GLM-5-Turbo trades open weights for speed in agent loops

Z.ai introduced GLM-5-Turbo as a speed-oriented GLM-5 variant for agent workflows. Its documentation describes training aimed at tool invocation, instruction following and sustained sequences of actions. The intended benefit is faster execution across tasks that require repeated model decisions and external tools.

At the March 15 launch, the company called Turbo an experimental closed-source version, available through hosted access, and said its findings would feed a later open-source release. That announcement establishes the release strategy; it does not establish that keeping the model closed caused its speed advantage.