Soniox TTS RT v1
Strong Proprietary · Soniox
57.2
/100
Soniox TTS RT v1 is a proprietary text-to-speech (TTS) API from Soniox, measured by two independent benchmarks: Coval and Speko.
Spanish German French Coval
+1
ElevenLabs Eleven v3 Conversational ElevenLabs · text-to-speech (TTS) API
Strong323.1 ms P50 time-to-first-audio on Coval; ranked 10th of 112 in the TTS for Voice Agents 2026 index.
Benchmark summary: 56.8/100, ranked 10th of 112 in the TTS for Voice Agents 2026 index (Strong). Strongest criterion: voice quality, 17.2 of 20 points.
ElevenLabs Eleven v3 Conversational is a proprietary text-to-speech (TTS) API from ElevenLabs, measured by three independent benchmarks: Artificial Analysis, Coval and Speko.
Proprietary Spanish German French Portuguese Japanese Chinese Arabic Hindi Vietnamese Korean Filipino Norwegian Tamil Knowledge sharing Assistants Entertainment Customer service Artificial Analysis Coval Speko ElevenLabs Strong
Overall score
56.8 /100
Score breakdown
- Real-time speed 29.9 /60
- Voice quality 17.2 /20
- Accuracy 8.2 /15
- Price 1.5 /5
Data checked on September 30, 2026
Updated after each benchmark capture
Every point comes from a public benchmark: sources · methodology
ElevenLabs Eleven v3 Conversational ranks 10th of 112 in the 2026 text-to-speech (TTS) API benchmark, with a score of 56.8/100.
Real-time latency
On Coval's 30-day window ending October 1, 2026 (11,842 runs), ElevenLabs Eleven v3 Conversational has a P50 (median) time-to-first-audio (TTFA) of 323.1 ms, 19th lowest of the 31 models Coval measures. Its P99, the slowest 1% of turns, is 497.7 ms (16th lowest of the 31 models Coval measures). Latency consistency: a standard deviation of 52.9 ms (10th lowest of the 31 models Coval measures). Mean time to first byte (TTFB) is 223.7 ms. Speko measures a P50 TTFA of 266 ms from US East (n=30), dated July 3, 2026: 16th lowest of the 29 models Speko measures.
Accuracy (word error rate)
Coval measures an accuracy (word error rate) of 4.3% over 11,822 samples on the same window, 5th lowest of the 31 models Coval measures. Speko rates its pronunciation robustness on numbers, dates and currency at 0.3 on a 0-to-1 scale (14th highest of the 28 models Speko rates).
Voice quality
In the Artificial Analysis Speech Arena (capture of October 1, 2026), ElevenLabs Eleven v3 Conversational has an Elo of 1,200 (±15, 95% confidence) from 2,039 blind comparisons, 13th highest of the 92 models Artificial Analysis rates. Speko rates its naturalness at 1,590 Elo from blind A/B votes (7th highest of the 41 models Speko rates).
Price
Artificial Analysis lists ElevenLabs Eleven v3 Conversational at $50 per 1M characters (57th lowest of the 78 models Artificial Analysis lists a price for). Speko lists a cost of about $50 per 1M characters.
Scoring note
Artificial Analysis changed its Speech Arena Elo between the captures of September 30, 2026 and October 1, 2026 without a new measurement date (1,199, then 1,200): the score uses the average, 1,199.5. The value shown is the latest. Speko changed its pronunciation robustness between the captures of September 30, 2026 and October 1, 2026 without a new measurement date (0.24, then 0.3): the score uses the average, 0.27. The value shown is the latest.
Languages tested
Tested in Spanish, German, French, Portuguese, Japanese, Chinese, Arabic, Hindi, Vietnamese, Korean, Filipino, Norwegian and Tamil by Artificial Analysis or Speko. A language missing here was not tested, which does not mean it is unsupported.
Benchmark measurements
Latency and accuracy — Coval (30-day window)
- Mean TTFA
- 329.6 ms
- P50 / median time-to-first-audio (TTFA)
- 323.1 ms
- P90 TTFA
- 386.6 ms
- P95 TTFA
- 410.2 ms
- P99 TTFA (slowest turns)
- 497.7 ms
- Latency consistency (standard deviation)
- 52.9 ms
- Leading silence
- 105.9 ms
- Time to first byte (network arrival)
- 223.7 ms
- Accuracy: word error rate (WER)
- 4.3%
- WER basis
- per-clip mean
- Latency runs
- 11,842 runs
- Accuracy runs
- 11,822 runs
- Window
- 30d
- Coval snapshot
- 2026-10-01T10:11:47.318752Z
- Coval model ID
- eleven_v3_conversational
- Coval model page
- benchmarks.coval.ai
Speko
- P50 TTFA, US East
- 266 ms
- Naturalness (Elo)
- 1,590
- Head-to-head win rate
- 66%
- Pronunciation robustness (0 to 1)
- 0.3
- Voice drift (lower is steadier)
- 20
- Price per 1M characters
- 50 USD
- Measured on
- 2026-07-03
- Speko model
- eleven_v3_conversational
- Speko model page
- benchmarks.speko.ai
Artificial Analysis Speech Arena
- Speech Arena Elo (blind listener preference)
- 1,200
- Elo 95% interval (±)
- 15
- Arena rank
- 13
- Arena rank range
- 9-14
- Arena appearances
- 2,039
- Voices tested
- 9
- Price per 1M characters
- 50 USD
- Released
- Aug 2026
- Artificial Analysis model
- v3 Conversational
- Artificial Analysis leaderboard
- artificialanalysis.ai
Similar TTS models
ElevenLabs Flash v2.5
Strong Proprietary · ElevenLabs
56.1
/100
ElevenLabs Flash v2.5 is a proprietary text-to-speech (TTS) API from ElevenLabs, measured by three independent benchmarks: Artificial Analysis, Coval and Speko.
Spanish German French Portuguese
+8
Cartesia Sonic 3.5
Strong Proprietary · Cartesia
58.5
/100
Cartesia Sonic 3.5 is a proprietary text-to-speech (TTS) API from Cartesia, measured by three independent benchmarks: Artificial Analysis, Coval and Speko.
Spanish German French Portuguese
+18