Alibaba generative AI model developments

Alibaba has expanded its generative AI capabilities through the release of new specialized models for audio and video processing.

In late July, Alibaba released the Qwen-Audio-3.0-ASR-Flash speech-recognition model. This model achieved high accuracy in medical testing and outperformed OpenAI in certain speech-to-speech benchmarks regarding reasoning and conversational dynamics, despite having higher latency. The Qwen family supports features such as full-duplex dialogue and custom voice cloning.

Following this, Alibaba Cloud launched Wan3.0, an AI video generation model capable of producing 30-second clips. Unlike previous models, Wan3.0 supports diverse document formats—including PDF and PPT—as prompts. The model is designed for professional use in advertising and media, with a tiered API pricing structure and integration into third-party platforms like Meitu.

As of late August 2026, Wan3.0 details reveal it doubles the duration capacity of its predecessor, Wan2.7, and supports resolutions up to 1080p. The model can transform documents such as PDFs, PowerPoint presentations, spreadsheets, and web pages directly into structured video content. It can process files up to 100MB or 50 pages in length. Technically, it focuses on maintaining consistency in character appearance, spatial layout, and object movement, while featuring synchronized facial expressions and multilingual voice outputs.

To encourage adoption, Alibaba Cloud is offering a 30% discount on API usage through September 23, 2026. This launch follows a $10 billion share offering by Alibaba intended to fund expanding artificial intelligence infrastructure and chip development.