The new artificial intelligence GLM-5.2, developed in China, is disrupting the AI market by outperforming ChatGPT on complex tasks and offering a significant advantage: running without relying on the cloud. This means users can take advantage of the technology without paying expensive subscriptions or sending information to third-party servers.
This new AI, released by the company Z.ai (formerly Zhipu AI), can run directly on the user's local hardware, which is a major advantage in terms of privacy and autonomy. Imagine having an accounting system that can process large volumes of data without sending information to the cloud, giving you more security and control over your data.
1. What makes GLM-5.2 so special?
GLM-5.2 is an open-weights AI model, meaning its "brain" is made publicly available for download and modification. In tests on the Code Arena platform, which evaluates the ability to create and fix programming code, the system reached second place globally, behind only Claude Fable 5 from Anthropic. It also outperformed GPT-5.5 on practical tasks, such as fixing real-world software bugs and math exams.
One of the main highlights of GLM-5.2 is its processing capacity. With a context window of 1 million tokens, it can analyze massive files, such as entire books or complete programming libraries, in a single pass. It is like having a librarian who can read and process every book in a library in seconds.
2. What are the technical challenges?
Despite the privacy and autonomy advantages, the technical requirements for running the system on your own workstation are high. GLM-5.2 was built on an architecture known as Mixture-of-Experts (MoE), a design technique that splits AI processing into smaller, specialized networks to save energy and speed up responses.
However, the model requires a massive amount of storage because it loads between 744 billion and 753 billion data parameters. In its original, uncompressed format, the AI files take up 1.51 terabytes of memory. For the system to run on commercial computers, developers need to apply a compression method called quantization, which reduces file size while sacrificing as little precision as possible.
3. How did the trade blockade accelerate the launch?
The launch of GLM-5.2 happened against a backdrop of strong geopolitical and commercial tension between the United States and China. On June 12, the U.S. Department of Commerce ordered Anthropic to immediately cut foreign access to its most advanced models. The Chinese company took advantage of the gap left by the unavailability of competing services to launch its open model under the permissive MIT license the following day.
The strategy pushed Z.ai's market value to around US$128 billion on the Hong Kong Stock Exchange. Since downloading the model is free and access to its cloud services costs about one tenth of what Western companies charge, the market expects financial pressure on the traditional Big Tech players, making next-generation digital tools cheaper and more private for the end user.
Frequently asked questions
What is GLM-5.2?
GLM-5.2 is a new artificial intelligence developed in China that outperformed ChatGPT on complex tasks and offers the ability to run without relying on the cloud.
What are the main advantages of GLM-5.2?
The main advantages of GLM-5.2 are its ability to run directly on the user's local hardware, offering privacy and autonomy, and processing large volumes of data without sending information to the cloud.
What are the technical challenges of GLM-5.2?
The main technical challenges of GLM-5.2 are the requirement for a large amount of storage and the need to apply compression methods for the system to run on commercial computers.