<div id="readability-page-1" class="page">China’s frontier artificial intelligence labs are ending the lunar year on a high note, racing to release a series of new AI models in the last few weeks of the Year of the Snake as the country gears up to celebrate the start of the Year of the Horse. Beijing-based Zhipu AI and Shanghai-quartered MiniMax, which just made their stock exchange debuts in Hong Kong, would be the next to update their flagship AI systems with across-the-board enhancements, according to sources familiar with their release plans. Zhipu AI, known as Z.ai internationally, is set to launch the fifth iteration of its flagship GLM model series, GLM-5, in the next two weeks, before the Lunar New Year starts the week of February 15. GLM-5 was expected to come with comprehensive and significant improvements in creative writing, coding, reasoning and agentic capabilities, according to two sources. Meanwhile MiniMax would release its M2.2 – a minor update to its M2.1 model with coding-focused enhancements – before the holiday, said the sources, who requested anonymity as they were not authorised to speak with the media. Zhipu AI did not immediately respond to a request for comment. MiniMax declined to comment. Alibaba Group Holding and Moonshot AI kicked off the recent flurry of releases with their Qwen3-Max-Thinking and Kimi 2.5 models last week. Alibaba owns the South China Morning Post. Chinese firms were seemingly following a pattern set by Hangzhou-based AI start-up DeepSeek when it released its high-performance and low-cost V3 and R1 models before Lunar New Year last year. The release sent shock waves through the global tech world while China celebrated the holiday. Despite widespread speculation that its next release would come around this Lunar New Year, DeepSeek had stayed mum on its plans and was not expected to release a major update as it did last year, according to a source. DeepSeek was likely to publish a minor update to its V3 model family, which had been receiving upgrades throughout 2025, the source said.  The company’s next flagship model was expected to be a trillion-parameter foundational model, according to the source. The ballooning of the model size had slowed down training and delayed its release, the source said, adding that under the current situation Moonshot AI’s Kimi 2.5 would remain the most advanced open-source release until a new festive release unseated it. DeepSeek did not respond to a request for comment. The term parameters refers to variables present in an AI system during training and is a key factor in the sophistication and effectiveness of the AI model. Alibaba-backed Moonshot’s Kimi 2.5 launched to global acclaim, with benchmark tests showing that it brought China’s open-source AI capabilities closer to leading US closed-source systems than ever before. The model is currently ranked fifth on consultancy Artificial Analysis’ Intelligence Index, tailing only top models from US labs. It is the only open-source model and the only Chinese model in the top five. More Chinese tech companies are looking to capitalise on the festive season, a key period for product visibility and user acquisition as the whole country is on holiday. China’s biggest tech firms are spending billions to help their AI chatbot apps win users. Social media and gaming company Tencent said its Yuanbao chatbot would give away 1 billion yuan (US$144 million) in digital red envelopes, the wired version of the traditional packets of money given for luck during the festival. Baidu is doing the same with 500 million yuan through its Wenxin app. Alibaba announced a 3 billion yuan campaign to promote its Qwen chatbot app on Monday. Additional reporting by Vincent Chow</div>

TCTZF

02513

00100

<p>China's AI labs are racing to release new models before the Lunar New Year. Zhipu AI plans to launch its GLM-5 model, featuring significant enhancements, while MiniMax will update its M2.1 model with coding improvements. This follows recent releases from Alibaba and Moonshot AI. The festive season is crucial for product visibility, with major firms like Tencent and Baidu offering substantial giveaways to attract users. Despite speculation, DeepSeek is not expected to release a major update this year, focusing instead on minor improvements to its existing models.</p>

China’s AI labs race to debut latest models before Lunar New Year