Generative AI firms on the Chinese mainland have kicked off a fresh wave of open-source innovation, launching rivals to DeepSeek just months after its highly anticipated debut.
Beijing-based Zhipu.AI released GLM-4.5, a platform combining reasoning, coding and autonomous agent features. Zhipu.AI highlighted that GLM-4.5 can produce interactive mini-games and physics simulations without extra tools.
Alibaba's AI division rolled out an upgrade to its Qwen3 reasoning model, delivering smarter responses, expanded knowledge and improved agent capabilities. The Qwen team teased that bigger updates are on the horizon.
Newcomer Moonshot AI from Beijing introduced Kimi K2, an open-access model that lets developers skip training from scratch and freely customize and extend it.
Meanwhile, Shenzhen-based Tencent unveiled HunyuanWorld-1, a generative AI that crafts fully editable 360-degree virtual environments on the fly, setting a new benchmark for immersive experiences.
On the open model leaderboard at Hugging Face, GLM-4.5 leads, with Qwen3 and Tencent not far behind. Kimi K2 ranks ninth, and eight of the top ten models originate on the Chinese mainland.
At a recent expo in Beijing, Nvidia founder and CEO Jensen Huang praised the wave of open models DeepSeek, Alibaba, Tencent, MiniMax and Baidu Ernie bot for driving AI breakthroughs worldwide.
To date, the Chinese mainland has launched 1,509 of the 3,755 AI models released globally, topping the chart for total open-model contributions.
The Chinese government has proposed creating a global AI cooperation organization to accelerate international tech collaboration, underscoring a shift toward more open innovation.
As these new releases hit the scene, developers and startups worldwide are tapping into powerful new tools, ensuring the generative AI race stays vibrant and full of surprises.
Reference(s):
cgtn.com
