Chinese artificial intelligence (AI) Lab ZAI announced the release of two new Open Source General Language Model (GLM) on Monday. GLM-4.5 and GLM-4.5- Air Dumb, AI firm calls them their latest flagship model. These are both hybrid reasoning models that offer complex arguments and how to think of the use of the device, and how to think of a quick response. Additionally, the company says these model supports agent’s capabilities. In particular, the AI firm claims that its latest models improve all other open source models worldwide.
Z.Ai introduced the open source GLM-4.5 AI model
In a blog post, the Chinese AI firm announced the release of these models. The idea behind the GLM model is to create a large language model (LLM) that is really common and can do different types of work. The company believes that despite numerous developments in Generative AI, Google, Open AI, and Entropic’s choice models are not so common, as they perform strongly in some areas while living behind others. The company said, “GLM-4.5 strives for the purpose of combining all different abilities.”
The GLM-4.5 AI model includes a total of 355 billion parameters with 32 billion active parameters. On the other hand, 106 billion total parameters are found in air variations, with 12 billion parameters active. Both models unite reasoning, coding and agent capabilities in the same architecture. It has a 1,28,000 token context window and it comes with a local function calling capacity.
Coming into the model architecture, Z.AI chose the compound specialist (MOE) architecture to improve the computing performance of both the training and the diagnosis. Instead of increasing the width of the MOE layers (hidden dimensions and experts) like the DEPSEC -V3, the GLM -4.5 series reduces the series width while increasing the height (layers). This was done when the company believed that deep models showed better reasoning for reasoning.
The Chinese AI firm also lodged a novel techniques used in the blog post before training and after training to help the developer understand how the model was made from the beginning.
GLM-4.5 Series AI model performance
Photo Credit: Z.Ai
Z.Ai claims that it has tested the GLM-4.5 model performance on 12 benchmarks in agent, reasoning and coding. He then claimed to compare the model’s overall score against different LLMs from Openi, Entropic, Google, Z, Alibaba, and more. Based on this internal evaluation, the Chinese AI firm claims that the GLM-4.5 is ranked third, behind the Openi O3 and Xi’s Gark 4.
Interested people can get the open weight of these models from Z.Ai’s gut hub and hugging facial lists. As an alternative, these LLMs can also be obtained through the company’s website and application programming interface (API).
