Qwen-Audio 3.0 TTS Plus is the new speech king — but is it too slow?

alibaba's qwen-audio 3.0 tts plus now ranks first on artificial analysis' text-to-speech leaderboard with an elo score of 1,236, edging out speechifyai's simba 3.2 by two points and leaving google's gemini 3.1 flash tts behind. the model is available in two variants: a flash version for real-time interaction and a plus version for high-fidelity output, supporting 16 languages including less common ones like tagalog and thai. performance-wise, it handles noisy reference audio well and allows emotional prompting via tags, but its generation speed of 16 chars/sec is significantly slower than sonic 3.5 and simba 3.2. priced at $27.60 per million characters, it sets a new quality benchmark for enterprise tts applications despite the latency trade-offs.