Chinese artificial intelligence (AI) companies are racing to launch new models during the Lunar New Year holiday period, seeking to create the next breakthrough moment, like the one achieved by DeepSeek.
A few days ago, a mysterious test version dubbed Pony Alpha became a hot topic in the AI community thanks to its exceptional coding performance, sparking widespread speculation about its developer. On Thursday (February 12th), it was revealed that the model was GLM-5 developed by Beijing-based Zhipu AI.
The new model scored high on various state-of-the-art AI agent benchmarks and has been successfully used in many AI agent workflows during its stealth testing period, according to a post by OpenRouter on the social media platform X. The closed testing approach demonstrates the growing confidence of Chinese developers in their products' capabilities.
Late last month, Moonshot, another Beijing-based technology company, launched the latest version of Kimi 2.5, surprising the AI community with its open-source performance across agent tasks, programming, image processing, video processing, and a range of general intelligence tasks. The open-
source model is expected to cut costs by up to 90 percent, creating a "Kimi 2.5 moment," said Chamath Palihapitiya, a prominent Silicon Valley analyst. He also called the model "very significant."
DeepSeek also began private grayscale trials of its app and web interface. The context window length (the amount of text that can be processed simultaneously) was immediately increased from 128,000 tokens to 1 million tokens, indicating that significant improvements are in the pipeline.
The push to replicate the significant global impact of DeepSeek's R1 launch in January 2025 has also extended to the realm of AI video and image processing.
ByteDance's recently tested Seedance 2.0 text-to-video platform, capable of generating multi-shot movie sequences in about 60 seconds with a relatively simple prompt, has sparked a global online short video creation frenzy.
"The infancy of AIGC is over," said Feng Ji, creator of the global hit game Black Myth: Wukong, discussing Seedance 2.0.
"Fortunately, at least for now, Seedance 2.0 is from China," added Feng, who was impressed by the AI's capabilities.
Additionally, Alibaba's AI division, Qwen, launched its next-generation image-generation baseline model, Qwen-Image 2.0, on Tuesday (February 10). The open-source Qwen model will power AI tools for the International Olympic Committee in preparation for the 2026 Milan Winter Olympics.
Chinese tech giants, including Alibaba, Tencent, and ByteDance, are leveraging the Lunar New Year holiday season to entice users to adopt their intelligent assistants. They are gearing up to accelerate ecosystem development, from research and development (R&D) to consumer adoption, based on their respective AI models and platforms.
China is projected to have 1.125 billion internet users by the end of 2025, and the number of generative AI technology users in the country will reach 602 million, a 141.7 percent jump from the end of 2024, according to the China Internet Network Information Center.
