Skip to content
IFAI

Model rankings

1492 models · 5 boards · 2 sources

‹
Week of September 7, 2026
October 19, 2026247 changesOctober 12, 2026127 changesOctober 5, 202634 changesSeptember 28, 202617 changesSeptember 21, 2026115 changesSeptember 14, 2026174 changesSeptember 7, 2026144 changesAugust 31, 202620 changesAugust 24, 202615 changesAugust 17, 202633 changesAugust 10, 2026116 changesAugust 3, 202613 changes
›

Text-to-speech arena · Anonymous head-to-head votes · Updated September 8, 2026

Best voice
Sonic 3.6Cartesia1,282Score
Best open model
Breeze TTS 2BreezeBlue1,215Score
Best from China
Qwen-Audio-3.0-TTS-PlusAlibaba1,241Score
Top three by Score
1Sonic 3.6Cartesia1,282
2Realtime TTS-2Inworld1,252
3Qwen-Audio-3.0-TTS-PlusAlibaba1,241
Top three open models
1Breeze TTS 2BreezeBlue · #7 overall1,215
2Fish Audio S2 ProFish Audio · #25 overall1,128
3Step Audio EditX (Mar 2026)StepFun · #32 overall1,104
This week's changes
Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback)Intelligence · #1New
GPT-6 Astra (max)Intelligence · #2New
MAI-Image-2.6Image · #2New
Realtime TTS-2Speech · #2New
Muse Spark 1.3 (max)Intelligence · #5New
AllOpenClosed
All regionsUSChina
One row per modelAll variants
RankChangeModelScore95% interval
1—Sonic 3.6ClosedCartesia1,282±17
2NewRealtime TTS-2ClosedInworld1,252±18
3▼ 1Qwen-Audio-3.0-TTS-PlusClosedAlibaba1,241±14
4▼ 1Simba 3.2ClosedSpeechifyAI1,240±14
5▼ 1Luna TTSClosedVUI Labs1,228±14
6NewRealtime TTS-2 Flash - Research PreviewClosedInworld1,222±15
7NewBreeze TTS 2OpenBreezeBlue1,215±17
8Newv3 ConversationalClosedElevenLabs1,210±15
9▼ 4Gemini 3.1 Flash TTSClosedGoogle1,208±12
10NewStepAudio 2.5 TTS (Aug 2026)ClosedStepFun1,205±17
11▼ 4Sonic 3.5ClosedCartesia1,193±12
12▼ 3Lightning V3.1 Pro (Jul 2026)ClosedSmallest.ai1,190±14
13NewSoniox TTS Real-Time v2ClosedSoniox1,179±11
14▼ 3Eleven v3ClosedElevenLabs1,175±11
15▼ 3Speech 2.8 HDClosedMiniMax1,171±11
16▼ 3Falcon 2ClosedMurf AI1,158±15
17▼ 3Speech 2.8 TurboClosedMiniMax1,152±11
18NewGradium TTS (Aug 2026)ClosedGradium1,149±15
19▼ 3Async Flash v1.5ClosedAsync1,141±13
20▼ 3Fish Audio S2.1 ProClosedFish Audio1,141±13
21▼ 3Step TTS 2 (Mar 2026)ClosedStepFun1,138±12
22▼ 3Speech 2.6 HDClosedMiniMax1,134±11
23▼ 3Lightning V3.1 Pro TTS (Jun 2026)ClosedSmallest.ai1,133±13
24▼ 2Azure HD 2.5ClosedMicrosoft1,131±12
25—Fish Audio S2 ProOpenFish Audio1,128±12
26▼ 2Speech 2.6 TurboClosedMiniMax1,127±11
27—SpaceXAI TTSClosedSpaceXAI1,123±16
28▼ 2Simba 3.0ClosedSpeechifyAI1,122±12
29▼ 6Async Pro v1.0ClosedAsync1,121±13
30▼ 2Speech-02-HDClosedMiniMax1,112±11
31▼ 2TTS-1 HDClosedOpenAI1,106±11
32▼ 2Step Audio EditX (Mar 2026)OpenStepFun1,104±13
33▼ 2Turbo v2.5ClosedElevenLabs1,100±10
34▼ 2Multilingual v2ClosedElevenLabs1,099±10
35▼ 2Chatterbox HDClosedResemble AI1,093±12
36▼ 2TTS-1ClosedOpenAI1,091±10
37▼ 1Gemini 2.5 Flash Lite TTSClosedGoogle1,084±12
38▲ 3Speech 2.6ClosedHithink1,084±13
39▼ 2Speech-02-TurboClosedMiniMax1,082±11
40▼ 1StudioClosedGoogle1,081±11
41▲ 1OpenAudio S1ClosedFish Audio1,080±11
42▼ 4Flash v2.5ClosedElevenLabs1,080±11
43▼ 3Voxtral TTSOpenMistral1,078±13
44▲ 3JourneyClosedGoogle1,075±13
45▲ 1Maya 2 FlashClosedMaya Research1,072±14
46▼ 2Sonic 3ClosedCartesia1,067±11
47▼ 4GPT-Realtime-2ClosedOpenAI1,067±13
48—T2A-01-HDClosedMiniMax1,066±11
49—Magpie-Multilingual 357M (Feb 2026)OpenNVIDIA1,065±13
50▲ 1Polly GenerativeClosedAmazon1,063±11

Intelligence over time

Each maker's best model at the time on the Artificial Analysis Intelligence Index, by the month it was released.

Gained most · Anthropic +57.2Highest now · Anthropic 64.6Gap between highest and lowest 18.1 ptsTime range Nov 2025 – Nov 2026
64.6Anthropic60.0Google56.9OpenAI50.1Meta46.5Kimi

Arena data comes from the leaderboard dataset it publishes on Hugging Face (CC BY 4.0). Only each board's overall leaderboard is used; scores and ranks are unchanged. dataset · CC BY 4.0

Artificial Analysis data comes from its API (artificialanalysis.ai). Scores are unchanged; ranks follow the scores. artificialanalysis.ai

Open or closed follows the license Arena gives; prices and Artificial Analysis intelligence show only where model names match, otherwise “—”.