ElevenLabs v3 was the company’s flagship expressive text-to-speech model: dramatic delivery, emotional range and multi-speaker dialogue across 70+ languages. On September 28, 2026 ElevenLabs replaced it with Eleven v4, and independent blind votes now place it well behind: #18 on the Artificial Analysis Speech Arena at Elo 1177, against 1330 and 1326 for Eleven v4 Turbo and Eleven v4.
For new projects, start with Eleven v4. It lists at the same $0.08 per 1,000 characters, supports 90+ languages and 10,000 characters per request, and is far ahead in listener preference.
What is still in the lineup
| Model | Best for | Languages | Speed | Request limit | API price per 1,000 characters |
|---|---|---|---|---|---|
| Eleven v3 | Expressive dialogue and performances | 70+ | — | 5,000 characters | $0.08 |
| v3 Conversational | Real-time voice agents | 70+ | ~280 ms | — | $0.04 |
| Multilingual v2 | Stable long-form narration | 29 | — | 10,000 characters | $0.08 |
| Flash / Turbo v2.5 | Low-cost real-time use | 32 | ~75 ms | 40,000 characters | $0.04 |
Speed figures exclude application and network latency, per ElevenLabs.
Where it ranks
On the Artificial Analysis Speech Arena (October 2026), Eleven v3 is #18 (Elo 1177), v3 Conversational is #15 (1210), Turbo v2.5 is #39 (1100), Multilingual v2 is #41 (1096) and Flash v2.5 is #49 (1078). Artificial Analysis shows prices of $50–$100 per 1M characters for these models, which differ from ElevenLabs’ own list prices above, so use the list prices for budgeting.
When it still makes sense
Keep v3 if you have prompts and voices tuned to it and no reason to re-test, or if you rely on the cheaper Conversational tier. For anything new, compare against Eleven v4, Qwen-Audio 3.1 and Gemini 3.8 Flash TTS, or see the best AI voice generators.
Best for
- Existing integrations and voices that already work well on v3
- Dialogue scripts and character reads where you have tuned prompts to v3
- Real-time apps on the cheaper v3 Conversational tier (about 280 ms)
Pros & cons
Strengths
- Expressive, emotionally rich delivery with audio tags and multi-speaker dialogue, across 70+ languages
- Still priced at $0.08 per 1,000 characters, with a cheaper $0.04 Conversational tier at about 280 ms for real-time use
- A mature ecosystem — a documented REST API, voice cloning and an established voice library
Limitations
- Superseded by Eleven v4, which ranks about 150 Elo points higher on the Artificial Analysis Speech Arena at the same list price
- Single requests are capped at 5,000 characters, versus 10,000 for v4, so longer scripts need splitting
- 70+ languages, against 90+ for v4