Image: Z.ai / zcode.z.ai Z.ai Launches ZCode, a Free AI Coding IDE Built on GLM-5.2, to Challenge Cursor and Claude Code
The Beijing-based AI company behind GLM-5.2 has released ZCode, a desktop coding environment for macOS, Windows, and Linux that lets users trigger agents remotely through WeChat or Telegram.
Z.ai — the Beijing-based company formerly known as Zhipu AI — launched ZCode on July 2, a desktop AI coding environment available for macOS, Windows, and Linux. The app is free to download and positions itself as a direct competitor to Cursor, Claude Code, and GitHub Copilot, built around Z.ai’s own GLM-5.2 model with a 1 million-token context window.
ZCode organizes work around what it calls a “Goal” — you describe what you want, the agent plans it out, edits files, checks its own work, and iterates until the task is done. That framing is similar to how Cursor’s Agent mode or Claude Code’s background agents operate, but ZCode adds something Western tools don’t: the ability to kick off and monitor coding tasks remotely through WeChat, Feishu, or Telegram. You can start a build from your phone in a chat app and get notified when it’s done.
The app ships with more than 20 bundled tools including Git and a terminal, and supports bring-your-own-key (BYOK) for third-party models from Anthropic, OpenAI, and Google if you’d rather not use GLM-5.2 directly.
Pricing is well below Western competitors
ZCode’s pricing puts a lot of pressure on the US market. The Lite plan is $16.20 per month. The Pro plan is $64.80 per month and includes MCP tool access. The Max plan is $144 per month with 20x usage quota. Cursor’s comparable tiers sit at $20, $60, and $200 per month. Claude Code doesn’t have a standalone product subscription; you pay per token through your Anthropic plan.
A promotional offer through July 31 gives subscribers 1.5x their normal usage quota.
Z.ai is also running a $30,000 credits grant program for developers who want to build on GLM-5.2 via the API.
The data question
ZCode’s differentiation comes with a tradeoff most enterprise teams will clock immediately: source code and prompts route through Z.ai’s infrastructure, which is subject to Chinese data law. The BYOK option moves the model compute to whichever provider you choose, but the ZCode desktop client itself still sits in the middle.
Z.ai provides some mitigation in the form of MIT-licensed open weights for GLM-5.2, which means self-hosting is an option for teams with the infrastructure to do it. That path removes the dependency on Z.ai’s servers entirely, though it requires meaningful setup.
For individual developers working on open-source projects or non-sensitive code, those concerns probably don’t apply. For any company handling proprietary IP, regulated data, or client code, the data governance question is going to be the first thing legal asks.
The market context
ZCode arrives when Chinese AI coding tools are having a moment. Kimi K2.7 Code just got added to GitHub Copilot as the first open-weight model in that catalog. DeepSeek V4 Pro is in Zed. GLM-5.2 itself is available in several third-party editors via BYOK configurations.
Z.ai is making a bet that having a first-party IDE — rather than just an API — is where the real usage happens. Cursor proved that thesis for Anysphere. Whether ZCode can replicate it with a Chinese-built model that carries its own set of enterprise objections is a different question.
Source: VentureBeat, ZCode product page