Zhipu Launches GLM-5-Turbo: The First Lobster-Specific Scene Native Model, Self-Developed Benchmark Remains Top in Domestic

Today, Zhipu officially released the GLM-5-Turbo base model, which is deeply optimized for complex Agent scenarios, aiming to address the industry pain point of general large models being prone to slowdowns in long-chain tasks.

This model integrates the "OpenClaw" scenario-native genes from the training phase, focusing on enhancing core capabilities such as tool calling, complex instruction decomposition, scheduled triggers, and high-throughput continuous execution. In the self-developed benchmark test ZClawBench, GLM-5-Turbo ranks first among domestic models and has received a 90% approval rate in developer blind tests.

As the usage ratio of Skills in the OpenClaw ecosystem rapidly increased from 26% to 45%, Agents are evolving from conversation tools to modular productivity tools. To this end, Zhipu also launched the "OpenClaw Subscription" system and an enterprise-level security management system (Claw for Enterprise Security), transforming the complex multi-Agent collaboration process from a "black box" into visualized management through permission orchestration and real-time monitoring.

The model is now first integrated into the world's first native AI Agent terminal, "OpenClaw Box." Starting from March 16, 2026, developers can call the relevant API via the Zhipu open platform BigModel.cn.

The release of GLM-5-Turbo marks that the focus of the large model competition has shifted from single semantic understanding to end-to-end execution efficiency. Through specialized optimization of base capabilities and the improvement of commercial support, Zhipu is accelerating the key transition of large models from "efficiency-enhancing tools" to "enterprise digital workforce," providing a standard model for the large-scale commercialization of AI Agents.