Alibaba’s Qwen-Audio TTS model takes the top spot on Speech Arena leaderboard, and crypto should pay attention
The Chinese tech giant's text-to-speech model narrowly edges out competitors from Google and SpeechifyAI, signaling a shift in who controls AI infrastructure that crypto projects increasingly depend on.
Alibaba just claimed the number one position on the Artificial Analysis Speech Arena Leaderboard with its Qwen-Audio-3.0-TTS-Plus model. The model earned an Elo rating of 1,236 based on 1,305 blind user evaluations, edging out SpeechifyAI’s Simba 3.2 by a razor-thin margin of two points.
The leaderboard breakdown
The rankings are determined by crowdsourced blind preference votes. Users listen to speech samples generated from identical text prompts and pick whichever sounds more natural, without knowing which model produced which output.
The current top five tells an interesting story about who’s winning the AI voice race. Alibaba sits at the top with 1,236 Elo. SpeechifyAI’s Simba 3.2 follows at 1,234. Google’s Gemini 3.1 Flash TTS comes in third at 1,214. Cartesia’s Sonic 3.5 takes fourth with 1,207. And Inworld’s Realtime TTS-2 Research Preview rounds out the top five at 1,205.
The model launched in July 2026 and offers eight distinct voices through Alibaba Cloud’s API. Pricing sits at $27.6 per million characters. For context, a million characters is roughly 200,000 words, or about three full-length novels worth of spoken audio.
This wasn’t a one-hit wonder either. Alibaba’s Tongyi Lab had already placed its Fun-Realtime-TTS model in the global top five back in May 2026. The trajectory here is clear: Alibaba has been steadily climbing and now sits at the summit.