News
China's Z.ai Launches Free Coding Agent ZCode to Rival Cursor and Claude Code

Beijing-based Z.ai, formerly Zhipu AI, has released ZCode, a free coding environment built on its GLM-5.2 model that undercuts Western rivals like Cursor and Claude Code on price, while raising questions about data security under China's national intelligence law.
Beijing-based Z.ai, previously operating under the Zhipu AI brand, launched ZCode in the week of July 1, 2026, a free coding environment built on its own GLM-5.2 model. The tool directly targets users of Cursor, Claude Code, and GitHub Copilot, offering similar agentic coding features at a fraction of Western prices.
ZCode operates as a so-called Agentic Development Environment, where the user describes a goal and the agent independently reads the code, plans changes, edits files, runs terminal commands, and iterates until it reaches a result. Unlike Cursor or Claude Code, which started as code editors with an agent layer bolted on, ZCode puts the conversation with the agent at the center of the interface.
Cheaper Plans, Free to Start
The desktop app is completely free to download, and new users get a five-day trial with five million tokens per day. Paid subscription plans start at $18 a month for the Lite tier, putting pricing well below Western competitors, which typically charge between $20 and $200 a month for comparable access levels. The company notes, however, that the promotional usage multiplier expires on July 31, after which peak- and off-peak-hour rates revert to standard levels starting in August.
Among the features that set ZCode apart from Western tools is the ability to launch coding tasks remotely straight from WeChat, Feishu, or Telegram without opening the desktop app, and to control a running agent from a phone after pairing via QR code. The tool also supports a bring-your-own-key (BYOK) model, letting teams connect their own models from Anthropic, OpenAI, OpenRouter, Moonshot, or MiniMax instead of relying solely on GLM-5.2.
Data Security Concerns
The biggest controversy surrounding ZCode is the fact that it comes from a Chinese company, which, as an entity registered in China, is subject to the country's 2017 National Intelligence Law and its 2021 Data Security Law. Both statutes require Chinese organizations to cooperate with intelligence agencies on request, a risk the U.S. Department of Homeland Security has previously flagged for data belonging to non-Chinese citizens and companies that passes through Chinese infrastructure.
The open weights of GLM-5.2, released under the MIT license, in theory let companies download, modify, and run the model on their own infrastructure, removing dependence on Z.ai's cloud at the model level. In practice, however, the full-precision version of the model requires about 1.5 terabytes of GPU memory, putting self-hosting out of reach for most development teams. The ZCode application itself remains closed-source, so task orchestration, work history, and bot integrations still pass through Z.ai's systems even when teams supply their own model key.
A closed agentic system from a Chinese vendor is, in practice, a black box with full user permissions - comment cited in Digital Applied's analysis of ZCode
What It Means for Developers in Poland
For Polish development teams looking for a cheaper alternative to Cursor or Claude Code, ZCode's pricing may be tempting, especially for hobby projects and smaller startups. But companies working with code covered by client confidentiality, personal data, or public-sector contracts should carefully examine what data actually flows through Z.ai's infrastructure before deploying the tool in a regulated environment. It's the same type of risk that previously led Alibaba to ban employees from using Claude Code, just in reverse geographically.
ZCode's launch also shows how quickly Chinese AI companies are closing the gap in developer tools, not just in language models themselves. Z.ai shipped eight public app releases in the eight days after launch, suggesting the company treats this market segment as a strategic priority rather than a side effect of GLM model development.
Sources: VentureBeat (venturebeat.com), Tech Times (techtimes.com), Digital Applied (digitalapplied.com), Let's Data Science (letsdatascience.com)

