Z.ai launches ZCode 3.2.2 as official GLM-5.2 coding harness

Z.ai has quietly stood up its own coding IDE. ZCode, pitched on the company's site as the 'Official Harness for GLM-5.2,' is now on version 3.2.2 and available for macOS, Windows and Linux, with the Linux build still marked Beta. It is a familiar move, a model vendor shipping the first-party tool around its own model, and it is starting to look less like a niche play and more like the default packaging for a serious coding LLM.

The pricing is where the interesting signal is. The GLM Coding Plans start at $16.20 a month for Lite, with Pro at $64.80 for what Z.ai calls 5x Lite usage, and Max at $144 for 20x. The tiers are gated by repo size and feature access as much as raw allowance. Lite is aimed at small-repo iteration, Pro adds MCP tools and faster generation, and Max promises first access to new features and dedicated resources. That undercuts the going rate for comparable Claude Code and Cursor tiers, which is a plausible pitch if you already trust GLM-5.2 on real work and a much harder one if you don't.

The feature list leans into where agent tooling has been going. Long-running Goals for continuous planning, execution and verification. Multi-agent collaboration called out as a v3.0 selling point. And remote bot control from WeChat, Feishu and Telegram, which tells you plainly who the primary audience is expected to be.

The honest caveat is that everything here is Z.ai's own copy on Z.ai's own site, so the multipliers, the harness quality and the actual GLM-5.2 coding ability all have to be taken as marketing until independent benchmarks or long-form user reports show up. What the product page does not give you is a token budget for those tiers, a data-residency story for teams that would be nervous about routing prompts through consumer chat apps, or any head-to-head against Claude Code or Cursor on real repositories.

If GLM-5.2 turns out to be as strong on code as this framing implies, the interesting question is whether a $16.20 entry price plus a first-party harness is enough to pull a chunk of hobby, student and small-team usage that currently lives on Anthropic and OpenAI's coding tools.