Alibaba launches AI video model Wan 3.0 after Qwen beats Gemini and ChatGPT in coding

Alibaba has launched Wan3.0, a new AI video model that can turn text, images, audio, videos and even PDFs, presentations and web pages into clips of up to 30 seconds.

In Short

• Wan3.0 turns documents, slides and spreadsheets into 30-second videos easily
• Alibaba says the model keeps characters, props and motion consistent throughout
• Pricing starts at five cents per second and rises with resolution

Alibaba is now taking its AI push into video generation. After its Qwen model surprised everyone with its capabilities and even claimed to outperform Google Gemini and ChatGPT on coding benchmarks, the company has launched Wan3.0. This new generative AI video model can generate clips of up to 30 seconds from text, images, audio, videos and even documents such as PDFs and presentations.

Alibaba says Wan3.0 can generate a 30-second video in a single pass, twice the 15-second limit of its predecessor, Wan2.7. The model supports output at up to 1080p and is designed to maintain consistency across a video, including characters, objects, backgrounds and spatial layouts. It can also generate faces with synchronised micro-expressions and supports multilingual voice output.

Wan 3.0 can turn information from PDFs into video

One of the highlighted features of Wan 3.0 is its ability to turn existing information into video. Users can feed the model a PDF, PowerPoint presentation or spreadsheet and have Wan3.0 generate a video based on that material.

The price game against US models

Just like its powerful Qwen models, Alibaba has also kept the price of its video generation model lower than its US counterparts. The company has priced Wan3.0 at $0.05 per second for 480p video, $0.10 per second for 720p and $0.20 per second for 1080p. For comparison, Google's Veo 3.1 standard tier costs $0.40 per second.

Availability of Wan3.0

Wan3.0 is now available for public beta users. The model entered public beta on August 6 through Alibaba Cloud's Model Studio and Qwen Cloud, and Alibaba is now expanding access. The company is also preparing a consumer-facing version through wan.video, while wider API access is being rolled out.

The launch comes just a day after Alibaba announced a $10.2 billion share sale in Hong Kong, with the company saying the proceeds will be used entirely to fund its AI push.

Meanwhile, Wan3.0 is part of Alibaba's broader AI offering. The company has been building up its Qwen AI models as it looks to compete with leading US AI companies. In May, Alibaba unveiled its Qwen3.7-Max model, which it said had outperformed OpenAI's ChatGPT and Google's Gemini on several coding benchmarks. The model was also positioned as a stronger option for coding, reasoning and agentic tasks.

The company has also committed 380 billion yuan over three years towards AI infrastructure, while its cloud and AI business has continued to grow. Wan3.0 is the latest step in that strategy, taking Alibaba's AI push beyond chatbots and coding and into the growing video-generation market.

- Ends