DeepSeek Photo: VCG

Chinese technology companies are accelerating releases and updates of large-language artificial intelligence (AI) models, moves that analysts said are expected to drive the development and application of AI technology. 

Chinse e-commerce company Alibaba updated three large-language models on Tuesday, the open-source multimodal model Qwen3-Omni, the open-source image editing model Qwen3-Image-Edit and the proprietary speech recognition model Qwen3-TTS, the Global Times learned from the company.

The highlight of this release is the Qwen3-Omni multimodal model, which can handle text, images, audio and video inputs simultaneously and support streaming output of texts and voices, providing broader possibilities for AI applications. Its open-source nature allows more developers and enterprises to leverage cutting-edge technology, accelerating the proliferation and practical implementation of AI solutions, the company told the Global Times on Tuesday.

This trend highlights the rapid advancement of large-language AI models in China, as tech companies intensify their efforts to promote the practical application of AI technology, said a Chinese expert.

DeepSeek, another prominent AI company, said on Monday that it has introduced its latest model, DeepSeek-V3.1-Terminus. This update retains the model's original capabilities while addressing issues raised by user feedbacks. The improvements include enhancing language consistency by reducing Chinese-English mixing and occasional abnormal characters, as well as improving the performance of Code Agent and Search Agent.

Wang Peng, a research fellow at the Beijing Academy of Social Sciences, said that these innovations reflect China's breakthroughs in the field of large-language AI models and the industry's relentless pursuit of enhanced model performance and user experience.

Wang noted that the focus of China's large-language AI models is shifting from technological competition to value creation. The core driving forces now include deepening technology, contextualizing application scenarios and fostering an open ecosystem, he told the Global Times on Tuesday.

By open-sourcing two of the three models, Alibaba aims to replicate the success of its Qwen family, which had been downloaded more than 40 million times since 2023. The company has since open-sourced more than 300 general-purpose large-language AI models, with derivative models surpassing 140,000, making it the world's largest open-source model series, according to Alibaba.

According to the 2025 AI Index Report released by Stanford University on April 10, Alibaba ranks third globally with six signature AI models worldwide. Its self-developed Tongyi Qwen large-scale model has open-sourced over 300 versions. 

The open-source model lowers technical barriers, creating a business model of "open-source foundational models + paid value-added services," which promotes technological inclusivity and industrial collaboration, Wang said. "This has accelerated AI adoption and feeds data accumulated from various sectors back to the vendor, tightening the innovation loop."

The application scenarios for these large models in China are becoming increasingly diverse. For instance, large-language AI models are penetrating industries such as finance, healthcare and manufacturing, achieving large-scale implementation that enhances efficiency and accuracy, Wang said. 

In the consumer market, there is explosive growth in AI-driven creative applications, virtual services and intelligent interactions, significantly upgrading user experiences, he added.

Although challenges remain due to the limitations in computing power, scarcity of quality datasets and ethics-related issues, the combined effects of policy support, technological breakthroughs, and strong market demand provide ample opportunities for the development of large-language AI models in China, Wang noted.

By 2027, China will make achievements in the extensive and deeper integration of AI in six key sectors, with the penetration rate of next-generation intelligent terminals and AI agents surpassing 70 percent, and this figure will exceed 90 percent by 2030, according to a guideline released by the State Council on deeply implementing the "AI Plus" initiative in August, the Xinhua News Agency reported.