Data read October 27, 2026
Fast enough TTS APIs in 2026: 9 models with a P95 from 120 to 293 ms
9 models fall in the Fast enough tier: a P95 from 120 to 293 ms on Coval's 30-day measure from the US East. The first is Alibaba Qwen3 TTS 1.7b.
Tiers are drawn at the widest gaps between consecutive P95 values, so they do not depend on any one model. A model Coval does not measure has no tier.
How the rows are ordered. Ranked models first, in ranking order (speed tier, then P95, ties by median); the 3 models without a rank follow alphabetically.
| Rank | Model | Median (ms) Typical wait before the first sound | 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 |
|---|---|---|---|---|---|---|
| 4 |  Alibaba Qwen3 TTS 1.7b | 103.6 | 136.6 | 0 | n/a | n/a |
| 5 |  SpeechifyAI Simba 3.0 | 100.5 | 149.4 | 4 | 1,116 ±11 | $6.60 to $10 |
| 6 |  Inworld Realtime TTS-2 | 149.1 | 194.7 | 12 | 1,267 ±14 | $15 to $25 |
| 7 |  Palabra TTS v1 | 105.1 | 214.3 | 0 | n/a | $12 |
| 8 |  ElevenLabs Eleven v4 Turbo | 188.9 | 238.6 | 9 | 1,333 ±16 | $40 |
| 9 |  ElevenLabs Flash v2.5 | 191.0 | 255.1 | 9 | 1,076 ±11 | $40 to $50 |
| not ranked |  Deepgram Flux TTS | 197.7 | 267.0 | 0 | 1,052 ±13 | $45 |
| not ranked |  Fish Audio Drama 3 Preview | 146.0 | 209.3 | 2 | 1,245 ±14 | $15 |
| not ranked |  SpeechifyAI Simba 3.2 | 104.3 | 161.1 | 0 | 1,249 ±12 | $6.60 to $10 |
Frequently asked questions
- What does the Fast enough speed tier mean?
- A TTS model is in the Fast enough tier when it has a P95 from 120 to 293 ms. P95 is the slow turn: one call in twenty is slower.
- Which TTS APIs are in the Fast enough tier?
- Starting with the best-ranked, Alibaba Qwen3 TTS 1.7b, SpeechifyAI Simba 3.0, Inworld Realtime TTS-2, Palabra TTS v1, and ElevenLabs Eleven v4 Turbo. There are 9 in total.