Breaking News: Zhipu.AI Open-Sources High-Speed AI Models
Beijing, China – April 15, 2025 – Chinese AI company Zhipu.AI has taken a major step toward global expansion by open-sourcing its latest General Language Models (GLM), including the ultra-fast GLM-Z1 inference model. The move signals a strategic push ahead of a potential initial public offering (IPO).
The company also launched an international domain, Z.ai, making the models freely accessible to developers worldwide.
“We believe open-source is the fastest path to AI innovation and global adoption,” said a Zhipu.AI spokesperson. “Our models are designed to empower developers with unprecedented speed and autonomy.”
Blazing-Fast Inference: 200 Tokens per Second
The GLM-Z1-32B-0414 model achieves inference speeds up to eight times faster than DeepSeek-R1, according to Zhipu.AI. Optimizations in GQA parameters, quantization, and speculative sampling enable 200 tokens per second on consumer-grade GPUs.
“That’s 50 times faster than human reading speed,” the company claims, positioning the model as a leader in efficient AI inference.
Autonomous ‘Rumination’ Model Unveiled
Zhipu also introduced the GLM-Z1-Rumination-32B-0414, a model capable of autonomous internet search, tool use, and self-verification. “This is a leap toward agents that can tackle open-ended problems without constant human prompting,” the spokesperson added.
Complete Open-Source Portfolio
The release includes:
- GLM-4-32B-0414: Enhanced for agent capabilities, with real-time code generation in HTML, CSS, JS, and SVG.
- 9B parameter versions of both GLM-4 and GLM-Z1, for resource-constrained environments.
- All models released under the permissive MIT license.
Smaller models retain strong performance in mathematical reasoning and general tasks, broadening adoption.
Background
Zhipu.AI, founded in 2019, has become one of China’s leading AI startups, competing with Baidu and Alibaba. The company has previously released GLM-130B and other models, gaining traction in both domestic and international markets.
Industry analysts view this open-source push as a prelude to a public listing. “Zhipu is building a developer ecosystem to boost valuation ahead of an IPO,” said Dr. Li Wei, a tech analyst at Beijing-based consultancy Horizon Insights.
What This Means
Open-sourcing high-speed models democratizes access to cutting-edge AI, potentially displacing proprietary offerings from Western firms. “Zhipu’s speed advantage could shift developer preference in the open-source community,” noted Sarah Chen, an AI researcher at Tsinghua University.
For enterprises, Zhipu continues its Model-as-a-Service (MaaS) platform, now integrating the new models with tiered pricing – including a free tier (GLM-Z1-Flash). The combination of open-source and commercial options positions Zhipu to capture a wide market share.
“The global AI race is heating up, and Zhipu is betting on openness to win,” Chen added.
Global Expansion via Z.ai
The Z.ai domain serves as an international portal for developers to test the models via web interface or dedicated app. This move mirrors strategies by Meta and Mistral AI, signaling Zhipu’s ambition to become a global AI standard.
This is a developing story. Check back for updates.