation is much reduced.
- Cloning. Instant Voice Clones work from about 10 seconds of audio, and Professional Voice Clones are supported. Request stitching for long-form work is more reliable than in v3.
How it ranks
| Artificial Analysis Speech Arena (October 2026) | Elo | Rank | Price per 1M characters |
|---|---|---|---|
| Eleven v4 Turbo | 1330 | #1 | $40.00 |
| Eleven v4 | 1326 | #2 | $80.00 |
| Qwen-Audio 3.1 TTS-Plus | 1298 | #3 | $19.30 |
| Cartesia Sonic 3.6 | 1286 | #4 | $49.00 |
| Gemini 3.8 Flash TTS | 1274 | #5 | $16.50 |
| Eleven v3 | 1177 | #18 | $100.00 |
The arena scores each provider’s own native voices (eight per model) from blind listener votes, and the two Eleven v4 scores are four points apart against error bars of about ±17, so treat them as tied. Eleven v4 leads the third-place model by 28 points. Artificial Analysis lists Eleven v3 at $100 per 1M characters, while ElevenLabs’ own price list shows $0.08 per 1,000 characters — the same as v4.
ElevenLabs separately reports that about 75% of listeners preferred v4 in its own blind tests against Cartesia Sonic 3.6, Inworld TTS-2 and two Gemini TTS models. That is a vendor-run test and we have not verified it.
Pricing and access
ElevenAPI bills $0.08 per 1,000 characters for v4 and $0.04 for v4 Turbo. Both are available on every plan including the free one, and through ElevenAPI, ElevenAgents and ElevenCreative. For comparison, Artificial Analysis lists Gemini 3.8 Flash TTS at $16.50 and Qwen-Audio 3.1 TTS-Plus at $19.30 per 1M characters.
Who it’s for
Choose Eleven v4 when delivery matters as much as the words: audiobooks, character voices, advertising, dubbing and localization where a brand voice must hold across languages. Choose Turbo for voice agents, where it ranks first and costs half of v4. If cost per character matters more than the last few Elo points, Qwen-Audio 3.1 and Gemini 3.8 Flash TTS sit about 30–60 points behind at a fraction of the price.
Best for
- Audiobooks, voice-over and character performances where delivery matters (v4)
- Voice agents and live phone flows (v4 Turbo, about 150 ms to first speech)
- Dubbing and localization with one brand voice across languages
- Games and animation with emotional range and multi-speaker scenes
Pros & cons
Strengths
- Ranks #1 (Eleven v4 Turbo, Elo 1330) and #2 (Eleven v4, 1326) on the Artificial Analysis Speech Arena, ahead of Qwen-Audio 3.1 and Gemini 3.8 Flash TTS
- Inline audio tags — [laughs], [excited, happy], even sound effects like [light rain] — plus natural-language delivery direction and better IPA phoneme support
- Natural multi-speaker dialogue with each speaker's identity preserved across a scene
- 90+ languages, with one voice carrying its identity and a native accent across all of them
- Instant Voice Clones from about 10 seconds of audio, and Professional Voice Clones on both models
Limitations
- The priciest of the leaders — $80 per 1M characters for v4, about 4x Qwen-Audio 3.1 TTS-Plus and nearly 5x Gemini 3.8 Flash TTS on Artificial Analysis's figures
- Eleven v4 requests are capped at 10,000 characters, so books and long scripts need request stitching
- The two arena scores are within each other's error bars, so Turbo at half the price is the better buy unless you need the produced-content tuning
- Closed and hosted only — no open weights and no self-hosting
Frequently asked questions
What is Eleven v4?
Eleven v4 is ElevenLabs' text-to-speech model family, launched on September 28, 2026 in two variants. Eleven v4 is tuned for produced content such as audiobooks, voice-over and dubbing. Eleven v4 Turbo is the low-latency variant for agents and interactive apps, with a median inference latency of about 100 ms and, in ElevenLabs' own measurement, a median time to first speech of about 150 ms. Both support more than 90 languages.
How much does Eleven v4 cost?
ElevenAPI lists $0.08 per 1,000 characters for Eleven v4 and $0.04 per 1,000 characters for Eleven v4 Turbo. Artificial Analysis converts those to $80 and $40 per 1M characters. The free plan includes 10,000 characters of v4 a month, and the Starter plan is $6 a month.
How does Eleven v4 rank against other text-to-speech models?
On the Artificial Analysis Speech Arena (blind listener votes, October 2026), Eleven v4 Turbo is
Is Eleven v4 better than Eleven v3?
In blind votes, yes by a wide margin — about 150 Elo points on Artificial Analysis. v4 also supports 90+ languages versus 70+ and a 10,000-character request limit versus 5,000, at the same $0.08 per 1,000 characters ElevenLabs lists for v3.
What do audio tags do?
They are inline instructions in the text — for example [excited, happy], [sleepy drowsy voice], [laughs] or [said angrily in French accent] — that change how a line is delivered, and can also add sound effects such as [light rain] or [phone buzzing]. ElevenLabs says v4 follows them more accurately than earlier models.
Can I clone my voice with Eleven v4?
Yes. Instant Voice Clones can capture a voice from about 10 seconds of audio, and Professional Voice Clones are supported for the highest-fidelity cases.