AI ranking
Audio
Audio and voice models (speech synthesis, voice cloning, sound generation) compared on output quality, prompt fidelity, control, cost, speed and defects.
Visual top 10
- 01 Eleven v4 turbo 100.00
- 02 Eleven v4 99.61
- 03 Qwen Audio 3.1 TTS Plus 93.86
- 04 Sonic 3.6 92.07
- 05 Gemini 3.8 flash tts 89.49
- 06 Fish audio drama 3 88.50
- 07 Qwen Audio 3.0 TTS Plus 86.72
- 08 Realtime tts 2 86.32
- 09 Lightning v3.1 pro (oct 2026) 86.12
- 10 Gemini 3.8 flash Lite tts 84.73
Compare models. Tick up to 3 models in the table below, then run the comparison: who wins on which use, by how much, and how solid the measurement is.
| # | Model | Vendor | Score | Monthly subscription | Assessment | |
|---|---|---|---|---|---|---|
| 01 | Elevenlabs | 100.00 / 100 | Freemium | Reliable | ||
| 02 | Elevenlabs | 99.61 / 100 | Freemium | Reliable | ||
| 03 | Alibaba | 93.86 / 100 | Not available | Reliable | ||
| 04 | Sonic 3.6 | Cartesia | 92.07 / 100 | API pricing | Reliable | |
| 05 | 89.49 / 100 | €22/month | Reliable | |||
| 06 | Fish audio drama 3 | Fish audio | 88.50 / 100 | API pricing | Reliable | |
| 07 | Alibaba | 86.72 / 100 | Not available | Reliable | ||
| 08 | Inworld | 86.32 / 100 | Not available | Reliable | ||
| 09 | Lightning v3.1 pro (oct 2026) | Smallest.ai | 86.12 / 100 | Not available | Reliable | |
| 10 | 84.73 / 100 | €22/month | Reliable | |||
| 11 | Simba 3.2 | Speechifyai | 82.75 / 100 | API pricing | Reliable | |
| 12 | Fish audio drama 3 preview | Fish audio | 80.96 / 100 | API pricing | Partial | |
| 13 | Breeze tts 2 | Breezeblue | 78.58 / 100 | Not available | Reliable | |
| 14 | Assemblyai TTS 1 | Assemblyai | 78.38 / 100 | Not available | Reliable | |
| 15 | Mistral | 77.59 / 100 | Not available | Reliable | ||
| 16 | Inworld | 76.40 / 100 | Not available | Reliable | ||
| 17 | Elevenlabs | 76.00 / 100 | Not available | Reliable | ||
| 18 | StepFun | 75.80 / 100 | API pricing | Reliable | ||
| 19 | 75.41 / 100 | €22/month | Reliable | |||
| 20 | Luna tts | Vui labs | 74.02 / 100 | Not available | Reliable | |
| 21 | Sonic 3.5 | Cartesia | 71.84 / 100 | API pricing | Reliable | |
| 22 | Elevenlabs | 70.64 / 100 | Freemium | Reliable | ||
| 23 | Fish audio s2.1 pro (oct 2026) | Fish audio | 70.63 / 100 | API pricing | Reliable | |
| 24 | Ming 3.0 TTS | Ant group | 70.45 / 100 | Not available | Reliable | |
| 25 | Soniox tts real Time v2 | Soniox | 69.45 / 100 | Not available | Reliable | |
| 26 | Gradium tts (sep 2026) | Gradium | 68.07 / 100 | API pricing | Reliable | |
| 27 | Minimax | 67.87 / 100 | API pricing | Reliable | ||
| 28 | Spacexai tts | Spacexai | 64.49 / 100 | Not available | Reliable | |
| 29 | Falcon 2 | Murf ai | 64.30 / 100 | Not available | Reliable | |
| 30 | Gradium tts (aug 2026) | Gradium | 63.90 / 100 | API pricing | Reliable | |
| 31 | Minimax | 63.70 / 100 | API pricing | Reliable | ||
| 32 | Microsoft | 62.11 / 100 | API pricing | Reliable | ||
| 33 | Async flash v1.5 | Async | 61.32 / 100 | Not available | Reliable | |
| 34 | StepFun | 60.53 / 100 | API pricing | Reliable | ||
| 35 | Async pro v1.0 | Async | 58.94 / 100 | Not available | Reliable | |
| 36 | Minimax | 58.54 / 100 | API pricing | Reliable | ||
| 37 | Minimax | 57.55 / 100 | API pricing | Reliable | ||
| 38 | Simba 3.0 | Speechifyai | 56.56 / 100 | API pricing | Reliable | |
| 39 | Chatterbox hd | Resemble ai | 55.17 / 100 | API pricing | Reliable | |
| 40 | Fish audio s2 pro | Fish audio | 54.97 / 100 | API pricing | Reliable | |
| 41 | Minimax | 54.77 / 100 | API pricing | Reliable | ||
| 42 | OpenAI | 54.38 / 100 | API pricing | Reliable | ||
| 43 | StepFun | 52.99 / 100 | API pricing | Reliable | ||
| 44 | Speech 2.6 | Hithink | 52.39 / 100 | API pricing | Reliable | |
| 45 | Elevenlabs | 52.19 / 100 | Freemium | Reliable | ||
| 46 | Elevenlabs | 51.80 / 100 | Freemium | Reliable | ||
| 47 | Minimax | 51.40 / 100 | API pricing | Reliable | ||
| 48 | Mistral | 51.20 / 100 | Not available | Reliable | ||
| 49 | OpenAI | 51.00 / 100 | API pricing | Reliable | ||
| 50 | Openaudio s1 | Fish audio | 50.61 / 100 | Not available | Reliable |
Indicative score, weighted from public sources. Transparent methodology. See the AI ranking page for the full methodology and the other categories.
AI audio and voice ranking
Le Recul's Audio ranking covers voice and sound models: speech synthesis (TTS), voice cloning and sound generation. The grid takes the criteria of the generative categories — output quality, prompt fidelity, control, cost, speed, defects — and applies them to what the ear actually perceives.
Audio has a particularity: the acceptability threshold is very high. A rough image often goes unnoticed; a vocal artefact is heard immediately. The gaps between models therefore play out less on technical feats than on regularity: natural prosody, breathing, handling of numbers and proper nouns, no drop-off at the end of a sentence.
The score aggregates public signals weighted according to an « Audio » grid, with a reliability status per evaluation. Voice cloning also raises questions of consent and use that this ranking does not settle: it measures capabilities, not whether a use is legitimate.
What this ranking is for
This ranking helps you choose a voice or sound model for a specific use: long-form reading, real-time conversation, dubbing or sound design.
It also places a limit clearly: the fact that a tool can clone a voice says nothing about your right to do so.
How to read the ranking
- Real listening quality: prosody, breathing, absence of audible artefacts.
- Regularity across a long passage, not on a single sentence.
- Handling of numbers, proper nouns and languages other than English.
- Control offered: voice, pace, emotion, pronunciation.
- Cost and speed for regular use.
FAQ — Audio
Which is the best voice AI?
It depends on the use. A model that sounds very natural in long-form reading is not necessarily the most responsive in real-time conversation. The ranking presents those trade-offs rather than a single winner.
Is voice cloning legal?
Cloning someone's voice without their agreement exposes you to serious legal risk, and the fact that a tool allows it does not cover you. The ranking only assesses technical capability.
Is transcription included?
This category mainly covers audio and voice generation. Speech recognition belongs to a different evaluation and is not mixed into the generation scores.
Why do audio scores move more slowly?
Because public sources here are fewer and less frequent than for text. Updates therefore mainly reflect the arrival of new models.
How often is the ranking updated?
About every 25 hours, automatically. The update date is shown at the top of the page.