Alibaba unveils Qwen2.5-Omni-7B: Its new open-source AI model for cost-effective AI agents in a challenge to DeepSeek

Less than a month after unveiling QwQ-32B (Qwen2.5-32B), an AI reasoning model that competes with DeepSeek’s R1, Alibaba is back with another release—this time, aimed at pushing multimodal AI forward.

On Thursday, Alibaba Cloud launched Qwen2.5-Omni-7B, the latest addition to its Qwen series. The new model processes not just text, but also images, audio, and video—offering real-time responses in both text and natural speech. In short, it’s built to do more than just chat.

The company says Qwen2.5-Omni-7B is efficient enough to run on edge devices like smartphones, giving developers flexibility without compromising output.

“This unique combination makes it the perfect foundation for developing agile, cost-effective AI agents that deliver tangible value, especially intelligent voice applications,” Alibaba said. 

One use case the company points to is helping visually impaired users navigate their surroundings through real-time audio descriptions.

The model is open-sourced on Hugging Face and GitHub—part of a broader shift happening across China’s AI scene since DeepSeek’s open-sourcing of its R1 model. Alibaba Cloud says it has already released over 200 generative AI models to the public.

China’s AI space has been moving fast. Since DeepSeek’s splash earlier this year, companies like Alibaba and Baidu have been rolling out new models and tools at an intense pace. Just last week, Baidu dropped a new multimodal foundation model alongside its first reasoning-focused version.

Alibaba, for its part, rolled out Qwen2.5 in January, updated its AI assistant Quark earlier this month, and now adds Qwen2.5-Omni-7B to the mix. The push is backed by serious funding. Last month, the company announced a $53 billion investment plan over the next three years to build out its cloud and AI infrastructure—more than what it spent in this space over the past decade.

The momentum appears to be paying off. Alibaba recently partnered with Apple to integrate its AI into iPhones sold in China. It also expanded a deal with BMW to bring its AI tech into the automaker’s next-gen intelligent vehicles.

Kai Wang, senior equity analyst at Morningstar, told CNBC that Alibaba’s position is strong—not just because of its AI models but also because it’s building the infrastructure needed to run them at scale.

Qwen2.5-Omni-7B is another piece of that puzzle, and a clear signal that Alibaba isn’t backing down in the fight for AI leadership in China.

Alibaba’s AI Push Extends Beyond Its Own Models

Alibaba isn’t just focused on building its own AI—it’s backing other Chinese AI startups as well. In late 2023, the company joined Tencent and other investors to invest $340 million in Zhipu, a rising competitor to OpenAI.

Zhipu has positioned itself as one of China’s top generative AI startups, playing a key role in shaping the country’s AI landscape. Meanwhile, tech giants like Alibaba, Tencent, and Baidu aren’t just investing—they’re actively developing their own AI models and rolling out new products powered by this technology.