Baidu и Alibaba обостряют конкуренцию в мире ИИ с новыми модельными инновациями Translation: Baidu and Alibaba intensify competition in the AI world with new model innovations.

Baidu has unveiled the latest iteration of its flagship AI model, Ernie 5.0, during the largest annual tech expo in Beijing, according to Bloomberg.

Ernie 5.0 is described as a «native omnimodal» neural network, capable of comprehending user commands across various media formats.

«Intelligence itself is the ultimate application, and the speed of technological iteration serves as the only true defensive barrier. Baidu will continue to invest in developing more advanced models to raise the bar even higher,» stated billionaire and company founder Robin Li.

In the live demonstrations of language, audio, and visual comprehension, Ernie 5.0 performed comparably to top competitors in the sector such as DeepSeek, Gemini, and GPT-5.

During his presentation, Li addressed concerns regarding the burgeoning bubble in the artificial intelligence market. He highlighted the «unhealthy, unstable» structure of the industry, where chip manufacturers garner the majority of the gains from technology, rather than application developers.

«Regardless of how much profit chip makers earn, models built upon them need to deliver ten times more value, and applications should yield a hundred times more,» he emphasized.

The introduction of Ernie 5.0 capped a three-hour showcase of Baidu’s new products. In addition to the model, the company also introduced two AI chips—M100 and M300—which are set for release in 2026 and 2027, with specific technical details yet to be disclosed.

Baidu has also launched an agency tool intended for industrial applications, such as traffic light management.

Additionally, the company is promoting some of its existing solutions in international markets, including a digital avatar for streamers and a no-code development tool.

Previously, Baidu held a competitive edge in China’s AI race and was regarded as a significant rival to OpenAI within the country. However, it has since lost its lead to ByteDance and Alibaba.

Alibaba is preparing to refresh its AI mobile app, Tongyi, in the coming months, aiming to make it more like ChatGPT.

The initial plan includes a rebranding to Qwen, named after its flagship AI model. Subsequent updates will introduce AI-driven features for shopping across various platforms, including the primary market of Taobao.

The ultimate goal is to transform the app into a fully functional AI agent and to scale globally with an international version launch.

In addition to Tongyi, Alibaba has Qwen Chat available on iOS and Android, though its functionality is notably limited. The company intends to streamline the interface and enhance user experience while developing consumer services under the Qwen brand.

Currently, the app will be free, but there are plans to introduce paid access to its services in the future.

Chinese startups significantly lag behind their American counterparts in terms of revenue. As of August, only four AI applications from private firms in China made it into the top 100 for annual revenue. Glority, Plaud, ByteDance, and Zuoyebang collectively generated $447 million, accounting for a mere 1.23% of a total of $36.4 billion.

However, China leads the world in the number of patents related to large language models (LLM). According to the World Intellectual Property Organization, China filed 1.8 million applications in 2024—almost half of the global total and more than three times that of the U.S.

Japan, South Korea, and Germany follow in subsequent rankings.

China has issued over 1 million patents, whereas the United States has granted 319,815.

It is worth noting that in September, China’s Cybersecurity Administration prohibited major tech firms in the country from purchasing Nvidia AI chips. Regulators in China recognized the high efficacy of domestic microchips.