Ranking da Arena de voz controladaArtificial Analysis
Compare modelos de texto para fala (TTS) usando as mesmas 8 vozes clonadas (4 dos EUA, 4 do Reino Unido).
Range | Creator | Model | Elo | 95% CI | Samples | Lançamento | Preço da API 1 | |
|---|---|---|---|---|---|---|---|---|
| 1 | 1-2 | Realtime TTS-2 | 1144 | -15/15 | 2,224 | Aug 2026 | US$ 20,8 /1 mi de caracteres | |
| 2 | 1-2 | Sonic 3.6 | 1139 | -15/15 | 2,233 | Aug 2026 | US$ 49,0 /1 mi de caracteres | |
| 3 | 2-4 | Qwen-Audio-3.0-TTS-Plus | 1124 | -16/16 | 2,009 | Sept 2026 | US$ 27,6 /1 mi de caracteres | |
| 4 | 3-4 | Sonic 3.5 | 1119 | -13/13 | 4,138 | May 2026 | US$ 49,0 /1 mi de caracteres | |
| 5 | 5-7 | Eleven v3 | 1074 | -13/13 | 3,581 | Feb 2026 | US$ 100,0 /1 mi de caracteres | |
| 6 | 5-7 | Realtime TTS-2 Flash | 1071 | -14/14 | 2,370 | Aug 2026 | US$ 10,4 /1 mi de caracteres | |
| 7 | 5-7 | StepAudio 2.5 TTS | 1070 | -14/14 | 2,919 | Apr 2026 | US$ 85,0 /1 mi de caracteres | |
| 8 | 7-11 | Gemini 3.8 Flash-Lite TTS | 1055 | -15/15 | 2,307 | Sept 2026 | US$ 11,0 /1 mi de caracteres | |
| 9 | 8-12 | v3 Conversational | 1053 | -15/15 | 2,028 | Aug 2026 | US$ 50,0 /1 mi de caracteres | |
| 10 | 8-12 | SpaceXAI TTS | 1052 | -14/14 | 2,652 | Mar 2026 | US$ 15,0 /1 mi de caracteres | |
| 11 | 8-12 | Gemini 3.8 Flash TTS | 1051 | -15/15 | 2,294 | Sept 2026 | US$ 16,5 /1 mi de caracteres | |
| 12 | 10-13 | Simba 3.0 | 1038 | -14/14 | 2,512 | Feb 2026 | US$ 6,6 /1 mi de caracteres | |
| 13 | 12-18 | Breeze TTS 2 | 1025 | -15/15 | 2,021 | Aug 2026 | US$ 34,0 /1 mi de caracteres | |
| 14 | 13-18 | Speech 2.8 HD | 1023 | -14/14 | 3,181 | Feb 2026 | US$ 100,0 /1 mi de caracteres | |
| 15 | 13-20 | Speech 2.8 Turbo | 1022 | -14/14 | 3,308 | Feb 2026 | US$ 60,0 /1 mi de caracteres | |
| 16 | 13-21 | Fish Audio S2.1 Pro | 1016 | -14/14 | 2,468 | Jun 2026 | US$ 15,0 /1 mi de caracteres | |
| 17 | 13-21 | Gradium TTS (Aug 2026) | 1016 | -15/15 | 2,037 | Aug 2026 | US$ 47,2 /1 mi de caracteres | |
| 18 | 13-22 | OpenAudio S1 Mini | 1014 | -15/15 | 2,480 | Jun 2025 | US$ 15,0 /1 mi de caracteres | |
| 19 | 15-22 | Fish Audio S2 Pro | 1009 | -13/13 | 3,724 | Mar 2026 | US$ 15,0 /1 mi de caracteres | |
| 20 | 15-23 | Voxtral TTS | 1008 | -14/14 | 2,432 | Mar 2026 | US$ 16,0 /1 mi de caracteres | |
| 21 | 16-23 | Multilingual v2 | 1005 | -14/14 | 3,182 | Aug 2023 | US$ 100,0 /1 mi de caracteres | |
| 22 | 22 | OpenAudio S1 | 1000 | 0/0 | 5,169 | Jun 2025 | US$ 15,0 /1 mi de caracteres | |
| 23 | 19-24 | Flash v2.5 | 995 | -14/14 | 3,193 | Dec 2024 | US$ 50,0 /1 mi de caracteres | |
| 24 | 22-25 | Fish Audio S1 | 990 | -14/14 | 3,266 | Oct 2025 | US$ 15,0 /1 mi de caracteres | |
| 25 | 23-26 | Turbo v2.5 | 980 | -15/15 | 2,544 | Jul 2024 | US$ 50,0 /1 mi de caracteres | |
| 26 | 25-27 | Kugel 3 | 974 | -14/14 | 3,789 | Jun 2026 | N/D | |
| 27 | 26-28 | Higgs Audio V3 TTS | 965 | -15/15 | 2,174 | Jun 2026 | US$ 15,0 /1 mi de caracteres | |
| 28 | 27-31 | Sonic 3 | 951 | -15/15 | 2,582 | Oct 2025 | US$ 49,0 /1 mi de caracteres | |
| 29 | 28-33 | Magpie TTS ZeroshotPesos abertos | 942 | -18/18 | 1,015 | Jul 2026 | N/D | |
| 30 | 28-33 | Raon SpeechLM | 941 | -15/15 | 2,005 | Apr 2026 | N/D | |
| 31 | 28-33 | Soniox TTS Real-Time v2 | 940 | -16/16 | 1,955 | Aug 2026 | US$ 14,2 /1 mi de caracteres | |
| 32 | 29-33 | Chatterbox | 936 | -15/15 | 2,561 | May 2025 | US$ 25,0 /1 mi de caracteres | |
| 33 | 29-33 | Noiz TTS | 935 | -15/15 | 2,080 | Nov 2025 | N/D | |
| 34 | 34 | Lightning v3.1 | 898 | -16/16 | 2,183 | Mar 2026 | US$ 25,0 /1 mi de caracteres | |
| 35 | 35 | XTTS v2 | 846 | -17/17 | 2,347 | Nov 2023 | US$ 40,4 /1 mi de caracteres | |
| 36 | 36 | OpenVoice v2 | 806 | -18/18 | 2,059 | Apr 2024 | US$ 8,3 /1 mi de caracteres |
1 O preço da API reflete o custo de gerar 1M de caracteres na API do criador do modelo com as configurações padrão do modelo
2 Pesos abertos inclui modelos que podem ser auto-hospedados usando pesos de modelo para download ou um ambiente de inferência fornecido pelo provedor. Pesos abertos para uso comercial inclui modelos cuja licença publicada permite o uso comercial, inclusive sujeito a condições da licença (por exemplo, uma assinatura do NVIDIA AI Enterprise), e exclui modelos que exigem um acordo comercial negociado separadamente.