Zhipu AI Upgrades ZCode Coding Harness with Agentic Features and Quota Resets
Zhipu AI has announced a major upgrade to ZCode, its coding harness optimized for GLM-5.2, introducing new features like Goal mode, Subagents, Remote Control, and Idle Tasks. Additionally, user quotas for the GLM Coding Plan have been fully reset, along with a temporary 1.5x quota bonus. The introduction of Goal mode enables ZCode to autonomously decompose, execute, and test complex tasks without constant developer supervision, pushing coding assistants toward true agentic autonomy. Furthermore, optimized context caching achieves a 98% hit rate, significantly reducing token costs and improving efficiency for developers. In benchmarks against Claude Code using GLM-5.2, the GLM+ZCode combination achieved a 2.39% higher overall task pass rate. The Remote Control feature allows developers to monitor and control desktop coding tasks via mobile apps like WeChat and Feishu without syncing project code to the phone.
## BACKGROUND
ZCode is Zhipu AI's official open-source coding agent harness designed to work with its GLM-5.2 model, competing with tools like Cursor and Anthropic's Claude Code. Context caching is an LLM optimization technique that stores and reuses previously processed input tokens, reducing API latency and inference costs for repetitive prompts.