Zhipu AI launched ZCode, a terminal-native coding agent optimized for its GLM-5.2 model. The product mirrors Anthropic's Claude Code in interface and workflow, but runs on Chinese LLM infrastructure.
Key facts
- ZCode launched by Zhipu AI, optimized for GLM-5.2 model.
- Three pricing tiers: Lite, Pro, Enterprise.
- Supports 20+ programming tools with deep ZCode integration.
- Remote execution via WeChat, Feishu, or Telegram.
- Goal management system for multi-step task decomposition.
Zhipu AI, the Beijing-based developer of the GLM family of large language models, has released ZCode, a terminal-native coding agent that closely resembles Anthropic's Claude Code. The product, announced via Zhipu's official site, is positioned as a direct competitor to Claude Code for developers working within Chinese AI ecosystems.
Product Architecture
ZCode operates as a command-line interface tool that integrates with Git repositories and executes autonomous coding tasks. The agent supports over 20 programming tools, including deep integration with Zhipu's own ZCode platform—a naming choice that creates some confusion between the agent and its hosting environment.
The core differentiator is optimization for GLM-5.2, Zhipu's latest flagship model. The company claims the combination delivers improved reasoning, coding accuracy, and multi-agent collaboration compared to prior GLM iterations. Zhipu did not release benchmark scores for ZCode on standard coding evaluations like SWE-bench or HumanEval.
Pricing and Tiers
ZCode offers three subscription tiers, all denominated in Chinese yuan:
Lite Light development tasks Baseline Pro Professional workflows 5x Lite quota Enterprise High-frequency, large-scale 20x Lite quotaThe pricing page notes that specific costs and quotas may change, directing users to Zhipu's subscription portal for final details. [According to ZCode's site], each tier includes access to the latest GLM models and features.
Remote Execution and Ecosystem
A notable feature is remote task invocation: users can trigger ZCode via WeChat, Feishu (ByteDance's enterprise messaging platform), or Telegram. This allows developers to queue coding tasks from mobile devices, with results delivered back through the same channels.
The product also includes a "Goal" management system for decomposing complex multi-step tasks into sub-goals, with planning, execution, and verification phases—similar to Claude Code's /loop command for long-horizon tasks.
Competitive Positioning
ZCode enters a market where Claude Code has established itself as a leading terminal agent, scoring 88.6% on SWE-bench Verified and 78.9% on Terminal-Bench 2.1 with Opus 4.8. Zhipu's offering targets developers who prefer or require Chinese-language interfaces and domestic AI infrastructure, particularly amid ongoing US-China technology restrictions.
The product's similarity to Claude Code—down to the CLI interaction pattern and agentic workflow—suggests Zhipu is prioritizing compatibility with existing developer workflows rather than introducing novel interaction paradigms. The company did not address whether ZCode uses any open-source components from Anthropic's Model Context Protocol (MCP), which Claude Code relies on for tool integration.
Community Reception
On Hacker News, the announcement garnered 274 points and 14 comments. The top comment noted the site is entirely in Chinese with no obvious language toggle on mobile, underscoring the product's domestic focus. [HN commenters] raised questions about data privacy and whether ZCode sends code to Chinese servers for processing.
Zhipu has not disclosed whether ZCode will be available internationally or whether it supports English-language codebases and documentation. The company's existing GLM models have primarily targeted Chinese-language applications, though they have demonstrated competitive performance on multilingual benchmarks.
What to watch
Watch for Zhipu to release benchmark scores on SWE-bench Verified or Terminal-Bench. If ZCode scores within 10% of Claude Code's 88.6% on SWE-bench Verified, it would signal GLM-5.2 has reached parity with frontier Western models on coding tasks—a milestone for Chinese LLM development.
Source: zcode.z.ai
