Foreign Media Focus: Chinese Companies Compete to Release New AI Models, with Costs Far Lower than Their U.S. Counterparts

【Wen/Observer Net Chen Sijia】At the beginning of last year, China's artificial intelligence (AI) model DeepSeek-R1 shocked the industry, achieving results comparable to top American AI models with far lower costs and computing power than U.S. tech giants like OpenAI. Some Western media called it the "DeepSeek moment," viewing it as a major breakthrough for China in the field of large AI models.

Reuters in the UK published an article on February 12, stating that more than a year has passed since the "DeepSeek moment," but Chinese tech companies have not slowed down at all, continuously launching new AI models with even stronger performance. Analysts pointed out that the reasoning capabilities of Chinese AI models are no less than those of their American counterparts, but the deployment costs are much lower, making them a focus of attention in the global industry.

IC photo

As the Spring Festival holiday approaches, Chinese tech companies have continuously launched impressive AI products. For example, ByteDance released a new video generation model, Seedance 2.0, which can understand complex natural language instructions and directly convert text or images into movie-level video content, unlike early AI video tools that could only generate fragmented clips.

The most groundbreaking capability of Seedance 2.0 is "single instruction to generate coherent multi-scene." The AI can automatically parse narrative logic, generate a sequence of connected shots, and maintain high consistency in characters, lighting, style, and atmosphere, achieving near "zero post-production" final output. This cross-shot narrative consistency has been regarded as a breakthrough innovation by the industry.

American billionaire Musk commented on a post about Seedance 2.0 on social media and exclaimed, "It's happening fast." Many industry professionals and foreign media outlets such as Reuters compared this to DeepSeek's breakthrough last year, calling it possibly the second "DeepSeek moment."

Meanwhile, several tech companies also have new moves. Zhipu released its new flagship model GLM-5, aimed at handling complex programming and intelligent agent tasks. Alibaba may also release Qwen 3.5, as developers have already submitted code support for the upcoming Qwen 3.5 series model on the HuggingFace project page in the AI open-source community.

Alfredo Montufar-Herrera, director general of the consulting company Ankaire based in China, said that despite the U.S. export restrictions on advanced AI chips, DeepSeek still achieved breakthroughs with its strong AI model last year, which both shocked the industry and made the market start paying attention to what products Chinese companies will launch next.

He told Reuters, "If these new models perform poorly, that would be surprising. I think the outside world generally has very high expectations for Chinese models."

The article points out that the usage cost of DeepSeek is much lower than many American products, and some investors have started to doubt that the statement that "only companies that spend hundreds of billions of dollars to build computing infrastructure can develop cutting-edge AI" is inaccurate.

Moreover, the industry previously believed that closed-source AI models would dominate, but the success of DeepSeek, an open-source model, has changed people's views. Now, more and more Chinese tech companies are embracing open source, significantly lowering the threshold for developers and enterprises to access cutting-edge AI technology, making "open source, low-cost" the norm.

Su Lianjie, chief analyst at the technology research company Omdia, said: "DeepSeek has shown the industry that even with limited resources, excellent models can be created. The combination of open-source access, strong reasoning capabilities, and low deployment costs has become a hallmark model for Chinese companies to build and promote foundational models."

Xiadong Bao, a fund manager at Edmunds Loehr Asset Management, told Bloomberg in the U.S.: "Before the real Spring Festival arrives, the Spring Festival for Chinese models has already begun. The performance gap between Chinese large language models and American large language models is rapidly narrowing."

This article is exclusive to Observer Net. Unauthorized reproduction is prohibited.

Original: toutiao.com/article/7605978413110772258/

Statement: The article represents the personal views of the author.