Fun Realtime Tts New Text To Speech Model Topping Artificial Analysis Leaderboard

Alibaba's Fun-Realtime-TTS took first place on the Artificial Analysis Speech Arena Leaderboard in a June 3, 2026 update (artificialanalysis.ai). The arena ranks text-to-speech models by listener preference: people hear pairs of samples generated from the same text in blind comparisons and choose which sounds more natural, and the votes are converted into Elo ratings, where a higher score means listeners preferred that model more often (leaderboard methodology). It is Alibaba's first model to reach #1 there; the earlier Fun-Realtime-TTS-Preview reached #7.

The lead is narrow. Fun-Realtime-TTS scores 1,219, based on 962 arena appearances, against 1,214 for Google's Gemini 3.1 Flash TTS, 1,209 for Inworld's Realtime TTS-2 Research Preview and 1,203 for Cartesia's Sonic 3.5. Artificial Analysis gives its score an uncertainty range of plus or minus 16 points and says only 24 Elo points separate the top five models, describing competition at the top of the arena as tighter than ever.

On price, Artificial Analysis places it in the middle of the field at $27.59 per million characters: above Gemini 3.1 Flash TTS at $18.3, and below Inworld Realtime TTS 1.5 Max at $35 and Sonic 3.5 at $39. The Inworld price quoted is for a different Inworld model than the one ranked third. Fun-Realtime-TTS generates speech in real time, supports voice cloning, voice design, multilingual output and regional accents and dialects, and is available to developers through an API on Alibaba Cloud.