Z.ai, a Chinese artificial intelligence startup formerly known as Zhipu AI, has introduced its latest innovation: the GLM-4.5, an open-source AI model designed for intelligent agent applications. Unlike traditional AI models, GLM-4.5 is built around “agentic” principles, meaning it can autonomously break down tasks into smaller, more manageable sub-tasks, allowing for more efficient problem-solving. This new release is significant, not only because of its cutting-edge features but also because it is freely available to developers, enabling them to incorporate the model into their own projects.
The introduction of GLM-4.5 comes at a time when competition in the global AI market is intensifying. Chinese tech giants are quickly advancing in the AI race, with Tencent unveiling its HunyuanWorld-1.0 model, which creates interactive 3D scenes from text or images, and Alibaba launching Qwen3-Coder for software development. Z.ai’s move to release an open-source model places it in direct competition with these players, reflecting China’s growing influence in the AI sector. As of mid-2025, China leads the world in large-language model releases, with over 1,500 models developed so far.
GLM-4.5 is powered by a Mixture of Experts (MoE) architecture, boasting 355 billion parameters with 32 billion active parameters. This architecture is designed for large-scale applications, excelling in reasoning, generation, and multi-agent tasks. There is also a more compact version, the GLM-4.5-Air, with 106 billion total parameters and 12 billion active parameters, offering flexibility for different use cases. Z.ai’s unique selling point lies in its ability to seamlessly integrate reasoning, coding, and agentic capabilities into a single model, making it an ideal solution for a variety of fast-evolving agent-based applications.
The GLM-4.5 model supports two distinct modes of operation: “thinking mode” and “non-thinking mode.” In “thinking mode,” the model takes more time to process complex tasks like math, coding, and logic problems, delivering highly accurate results. On the other hand, the “non-thinking mode” is optimized for quick, direct responses, making it perfect for more casual or general-purpose tasks. This flexibility enables the model to handle both high-level reasoning and everyday queries efficiently.
Z.ai’s GLM-4.5 models are equipped with multi-step reasoning, external tool integration, and function calling capabilities. This means users can give natural language instructions to perform tasks such as web browsing, creating presentations, or building websites. Additionally, the model can assist in app design, code generation, and even interactive game creation. It is fully compatible with existing coding environments like Claude Code and CodeGeex, offering a seamless experience for developers.
In terms of performance, the GLM-4.5 series has been benchmarked across 12 industry-standard tests, including MMLU, GSM8K, and HumanEval. The full GLM-4.5 version scored an impressive 63.2, ranking it third overall and second globally, while the GLM-4.5-Air scored 59.8, making it the leader among models with approximately 100 billion parameters. The model also excels in specific areas, such as tool-calling, with a success rate of 90.6%, surpassing other leading models like Claude 3.5 Sonnet and Kimi K2. Additionally, GLM-4.5 has proven to be particularly strong in Chinese-language tasks and coding, delivering state-of-the-art results across various open benchmarks.
When it comes to cost efficiency, Z.ai’s GLM-4.5 series is highly competitive. API calls are priced at just $0.11 per million input tokens and $0.28 per million output tokens. For those seeking even faster performance, a high-speed version can generate over 100 tokens per second, offering both speed and affordability.
In conclusion, Z.ai’s GLM-4.5 is a game-changer in the world of artificial intelligence, merging advanced agentic capabilities with outstanding performance across a wide range of tasks. With its robust architecture, dual-mode reasoning, and impressive benchmarks, it’s poised to become a key player in both academic and commercial AI applications. As the AI landscape continues to evolve, GLM-4.5 is setting a new standard for what is possible in the field.
