Alibaba Cuts Prices on Qwen Audio 3.1 Voice Models

Alibaba released five new audio models in the Qwen-Audio-3.1 line. The ASR and ASR-Next models recognize speech in several languages and dialects and tell speakers apart by voice and by emotion, while TTS and TTS-Next read text out loud and carry a voice across languages, with intonation controlled through a normal text prompt. Along with the release Alibaba cut prices - TTS got cheaper by about 70%, the real-time model by about 85%, and ASR by up to 95%.

Here the value is not in new quality but in price - there is not one benchmark in the source, just a list of features and new API rates. I follow releases like this because the voice channel in agents stays one of the most expensive parts of the chain, and a price cut of 70-95% across different models changes that bill quite a bit. But the numbers are only from Alibaba itself, no independent check, and I would first run my own cases on it and only then move it into production.