Introducing GLM-4.5, the latest flagship model from ZhiPu AI, a groundbreaking foundation model designed for intelligent agent applications. This cutting-edge model is now available on the Hugging Face and ModelScope platforms, with its model weights released under the MIT License, making it accessible to developers worldwide.
Achieving State-of-the-Art Performance with GLM-4.5
GLM-4.5 has set a new standard in open-source models, demonstrating exceptional performance in a comprehensive set of abilities, including reasoning, coding, and intelligent agent capabilities. In real-world evaluations, it has emerged as the top-performing model in China, showcasing its superiority in artificial intelligence. Across 12 benchmark tests, including MMLU Pro, AIME24, MATH500, and SciCode, GLM-4.5 has secured a remarkable third place globally, first among domestic models, and first among open-source models.
A Milestone in AI: Integrating Reasoning, Coding, and Agent Capabilities
GLM-4.5 is the first model to natively integrate reasoning, coding, and intelligent agent capabilities within a single framework, meeting the complex demands of intelligent agent applications. This technical breakthrough is a testament to ZhiPu AI's commitment to the first principles of Artificial General Intelligence (AGI), blending a wide range of general intelligence capabilities without compromising existing abilities.
Architectural Excellence with MoE: GLM-4.5 and GLM-4.5-Air
GLM-4.5 employs a Mixture of Experts (MoE) architecture, offering two versions to cater to different needs. The standard GLM-4.5 boasts 35.5 billion total parameters and 3.2 billion active parameters, while the GLM-4.5-Air variant contains 10.6 billion total parameters and 1.2 billion active parameters. Despite having fewer parameters than DeepSeek-R1 and Kimi-K2, GLM-4.5 outperforms these models in multiple standard benchmarks, positioning itself at the forefront of performance-to-parameter ratio on the SWE-bench Verified leaderboard.
Dual Operating Modes for Enhanced Flexibility
GLM-4.5 supports two operating modes to cater to different application scenarios. The 'thinking mode' is designed for complex reasoning and tool usage, while the 'non-thinking mode' facilitates immediate responses. The training process involves pre-training on 15 trillion tokens of general data, followed by targeted training on 8 trillion tokens of data in coding, reasoning, and intelligent agent domains. The model's capabilities are further enhanced through reinforcement learning.
Cost-Efficient and High-Performance: GLM-4.5 Series
The GLM-4.5 series represents a significant breakthrough in cost and efficiency. With API call prices significantly lower than mainstream models, it offers an economical solution for developers. The input cost is as low as 0.8 yuan per million tokens, and the output cost is 2 yuan per million tokens. The high-speed version can generate up to 100 tokens per second, supporting low-latency, high-concurrency deployment requirements.
Full-Stack Development Capabilities with GLM-4.5
GLM-4.5 is capable of handling full-stack development tasks, including front-end website coding, back-end database management, and tool invocation interface support. The model is deeply optimized for full-stack programming and tool invocation, compatible with mainstream code intelligence frameworks such as Claude Code, Cline, and Roo Code, offering one-click compatibility with the Claude Code framework.
Accessing GLM-4.5 and Ensuring Transparency
Users can access the GLM-4.5 API through the BigModel.cn open platform or experience the full functionality for free at chatglm.cn and z.ai. To ensure transparency in evaluations, ZhiPu AI has published 52 test questions and agent trajectories for industry verification and replication.
The launch of GLM-4.5 signifies a new technological foundation for intelligent agent application development. Its core capabilities of native integration, exceptional cost-performance ratio, and extensive ecosystem compatibility are poised to drive the规模化 application and commercial value realization of AGI technology across various industries.