In two demonstrations, Alibaba’s Tongyi Lab researchers showed their new open source R1-Omni inferring the emotional state of a person in a video while also offering descriptions of their clothes and environment. It adds another layer of understanding to so-called computer vision, and is an enhanced version of another open source model, HumanOmni, authored by the same lead researcher, Jiaxing Zhao.

Alibaba’s effort to carve out a leading position in AI was accelerated by DeepSeek’s splashy debut in January, and the ecommerce leader is now pushing out new releases of AI tools and apps in several arenas. It benchmarked its Qwen model against DeepSeek, secured a major partnership with Apple Inc. for AI on iPhones, and now looks to be taking on OpenAI as well. It’s offering R1-Omni for users to download for free on Hugging Face.

Attempts at achieving emotional intelligence—which empowers computers to recognize and respond to human feelings—have been widespread already. Technology that identifies a person’s state of mind and wellbeing is being used to help customer service chatbots detect frustration and Tesla Inc. cars to spot drowsy drivers.

OpenAI pushed out its GPT-4.5 model earlier this year, saying it was better at identifying and responding to subtle cues from users’ written prompts. But the model comes with a hefty price tag: it is initially available only to users who pay $200 a month. Alibaba, locked in a price war for customers in China, is asking for no fee and letting everyone make use of its new model. The demonstrations only show it surfacing general emotional descriptors like “happy” or “angry,” however its purported ability to derive those from visual cues is significant.

The Hangzhou-based tech company’s chief executive officer Eddie Wu told analysts in February that artificial general intelligence is now Alibaba’s “primary objective.” Emotional intelligence is a key step on the route to that goal.