ZCode is a China-developed Coding Harness deeply optimized for Zhipu GLM. Today, Zhipu announced a major ZCode upgrade, officially launching four features: Goal, Subagents, Remote Control, and off-peak tasks.
As a special reward marking the milestone of one million users, the quotas for all GLM Coding Plan users were reset uniformly today at 13:00, fully replenishing everyone’s quota.
In Zhipu’s internally developed Z.ai Code Bench, the official team tested GLM-5.2 paired separately with ZCode and Claude Code. The results showed that GLM+ZCode achieved a 2.39% higher overall task pass rate than GLM+Claude Code; in terms of checkpoint pass rate, ZCode was 1.22% lower.
ZCode has been specifically optimized for context cache reuse. According to Zhipu’s official tests of multiple Coding Agents, GLM achieved a cache hit rate of over 98% in ZCode, allowing more repeated context to hit the cache and be deducted at a lower credit multiplier, thereby increasing the effective Token volume of the GLM Coding Plan by approximately 30%.
From now through August 31, 2026, using the GLM Coding Plan in ZCode will also provide a limited-time 1.5x quota boost. With the two effects combined, overall usage will approach 1.8 times the regular quota.
In addition, ZCode has launched an off-peak task feature. After GLM Coding Plan users submit an off-peak task, ZCode will execute it automatically during low-traffic periods. Within the benefits available for off-peak tasks, users’ GLM Coding Plan credits will not be deducted, making the feature suitable for tasks that are not urgent but take a long time to complete.
The most important change in this upgrade is that the Agent can now complete complex tasks on its own.
When using a Coding Agent in the past, developers often had to stay nearby and keep pushing it forward: after submitting a request, they had to wait for the Agent’s response and then ask it to continue making changes; if a test failed, they had to follow up, and if something was overlooked, they had to provide the context again.
Goal mode changes this process. Now, you only need to set a clear, verifiable objective for ZCode, such as keeping the initial page load time under 2 seconds while ensuring that all existing tests pass.
ZCode will then automatically break down the task, modify code, run commands, and execute tests, before determining whether the goal has been completed based on files, command output, and test results. If the requirements are not met, it will continue to the next round until the task is complete. The progress, time spent, and execution results for each round are all clearly displayed in the Goal panel.
In addition, Remote Control temporarily makes the current ZCode desktop window accessible from a phone. When a long-running task is running on a computer, developers can use their phones to check progress, continue entering commands, create new tasks, or reconnect to a disconnected workspace.
In addition to QR codes and browser links, ZCode supports WeChat, Feishu, and Lark. Users can open their own workspace directly from a WeChat or Feishu conversation and track the Agent’s execution progress. The phone serves only as a control interface; code and commands continue running in the original desktop environment. Projects are not synchronized to the phone, nor is an additional cloud environment created. Local, SSH, WSL, and Docker workspaces are all supported.
