Alibaba Launches AI Model That Can Read Human Emotions

Chinese e-commerce giant Alibaba has continued to make headlines by launching a new AI model that it claims is capable of reading human emotions. The model, called "R1-Omni", was developed by a team of engineers at Alibaba's Tongyi Lab. With the ability to read emotions at the core of user interactions, Alibaba confidently claims that R1-Omni can completely surpass competitors such as OpenAI's ChatGPT and many other LLM models currently on the market in terms of real-world experience.

Notably, R1-Omni is open source. In demos, the AI ​​model was able to understand human emotions from videos, as well as describe their clothing and surroundings. Alibaba's new AI model takes computer vision to the next level, building on a previously released model called HumanOmni, also created by researcher Jiaxing Zhao and his colleagues at Tongyi Lab.

Technically, R1-Omni can be considered the first comprehensive multimodal language model that applies reinforcement learning with verifiable rewards (RLVR) – a technology that promises to improve AI’s ability to reason and analyze emotions. As a result, R1-Omni can quickly analyze and understand more accurately how visual and auditory information supports emotion recognition, and clearly identify which factors play a more important role in assessing each type of emotion.

Alibaba is investing heavily in AI tools and applications. The Chinese tech giant recently signed a partnership with Apple to bring AI features to iPhones in China, and challenged OpenAI by offering R1-Omni for free on its Hugging Face platform.

Alibaba’s ultimate goal is “artificial general intelligence” (AGI). CEO Eddie Wu made it clear to analysts in February that AGI is a top priority for the company. And the latest developments in AI are showing Alibaba’s efforts to position itself as a major player in the global AI race.

On the other side, OpenAI is also aware of the growing competition from rival AI platforms. The company released an updated GPT-4.5 model earlier this year that can detect subtle nuances in user prompts. However, the model is not available for free, but only for premium subscribers, which cost a hefty $200/month.