At Apsara 2026 in Hangzhou, Alibaba gave each part of the Qwen family a clearer job. Some models translate or transcribe speech; another can turn a script into a full audio scene. But a name on stage is not necessarily a model developers can use today. Alibaba has yet to spell out access and pricing for much of the lineup.

What the multimodal models do

  • Qwen3.8-LiveTranslate: Translates speech as a conversation happens. Alibaba says its latency measure, LAAL, fell from 2.8 to 2.3 seconds, nearly 20%, while translation quality improved. The company has not said which language pairs it supports or how developers can access it.
  • Qwen-Audio-3.1-TTS-Next: Turns a text script into speech and background sound together. Alibaba points to audiobooks, film and television, podcasts and games as possible uses. It has not said how developers can access or license the model.
  • Qwen-Audio-3.1-ASR, TTS and Realtime: Three speech tools with different jobs. ASR turns speech into text, TTS turns text into speech, and Realtime handles live voice interaction. Alibaba gave no separate performance figures or release dates for these updates.
  • Qwen-Image 3.1: A model for generating and editing images, including designs with transparent backgrounds. Alibaba is aiming it at creative and e-commerce work, but says it will arrive later in 2026.

Qwen 4 isn't out yet

Alibaba said Qwen 4 is currently in training. It gave no release date, model card, public weights, price or parameter count for it. The company said later Qwen 4.5 and Qwen 5 models could reach 5–10 trillion parameters. Those numbers are plans for future models, not specifications for Qwen 4.

Alibaba also described work on its existing Qwen3.8-Max model. It says 33 automated improvement cycles over a month raised the model's Artificial Analysis score from 40 to 45. Those are Alibaba's own results, and the company has not said whether developers can access the updated version yet.

For phone makers, Alibaba launched Qwen Intelligence. The company says it will help AI assistants handle tasks across apps. It is a product built around Qwen, rather than a separate model.

Alibaba's new Zhenwu V900 chip is covered in our earlier Apsara report.