Data read October 27, 2026
Arabic TTS APIs in 2026: 24 models tested in Arabic
24 models were tested in Arabic by Artificial Analysis or Speko. 8 of the 24 are in the multilingual speed ranking; the others are not tagged multilingual by Coval or are not measured by it. The fastest ranked model tested in Arabic is Inworld Realtime TTS-2 Flash.
No public source publishes naturalness or accuracy for Arabic on its own, and all latency is measured in English from the US East. The page therefore says who was tested in Arabic, and how fast each model is in general.
How the rows are ordered. Ranked models first, in ranking order (speed tier, then P95, ties by median); the 16 models without a rank follow alphabetically.
| Rank | Model | Speed tier From the Coval P95 | P95 (ms) 1 turn in 20 is slower | Languages tested | Listener Elo Artificial Analysis, with 95% interval | Price per 1M characters List price; a range when sources differ |
|---|---|---|---|---|---|---|
| 1 = |  Inworld Realtime TTS-2 Flash | Headroom | 100.9 | 9 | 1,218 ±13 | $10.40 to $15 |
| 6 |  Inworld Realtime TTS-2 | Fast enough | 194.7 | 12 | 1,267 ±14 | $15 to $25 |
| 8 |  ElevenLabs Eleven v4 Turbo | Fast enough | 238.6 | 9 | 1,333 ±16 | $40 |
| 9 |  ElevenLabs Flash v2.5 | Fast enough | 255.1 | 9 | 1,076 ±11 | $40 to $50 |
| 10 = |  Soniox TTS Real-Time v2 | Visible delay | 324.7 | 10 | 1,180 ±11 | $13 to $14.81 |
| 14 |  Cartesia Sonic 3.6 | Visible delay | 374.4 | 9 | 1,292 ±14 | $39.20 to $50 |
| 15 |  Cartesia Sonic 3.5 | Visible delay | 390.5 | 15 | 1,190 ±11 | $39.20 to $50 |
| 18 |  SpaceXAI Grok TTS | Visible delay | 473.9 | 11 | 1,155 ±13 | $15 |
| not ranked |  Alibaba Qwen-Audio-3.0-TTS-Plus | n/a | n/a | 8 | 1,269 ±13 | $19.30 |
| not ranked |  Boson AI Higgs Audio V3 TTS | n/a | n/a | 9 | 1,045 ±13 | $15 |
| not ranked |  Cartesia Sonic 3 | n/a | n/a | 9 | 1,054 ±11 | $49 to $50 |
| not ranked |  Coqui XTTS v2 | n/a | n/a | 7 | 913 ±15 | $40.40 |
| not ranked |  ElevenLabs Eleven v3 | n/a | n/a | 10 | 1,184 ±10 | $100 |
| not ranked |  ElevenLabs Eleven v3 Conversational | n/a | n/a | 13 | 1,213 ±12 | $50 |
| not ranked |  ElevenLabs Eleven v4 | n/a | n/a | 9 | 1,333 ±16 | $80 |
| not ranked |  ElevenLabs Multilingual v2 | n/a | n/a | 8 | 1,093 ±10 | $100 |
| not ranked |  ElevenLabs Turbo v2.5 | n/a | n/a | 9 | 1,093 ±10 | $50 |
| not ranked |  Fish Audio OpenAudio S1 | n/a | n/a | 7 | 1,084 ±11 | $15 |
| not ranked |  Fish Audio OpenAudio S1 Mini | n/a | n/a | 7 | 1,043 ±20 | $15 |
| not ranked |  Fish Audio S2 Pro | n/a | n/a | 9 | 1,108 ±12 | $15 |
| not ranked |  Google Gemini 3.8 Flash TTS | n/a | n/a | 9 | 1,280 ±14 | $15 to $16.50 |
| not ranked |  Google Gemini 3.8 Flash-Lite TTS | n/a | n/a | 9 | 1,256 ±13 | $10 to $11 |
| not ranked | | n/a | n/a | 11 | 1,174 ±10 | $100 |
| not ranked | | n/a | n/a | 9 | 1,153 ±10 | $60 |
Frequently asked questions
- Which TTS APIs support Arabic?
- 24 models were tested in Arabic. Starting with the best-ranked, Inworld Realtime TTS-2 Flash, Inworld Realtime TTS-2, ElevenLabs Eleven v4 Turbo, and ElevenLabs Flash v2.5. Being tested in a language is evidence of support, not a quality score.
- Is TTS latency measured in Arabic?
- No. Coval measures latency in English from the US East, so the latency columns show how fast each model starts in general, not in Arabic.
- Which Arabic TTS API is fastest?
- Among models tested in Arabic that are also ranked, Inworld Realtime TTS-2 Flash is first, with a P95 of 100.9 ms.
- How many of the 24 models are ranked?
- 8 of the 24 are in the multilingual speed ranking; the others are not tagged multilingual by Coval or are not measured by it.