Controlled Voice Arena LeaderboardArtificial Analysis
Compare Text to Speech (TTS) models using the same 8 cloned voices (4 US, 4 UK).
Range | Model | Elo | Samples | Released | API Pricing 1 | |
|---|---|---|---|---|---|---|
| 1 | 1 | Qwen-Audio-3.1-TTS-Plus | 1184±1695% interval 1168 to 1200 | 1,426 | Sept 2026 | $19.3 /1M chars |
| 2 | 2-3 | Eleven v4 | 1156±1595% interval 1141 to 1171 | 1,606 | Sept 2026 | $80.0 /1M chars |
| 3 | 2-4 | Realtime TTS-2 | 1145±1595% interval 1130 to 1160 | 2,365 | Aug 2026 | $20.8 /1M chars |
| 4 | 3-5 | Sonic 3.6 | 1138±1595% interval 1123 to 1153 | 2,381 | Aug 2026 | $49.0 /1M chars |
| 5 | 4-6 | Qwen-Audio-3.0-TTS-Plus | 1126±1595% interval 1111 to 1141 | 2,145 | Sept 2026 | $19.3 /1M chars |
| 6 | 5-6 | Sonic 3.5 | 1120±1395% interval 1107 to 1133 | 4,336 | May 2026 | $49.0 /1M chars |
| 7 | 7-10 | StepAudio 2.5 TTS | 1072±1495% interval 1058 to 1086 | 3,039 | Apr 2026 | $85.0 /1M chars |
| 8 | 7-10 | Eleven v3 | 1071±1395% interval 1058 to 1084 | 3,697 | Feb 2026 | $100.0 /1M chars |
| 9 | 7-10 | Realtime TTS-2 Flash | 1070±1495% interval 1056 to 1084 | 2,480 | Aug 2026 | $10.4 /1M chars |
| 10 | 7-13 | Gemini 3.8 Flash-Lite TTS | 1059±1595% interval 1044 to 1074 | 2,438 | Sept 2026 | $11.0 /1M chars |
| 11 | 10-14 | v3 Conversational | 1054±1595% interval 1039 to 1069 | 2,105 | Aug 2026 | $50.0 /1M chars |
| 12 | 10-14 | SpaceXAI TTS | 1053±1495% interval 1039 to 1067 | 2,763 | Mar 2026 | $15.0 /1M chars |
| 13 | 10-14 | Gemini 3.8 Flash TTS | 1047±1595% interval 1032 to 1062 | 2,402 | Sept 2026 | $16.5 /1M chars |
| 14 | 12-14 | Simba 3.0 | 1039±1495% interval 1025 to 1053 | 2,630 | Feb 2026 | $6.6 /1M chars |
| 15 | 15-20 | Speech 2.8 HD | 1022±1495% interval 1008 to 1036 | 3,273 | Feb 2026 | $100.0 /1M chars |
| 16 | 15-21 | 1022±1595% interval 1007 to 1037 | 2,125 | Aug 2026 | $34.0 /1M chars | |
| 17 | 15-21 | Speech 2.8 Turbo | 1020±1395% interval 1007 to 1033 | 3,447 | Feb 2026 | $60.0 /1M chars |
| 18 | 15-22 | Fish Audio S2.1 Pro | 1016±1495% interval 1002 to 1030 | 2,542 | Jun 2026 | $15.0 /1M chars |
| 19 | 15-24 | Gradium TTS (Aug 2026) | 1013±1595% interval 998 to 1028 | 2,151 | Aug 2026 | $47.2 /1M chars |
| 20 | 15-24 | 1013±1595% interval 998 to 1028 | 2,552 | Jun 2025 | $15.0 /1M chars | |
| 21 | 17-24 | 1007±1395% interval 994 to 1020 | 3,900 | Mar 2026 | $15.0 /1M chars | |
| 22 | 18-26 | 1002±1495% interval 988 to 1016 | 2,543 | Mar 2026 | $16.0 /1M chars | |
| 23 | 23 | OpenAudio S1 | 1000AnchorFixed at 1000 as the reference point for this board's Elo scale. | 5,461 | Jun 2025 | $15.0 /1M chars |
| 24 | 19-26 | Multilingual v2 | 1000±1495% interval 986 to 1014 | 3,326 | Aug 2023 | $100.0 /1M chars |
| 25 | 22-26 | Flash v2.5 | 992±1495% interval 978 to 1006 | 3,312 | Dec 2024 | $50.0 /1M chars |
| 26 | 22-26 | Fish Audio S1 | 990±1495% interval 976 to 1004 | 3,374 | Oct 2025 | $15.0 /1M chars |
| 27 | 27-29 | Turbo v2.5 | 974±1595% interval 959 to 989 | 2,616 | Jul 2024 | $50.0 /1M chars |
| 28 | 27-29 | Kugel 3 | 972±1495% interval 958 to 986 | 3,864 | Jun 2026 | N/A |
| 29 | 27-31 | 962±1595% interval 947 to 977 | 2,265 | Jun 2026 | $15.0 /1M chars | |
| 30 | 29-33 | Soniox TTS Real-Time v2 | 948±1595% interval 933 to 963 | 2,104 | Aug 2026 | $14.2 /1M chars |
| 31 | 29-34 | Sonic 3 | 947±1595% interval 932 to 962 | 2,650 | Oct 2025 | $49.0 /1M chars |
| 32 | 30-35 | 939±1795% interval 922 to 956 | 1,028 | Jul 2026 | N/A | |
| 33 | 30-35 | Raon SpeechLM | 938±1595% interval 923 to 953 | 2,102 | Apr 2026 | N/A |
| 34 | 31-35 | 932±1595% interval 917 to 947 | 2,626 | May 2025 | $25.0 /1M chars | |
| 35 | 32-35 | Noiz TTS | 928±1595% interval 913 to 943 | 2,166 | Nov 2025 | N/A |
| 36 | 36 | Lightning v3.1 | 893±1695% interval 877 to 909 | 2,257 | Mar 2026 | $25.0 /1M chars |
| 37 | 37 | 839±1795% interval 822 to 856 | 2,426 | Nov 2023 | $40.4 /1M chars | |
| 38 | 38 | 810±1895% interval 792 to 828 | 2,125 | Apr 2024 | $8.3 /1M chars |
1 API Pricing reflects the cost to generate 1M characters on the model creator's API with the model's default settings
2 Open Weights includes models that can be self-hosted using downloadable model weights or a provider-supplied inference runtime. Open Weights Commercial includes models whose published license permits commercial use, including use subject to license conditions (e.g. an NVIDIA AI Enterprise subscription), and excludes models that require a separately negotiated commercial agreement.