AI News (2026/2/12): GLM-5, the Latest Flagship Model from Zhipu, Open-Sourced: Achieves SOTA in Programming and Agent Capabilities
Executive Summary:
Zhipu AI has announced the open-sourcing of its latest flagship model, GLM-5, which has been scaled up to 744B parameters (40B active parameters) and trained on 28.5T of data. The model achieves state-of-the-art (SOTA) performance in programming and agent capabilities, with overall performance approaching Claude Opus 4.5. GLM-5 leads in multiple programming benchmarks and can complete complex system engineering and long-term agent tasks, supporting resource planning scenarios such as vending machine operation simulations.
GLM-5, the Latest Flagship Model from Zhipu, Open-Sourced: Achieves SOTA in Programming and Agent Capabilities
News Details
Zhipu AI announced on February 12 that it has open-sourced its latest flagship model, GLM-5. The model has been scaled up to 744B parameters, with 40B active parameters, and has been trained on 28.5T of data. GLM-5 has made significant progress in programming and agent capabilities, with overall performance approaching Claude Opus 4.5. This release marks another important breakthrough for Zhipu in the field of large models.
Key Points
Parameter Scale: GLM-5 has been scaled up to 744B parameters, with 40B active parameters. The pre-training data volume reaches 28.5T, ensuring high-performance across multiple tasks.
Programming Capabilities: GLM-5 performs exceptionally well in multiple programming benchmarks, including SWE-bench and Terminal Bench. These tests cover various aspects such as code generation, debugging, and optimization, where GLM-5 significantly outperforms existing open-source models.
Agent Capabilities: GLM-5 can complete complex system engineering tasks and long-term agent tasks end-to-end. For example, it supports simulations of vending machine operations and other resource planning scenarios, demonstrating its strong potential in practical applications.
AI-ALL In-Depth Commentary
The open-sourcing of Zhipu AI's GLM-5 model not only sets a new benchmark in parameter scale but also achieves the best performance in the open-source domain for programming and agent capabilities. This advancement is of great significance to the AI developer community, providing them with a powerful tool to tackle complex programming and resource planning problems. GLM-5's performance is on par with Claude Opus 4.5, indicating that open-source models can now match commercial models in certain areas. This will further drive the adoption and application of AI technology, especially in resource-constrained environments, where developers can leverage this model for efficient task processing and system optimization. Additionally, the open-sourcing of GLM-5 will foster collaboration and innovation between academia and industry, accelerating the development of AI technology.
