Language tested · read on October 27, 2026
Text-to-speech models tested in Korean
8 of the 96 text-to-speech (TTS) models in this index are filed under Korean , led by Inworld Realtime TTS-2, Cartesia Sonic 3.5 and Rime Coda. Same order and same measures as the full ranking.
| Rank | Model | Speed tier | P95 | Speed (shorter is faster) | Median | Listener Elo | Price per 1M chars | Best for |
|---|---|---|---|---|---|---|---|---|
| 8 |  Inworld Realtime TTS-2Inworld | Fast enough | 194.7 ms | 149.1 ms | 1,267 | $15 to $25 | Customer service, Voice agents, Assistants, Entertainment, Knowledge sharing | |
| 19 |  Cartesia Sonic 3.5Cartesia | Visible delay | 390.5 ms | 279.3 ms | 1,190 | $39.2 to $50 | Customer service, Assistants, Entertainment, Knowledge sharing | |
| 21 |  Rime CodaRime | Visible delay | 428.1 ms | 317.5 ms | 1,054 | $50 | Customer service, Assistants, Entertainment, Knowledge sharing | |
| 37 (tie) |  ElevenLabs Eleven v3 ConversationalElevenLabs | No tier | Not timed | Not timed | Not timed | 1,213 | $50 | Customer service, Assistants, Entertainment, Knowledge sharing |
| 37 (tie) |  Google Gemini 3.1 Flash TTSGoogle | No tier | Not timed | Not timed | Not timed | 1,212 | $18.3 to $33.3 | Customer service, Assistants, Entertainment, Knowledge sharing |
| 46 (tie) | | No tier | Not timed | Not timed | Not timed | 1,126 | $100 | Customer service, Assistants, Entertainment, Knowledge sharing |
| 61 |  Amazon Polly GenerativeAmazon | No tier | Not timed | Not timed | Not timed | 1,063 | $30 | Customer service, Assistants, Entertainment, Knowledge sharing |
| Unranked |  OpenAI GPT-4o mini TTSOpenAI | No tier | Not timed | Not timed | Not timed | Not rated | $20 | Not rated |
Frequently asked questions
- Which Korean TTS model is the fastest?
- Inworld Realtime TTS-2, with a P95 time to first audio of 194.7 ms in Coval's 30-day window. Of the 8 models on this page, 3 are timed.
- How are these models ordered?
- As in the full ranking: speed tier, then P95 time to first audio, with ties within 10 ms ordered by median time to first audio. Models Coval does not time follow, ordered by listener Elo; models with neither measure come last, in alphabetical order.