Chinese artificial intelligence (AI) company Zhipu AI has launched its latest flagship model, the GLM-5, with a leap in coding capabilities amid a heated race among Chinese tech firms to roll out major new models ahead of the Spring Festival holiday.
Zhipu AI, known internationally as Z.ai, said that the GLM-5 represented a shift in AI development from “vibe coding” to “agentic engineering” – AI-automated coding at a larger scale – thanks to its latest model’s enhanced performance.
The company said that GLM-5 achieved industry-leading scores for open models in key areas such as coding and agentic performance according to internal tests, even surpassing Google DeepMind’s most powerful model, Gemini 3 Pro .
However, the model still lagged Anthropic’s Claude, a proprietary model and widely considered the world’s best coding model, across the board on coding benchmarks, according to the company’s self-reported scores.
The new model doubled the size of its predecessor, GLM-4.7, from 355 billion to 744 billion parameters and increased the training data to 28.5 trillion tokens. It also adopted a new architecture, first pioneered by the Hangzhou start-up DeepSeek , called DeepSeek Sparse Attention , designed to maximise computational and cost efficiency.
05:04
China creates analogue AI chip said to be 1,000 times faster than Nvidia GPU
China creates analogue AI chip said to be 1,000 times faster than Nvidia GPU
The new model is available on Zhipu’s official website and has been open-sourced on GitHub and Hugging Face .