Cartesia · Proprietary · Read 2026-10-27
Cartesia Sonic 3.5: latency, voice quality, price and languages
- Rank
- 15 / 23
- Speed tier
- Visible delay
- P95
- 390.5 ms
- Price / 1M chars
- $39.20 to $50
Cartesia Sonic 3.5 is a proprietary text-to-speech model from Cartesia. Coval measures a median of 279.3 ms and a P95 of 390.5 ms to first audio, which puts it in the Visible delay speed tier, rank 15 of 23 in the multilingual ranking. Artificial Analysis or Speko tested it in 15 languages.
This summary is generated from the measures below; it is not a vendor description. Measures a source does not hold are shown as n/a and are never estimated.
Speed (Coval, 30 days, US East, English)
| Speed tier | Visible delay |
| Median time to first audio | 279.3 ms |
| P90 | 349.6 ms |
| P95 | 390.5 ms |
| Mean | 290.5 ms |
| Standard deviation | 122.5 ms |
| Calls in the window | 2,807 |
| Cekura median, streamed text | 244 ms |
| Cekura median, full text | 261 ms |
Voice quality (never added together)
| Listener Elo (Artificial Analysis) | 1,190 ±11 |
| Customer-service Elo (Artificial Analysis) | 1,203 ±24 |
| Naturalness (Speko) | 1,574 |
| Naturalness estimate (Cekura, 1 to 5) | 4.41 |
Accuracy
| Reading accuracy (Cekura) | 73.1% |
| Numbers (Cekura) | 69% |
| Codes (Cekura) | 75% |
| Context (Cekura) | 54% |
| Medical terms (Cekura) | 97% |
| Hard passages judged correct (Artificial Analysis) | 68.8% |
| Numbers and dates (Speko score) | 0.18 |
| Word error rate (Coval) | 1.6% |
Price per 1 million characters
| Range across sources | $39.20 to $50 |
| Artificial Analysis | $49 |
| Speko | $50 |
| Coval | n/a |
| Cekura | $39.20 |
Coverage
| Access | Proprietary |
| Streaming | Measured |
| Coval tags | emotion-control, multilingual, voice-cloning |
| Rated for | Knowledge sharing, Assistants, Entertainment, Customer service |
| Languages tested | Arabic, Chinese, Filipino, French, German, Hindi, Japanese, Korean, Norwegian, Portuguese, Spanish, Tamil, Telugu, Thai, Vietnamese |
Not collected
Every measure of the database is filled for this model.
Sources
- https://artificialanalysis.ai/text-to-speech (read 2026-10-27)
- https://artificialanalysis.ai/text-to-speech/leaderboard/provider-voice (read 2026-10-27)
- https://benchmarks.cekura.ai/text-to-speech (read 2026-10-27)
- https://benchmarks-api-6wxgp27p2a-ue.a.run.app/v1/results/aggregates?benchmark=TTS&window=30d&include_series=false (read 2026-10-27)
- https://benchmarks.speko.ai/tts/ (read 2026-10-27)
- Vendor site: https://www.cartesia.ai/
Keep exploring
Frequently asked questions
- How fast is Cartesia Sonic 3.5?
- Coval measures a median time to first audio of 279.3 ms and a P95 of 390.5 ms for Cartesia Sonic 3.5, over 30 days from the US East in English (Visible delay tier).
- How much does Cartesia Sonic 3.5 cost?
- The benchmarks print a list price of $39.20 to $50 per million characters for Cartesia Sonic 3.5.
- Which languages was Cartesia Sonic 3.5 tested in?
- Artificial Analysis or Speko tested Cartesia Sonic 3.5 in Arabic, Chinese, Filipino, French, German, Hindi, Japanese, Korean, Norwegian, Portuguese, Spanish, Tamil, Telugu, Thai, Vietnamese.