XTTS v2 text-to-speech model
CoquiOpenReleased Nov 2023.
XTTS v2 is a text-to-speech (TTS) model from Coqui. It ranks #90 of 98 on the Artificial Analysis TTS arena with Elo 915 (95% CI ±15, 2,703 samples). List price is about $40.4 per 1M characters. Weights are openly available for self-hosting. Released Nov 2023.
Key facts
- Provider
- Coqui
- AA rank
- #90 / 98
- AA Elo (95% CI)
- 915 ±15
- AA samples
- 2,703
- TTS Arena Elo (uncertainty)
- –
- TTS Arena votes
- –
- TTS Arena win rate
- –
- Price (per 1M chars)
- $40.4
- Released
- Nov 2023
- Open weights
- Yes
- License
- –
- Voices tested by AA
- –
- Languages
- –
- Voice cloning
- –
"–" means the upstream sources do not provide this value; we do not guess.
Nearby models
By AA Elo; this model highlightedAlternatives · Cheaper with AA Elo ≥ 885
| # | Model | AA Elo | Arena Elo | Price $/1M chars | Open weights |
|---|---|---|---|---|---|
| 1 | 1336 | – | $40 | – | |
| 3 | 1307 | – | $19.3 | – | |
| 5 | 1280 | – | $16.5 | – | |
| 6 | IRealtime TTS-2 Inworld | 1270 | – | $20.8 | – |
| 7 | 1266 | – | $19.3 | – |
Alternatives · Top 5 open-weight by AA Elo
| # | Model | AA Elo | Arena Elo | Price $/1M chars | Open weights |
|---|---|---|---|---|---|
| 9 | 1248 | – | $15 | Open | |
| 11 | BBreeze TTS 2 BreezeBlue | 1230 | – | $34 | Open |
| 19 | 1191 | – | – | Open | |
| 38 | 1110 | – | $15 | Open | |
| 41 | 1100 | – | – | Open |
FAQ
Where does XTTS v2 rank among TTS models?
As of Oct 25, 2026, XTTS v2 ranks #90 of 98 on the Artificial Analysis TTS leaderboard with Elo 915 (±15).
Is XTTS v2 open source?
Yes. XTTS v2 has open weights and can be self-hosted.
What is the best Chinese TTS model?
By AA Elo, the top model from a Chinese provider is Qwen-Audio-3.1-TTS-Plus by Alibaba (Elo 1307, #3 overall).
Which top-10 TTS model is cheapest?
Among the AA top 10, the lowest list price is Simba 3.2 (SpeechifyAI) at $6.6 per 1M characters, Elo 1246.
What is the best open-source TTS model?
By AA Elo, the top open-weight model is Fish Audio Drama 3 Preview by Fish Audio (Elo 1248, #9 overall); 18 open-weight models have an AA score.
Sources
"–" means the upstream sources do not provide this value; we do not guess.