Prototype · closed beta
unOpen.AI
In closed beta: Roxy, a private AI assistant with permanent memory running on infrastructure the user owns. Brand, site, product, backend and conversational design.
Model families
Voice models.
Ranked by listeners, weekly.
Every text-to-speech model in the Artificial Analysis arena, where people listen to two anonymous voices reading the same text and pick one. Refreshed every week; each model has its own page.
All Model Families →
LEADING MODELS
The highest-rated speech models.
Ranked by blind preference votes in the Artificial Analysis text to speech arena. Each model has its own page with every arena it is ranked in.
TOP TEN
The top ten, side by side.
Elo rating from blind preference votes · Higher is preferred more often · Linear scale from 0 to 1,500
- Eleven v4 Turbo
- 1,336
- Eleven v4
- 1,334
- Qwen-Audio-3.1-TTS-Plus
- 1,305
- Sonic 3.6
- 1,296
- Gemini 3.8 Flash TTS
- 1,283
- Fish Audio Drama 3
- 1,278
- Qwen-Audio-3.0-TTS-Plus
- 1,269
- Realtime TTS-2
- 1,267
- Lightning V3.1 Pro (Oct 2026)
- 1,266
- Gemini 3.8 Flash-Lite TTS
- 1,259
Elo is relative: a 100-point gap means voters preferred the higher model about 64% of the time. Gaps inside the confidence intervals in the table below are not meaningful, and preference votes say nothing about price, licence or speed.
Source: Artificial Analysis arenas. Checked 30 October 2026. Measured independently by Artificial Analysis, not by QQuantum.ai or the providers.
EVERY RANKED MODEL
All 100 speech models, ranked.
Sort by any column. Select a model for its full page.
| 1 | Eleven v4 Turbo | ElevenLabs | 1336 -16/16 | — |
|---|---|---|---|---|
| 2 | Eleven v4 | ElevenLabs | 1334 -15/15 | — |
| 3 | Qwen-Audio-3.1-TTS-Plus | Alibaba | 1305 -15/15 | — |
| 4 | Sonic 3.6 | Cartesia | 1296 -13/13 | — |
| 5 | Gemini 3.8 Flash TTS | 1283 -13/13 | — | |
| 6 | Fish Audio Drama 3 | Fish Audio | 1278 -18/18 | — |
| 7 | Qwen-Audio-3.0-TTS-Plus | Alibaba | 1269 -13/13 | — |
| 8 | Realtime TTS-2 | Inworld | 1267 -14/14 | — |
| 9 | Lightning V3.1 Pro (Oct 2026) | Smallest.ai | 1266 -16/16 | — |
| 10 | Gemini 3.8 Flash-Lite TTS | 1259 -13/13 | — | |
| 11 | Simba 3.2 | SpeechifyAI | 1249 -12/12 | — |
| 12 | Fish Audio Drama 3 Preview | Fish Audio | 1240 -14/14 | — |
| 13 | Breeze TTS 2 | BreezeBlue | 1228 -13/13 | — |
| 14 | AssemblyAI-TTS-1 | AssemblyAI | 1227 -16/16 | — |
| 15 | Voxtral 3 TTS | Mistral | 1223 -14/14 | — |
| 16 | Realtime TTS-2 Flash | Inworld | 1217 -12/12 | — |
| 17 | v3 Conversational | ElevenLabs | 1215 -12/12 | — |
| 18 | StepAudio 2.5 TTS (Aug 2026) | StepFun | 1214 -13/13 | — |
| 19 | Gemini 3.1 Flash TTS | 1212 -11/11 | — | |
| 20 | Luna TTS | VUI Labs | 1205 -11/11 | — |
| 21 | Sonic 3.5 | Cartesia | 1194 -11/11 | — |
| 22 | Fish Audio S2.1 Pro (Oct 2026) | Fish Audio | 1188 -13/13 | — |
| 23 | Eleven v3 | ElevenLabs | 1188 -10/10 | — |
| 24 | Ming-3.0-tts | Ant Group | 1187 -13/13 | — |
| 25 | Soniox TTS Real-Time v2 | Soniox | 1182 -10/10 | — |
| 26 | Gradium TTS (Sep 2026) | Gradium | 1175 -13/13 | — |
| 27 | Speech 2.8 HD | MiniMax | 1174 -10/10 | — |
| 28 | SpaceXAI TTS | SpaceXAI | 1157 -13/13 | — |
| 29 | Falcon 2 | Murf AI | 1156 -12/12 | — |
| 30 | Gradium TTS (Aug 2026) | Gradium | 1154 -12/12 | — |
| 31 | Speech 2.8 Turbo | MiniMax | 1153 -10/10 | — |
| 32 | Azure HD 2.5 | Microsoft | 1145 -11/11 | — |
| 33 | Async Flash v1.5 | async | 1141 -11/11 | — |
| 34 | Step TTS 2 (Mar 2026) | StepFun | 1137 -11/11 | — |
| 35 | Async Pro v1.0 | async | 1129 -12/12 | — |
| 36 | Speech 2.6 HD | MiniMax | 1127 -11/11 | — |
| 37 | Speech 2.6 Turbo | MiniMax | 1122 -10/10 | — |
| 38 | Simba 3.0 | SpeechifyAI | 1117 -11/11 | — |
| 39 | Chatterbox HD | Resemble AI | 1110 -12/12 | — |
| 40 | Fish Audio S2 Pro | Fish Audio | 1109 -11/11 | — |
| 41 | Speech-02-HD | MiniMax | 1108 -11/11 | — |
| 42 | TTS-1 HD | OpenAI | 1106 -11/11 | — |
| 43 | Step Audio EditX (Mar 2026) | StepFun | 1099 -12/12 | — |
| 44 | Speech 2.6 | Hithink | 1096 -12/12 | — |
| 45 | Turbo v2.5 | ElevenLabs | 1095 -10/10 | — |
| 46 | Multilingual v2 | ElevenLabs | 1093 -10/10 | — |
| 47 | Speech-02-Turbo | MiniMax | 1091 -11/11 | — |
| 48 | Voxtral TTS | Mistral | 1090 -12/12 | — |
| 49 | TTS-1 | OpenAI | 1089 -10/10 | — |
| 50 | OpenAudio S1 | Fish Audio | 1087 -11/11 | — |
| 51 | Gemini 2.5 Flash Lite TTS | 1085 -11/11 | — | |
| 52 | Flash v2.5 | ElevenLabs | 1077 -10/10 | — |
| 53 | Journey | 1074 -12/12 | — | |
| 54 | GPT-Realtime-2 | OpenAI | 1073 -12/12 | — |
| 55 | Studio | 1073 -11/11 | — | |
| 56 | T2A-01-HD | MiniMax | 1072 -11/11 | — |
| 57 | Gemini 2.5 Flash TTS (Dec 2025) | 1068 -12/12 | — | |
| 58 | Magpie-Multilingual 357M (Feb 2026) | NVIDIA | 1067 -12/12 | — |
| 59 | Polly Generative | Amazon | 1063 -11/11 | — |
| 60 | Maya 2 Flash | Maya Research | 1062 -13/13 | — |
| 61 | Chirp 3: HD | 1062 -11/11 | — | |
| 62 | Kokoro 82M v1.0 | Kokoro | 1062 -11/11 | — |
| 63 | Octave 2 | Hume AI | 1060 -12/12 | — |
| 64 | Sonic 3 | Cartesia | 1056 -11/11 | — |
| 65 | MiMo-V2.5-TTS | Xiaomi | 1055 -12/12 | — |
| 66 | Coda | Rime | 1054 -12/12 | — |
| 67 | Flux TTS | Deepgram | 1052 -13/13 | — |
| 68 | Async Flash v1.0 | async | 1048 -11/11 | — |
| 69 | Maya 2 Global | Maya Research | 1048 -13/13 | — |
| 70 | Maya1 | Maya Research | 1046 -12/12 | — |
| 71 | OpenAudio S1 Mini | Fish Audio | 1044 -20/20 | — |
| 72 | Gemini 2.5 Pro (Dec 2025) | 1042 -12/12 | — | |
| 73 | Octave TTS | Hume AI | 1040 -11/11 | — |
| 74 | Higgs Audio V3 TTS | Boson AI | 1039 -13/13 | — |
| 75 | Bland Speech v3 | Bland AI | 1035 -12/12 | — |
| 76 | Azure Neural | Microsoft | 1033 -20/20 | — |
| 77 | MAI-Voice-1 | Microsoft | 1030 -12/12 | — |
| 78 | T2A-01-Turbo | MiniMax | 1028 -11/11 | — |
| 79 | Polly Long-Form | Amazon | 1027 -13/13 | — |
| 80 | Chatterbox | Resemble AI | 1024 -11/11 | — |
| 81 | Lightning v3.1 | Smallest.ai | 1022 -12/12 | — |
| 82 | Raon SpeechLM | Krafton | 1007 -13/13 | — |
| 83 | Zonos-v0.1 | Zyphra | 1000 0/0 | — |
| 84 | Murf Speech Gen 2 | Murf AI | 986 -12/12 | — |
| 85 | OpenVoice v2 | OpenVoice | 956 -13/13 | — |
| 86 | VibeVoice 1.5B | Microsoft | 955 -14/14 | — |
| 87 | Neuphonic TTS | Neuphonic | 941 -14/14 | — |
| 88 | Qwen3 TTS Flash | Alibaba | 941 -14/14 | — |
| 89 | Qwen3 TTS | Alibaba | 929 -13/13 | — |
| 90 | Kugel 3 | KugelAudio | 925 -13/13 | — |
| 91 | WaveNet | 919 -12/12 | — | |
| 92 | XTTS v2 | Coqui | 915 -15/15 | — |
| 93 | Mist V2 | Rime | 903 -15/15 | — |
| 94 | Polly Neural | Amazon | 895 -14/14 | — |
| 95 | Neural2 | 894 -12/12 | — | |
| 96 | StyleTTS 2 | StyleTTS | 891 -15/15 | — |
| 97 | Standard | 888 -12/12 | — | |
| 98 | Noiz TTS | Noiz | 869 -15/15 | — |
| 99 | MetaVoice v1 | MetaVoice | 847 -18/18 | — |
| 100 | Polly Standard | Amazon | 832 -15/15 | — |
Elo from the Artificial Analysis arenas: people choose between two anonymous outputs of the same prompt, and the ratings follow from those votes. The figure after each score is its 95% confidence interval; gaps inside it are not meaningful. Rank and votes are for text to speech. Refreshed weekly; not measured by QQuantum.ai. Select a heading to sort.
How we evaluate
A listening preference does not show latency, the languages and accents you need, how a voice handles your product names, or whether it can be streamed fast enough for a live conversation. For voice agents we test the shortlist inside the real call flow, where time to first audio matters as much as naturalness. Our voice AI page covers the rest of that stack.
Voice cloning raises consent and likeness questions: use only voices you have the rights to, and tell callers they are speaking to an automated system where the law requires it. Check each provider's terms for retention of the text and audio you send.
Questions
Questions about Voice Models (Text to Speech).
What does the Elo rating measure?
How often listeners preferred a model's speech over another model's for the same text, in blind side-by-side tests run by Artificial Analysis. It reflects naturalness to a general audience, not latency or language coverage.
Which voice model is best for a phone agent?
Usually not simply the top-ranked one: streaming latency, language support and cost per minute decide most phone deployments. Shortlist from the table and test in the live call flow.
Can we clone a voice?
Several providers offer it. Use only voices you have explicit rights to, and check the provider's consent and likeness rules before building on it.
