Alibaba Cloud continues the industrialization of its artificial intelligence strategy by revealing Qwen2.5-OMNI-7Ba new generation model designed for on -board uses. This multimodal model, capable of treating texts, images, audio and video, is optimized for execution on EDGE terminals such as mobile phones, without performance degradation.
By focusing on local execution rather than on the only power of cloud infrastructure, Alibaba wants to send an increasing need for autonomy and responsiveness in AI applications. The model can generate textual and vocal responses in real time, which makes it particularly suitable for use cases with high temporal or contextual constraint.
According to the company, Qwen2.5-OMNI-7B is a relevant technical basis for the development of low-cost intelligent agents. Among the examples of integration mentioned: real -time vocal assistance for visually impaired, or integration into future BMW vehicles, as part of an expanded partnership formalized this week.
This technical approach is part of an opening logic: the model is available in open source on Hugging Face and Github, like the dynamics launched in China by the Deepseek R1 model. Alibaba claims to date more than 200 generative models open for three years, thus consolidating an ecosystem strategy favorable to rapid adoption.
This change of course comes as the competition intensifies in the generative AI in China. Baidu recently presented its own multimodal model and an LLM oriented reasoning, while Alibaba multiplies iterations: new version of its Quark assistant, update of the Qwen series, and integration projects in the Apple universe. The company also announced an investment of $ 53 billion over three years in its Cloud and AI capacities, an unprecedented effort on the group level.
