whatmodel.sh / tts / Eleven v3
Eleven v3
↗ share
+ contribute
Eleven v3 is a closed-weights text-to-speech model from ElevenLabs, released February 2026. Licence: proprietary.
Score within the Vapi Humanness · Bradley–Terry rating, 2026-09-25 snapshot comparison set only.
$ how it compares
The text-to-speech models ranked either side of Eleven v3, out of 148 with a score, then more from ElevenLabs.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 19 |  MiniMax Speech 2.8 HD | 66±2 | api $100.00/M chars | ||
| 20 |  dots.tts SOAR 2B | 65±2 | local | ||
| 21 | Eleven v3 | 65±1 | api $100.00/M chars | ||
| 22 |  Chirp 3: HD | 65±4 | api $30.00/M chars | ||
| 23 |  Higgs Audio V3 TTS | 65±3 | api $15.00/M chars +local | ||
| more from ElevenLabs | |||||
| 30 | Eleven Turbo v2.5 | 64±4 | api $50.00/M chars | ||
| 38 | Eleven Flash v2.5 | 63±2 | api $50.00/M chars | ||
| 45 | Eleven Flash v2 | 62±3 | api $50.00/M chars | ||
see all text-to-speech models · find the ones that fit your machine
$ what it's good at
voice cloning█···█······█ #3 of 3 · 11% of its score · 4 tests
Trails Eleven Flash v2.5 here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Seed-TTS · report split unspecified, speaker similarity | 39.2% ✎ | ██·········█ beats 0/2 | 3.6% |
| MiniMax TTS · 9-language speaker similarity | 49.7% ✎ | ██·········█ beats 0/2 | 3.5% |
| MiniMax TTS · 9-language WER (Voxtral ASR) | 2.2% ✎ | █····█·····█ beats 0/2 | 1.9% |
| Seed-TTS · report split unspecified (Voxtral ASR) | 1.3% ✎ | █··········▄ beats 0/2 | 1.7% |
other tests16 not in a task · 89% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Vapi Humanness · Bradley–Terry rating, 2026-09-25 snapshot | 1281.1 ✎ | ▁···▁▃··█▄▁▇ beats 18/18 | 16% |
| CV3-Eval hard-en · WER | 5.8% ✎ | █····█·█████ beats 6/6 | 8.2% |
| Vapi Humanness · historical Elo, 2026-06-10 | 1231.8 ✎ | ▃·▃▃▃·▃██▃·▃ beats 9/12 | 8.2% |
| ZTTS1-Eval in-the-wild · 10-language mean WER | 2.4% ✎ | ▄·····█··▄██ beats 6/7 | 7.1% |
| CV3-Eval hard-en · DNSMOS (predicted quality) | 3.92 / 5 ✎ | █·▄·▄▄·▄···▄ beats 5/6 | 6.5% |
| ZTTS1-Eval Clean (FLEURS-R) · 9-language mean WER | 3.8% ✎ | ▄·····▄█▄█·▄ beats 5/7 | 6.2% |
| ZTTS1-Eval in-the-wild · 10-language mean MSR-UTMOS | 3.39 / 5 ✎ | █··▄█▄····▄▄ beats 6/7 | 6.0% |
| CV3-Eval hard-zh · CER | 10.7% ✎ | ▄▄▄···▄··▄·█ beats 2/6 | 5.8% |
8 more tests
| test | result | vs the others in its run | counts |
|---|---|---|---|
| CV3-Eval multilingual voice clone · 9-language mean error | 5.1% ✎ | ▄···▄··▄··██ beats 3/6 | 5.6% |
| ZTTS1-Eval Clean (FLEURS-R) · 9-language mean MSR-UTMOS | 3.35 / 5 ✎ | ▃▃█···▃···▃▃ beats 5/7 | 5.2% |
| CV3-Eval protocol, 7 added languages · mean WER (Qwen-Audio 3.0 extension) | 5.1% ✎ | █··█······██ beats 0/3 | 3.9% |
| CV3-Eval hard-en · speaker similarity (ERes2Net SV) | 48.9% ✎ | ▄·····▄▄·▄▄█ beats 0/6 | 3.0% |
| RW-Voice-EQ TTS · Language Code-Switching (human) | 3.33 / 5 ✎ | █··▄·······▄ beats 0/3 | 2.8% |
| Voice-First Nation · 10 Indian languages overall pairwise preference (Bradley–Terry) | 1056.3 ✎ | ▄····▄·····█ beats 3/3 | 2.3% |
| RW-Voice-EQ TTS · Reliability (human) | 3.59 / 5 ✎ | █·▄······▄·▄ beats 0/4 | 2.1% |
| Amharic TTS · 87-prompt CER (Chirp 3 judge) | 111.7% ✎ | ▄·····▄·▄·██ beats 0/6 | 0.0% |
$ where to run it
Through an API
api · $100.00/M chars · elevenlabs.io ↗ · price sourceUSD per 1M input characters: ElevenLabs lists v3 Text to Speech at $0.10 per 1K characters (x1000). Product: ElevenAPI text to speech. API model: eleven_v3. Settings: standard API rate (same per-1K rate on pay-as-you-go and plan tiers shown). Price basis: published list rate, per-1K to per-1M scaling only. Exclusions: subscription fees, taxes. Currency: USD. ElevenLabs official API pricing page, observed 2026-09-25.
$ how the score was worked out · Qwen-Audio-3.0-TTS technical… 33%, ZONOS2 Technical Report (ZTTS… 24%, The Humanness Index by Vapi ·… 16% and 5 more
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 Qwen-Audio-3.0-TTS technical report | 6 | 33% |
| ZONOS2 Technical Report (ZTTS1-Eval results) | 4 | 24% |
 The Humanness Index by Vapi · September 2026 public standings | 1 | 16% |
| Voxtral TTS | 4 | 11% |
| The Humanness Index by Vapi · June 2026 production export | 1 | 8% |
| RW-Voice-EQ Bench (Hume AI) | 2 | 5% |
| Preferences of a Voice-First Nation: pairwise human evaluation of TTS in 10 Indian languages (arXiv:2604.21481v2) | 1 | 2% |
| Addis AI Amharic TTS benchmark | 1 | 0% |
| prior · pull toward the middle | 1% | |
| weighted average | = 65.1 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json