Z.ai's ZCode chases Claude Code with GLM-5.2, and looks a lot like Codex
Z.ai, the Zhipu team behind the open-weight GLM models, shipped ZCode: a desktop coding agent wired to GLM-5.2. What several developers spotted is that the app is closed source and its UI reads as a near-copy of OpenAI's Codex.
Plans run from $16.20 to $144 a month, and GLM Coding subscribers get a quota discount inside ZCode: peak hours (14:00-18:00) drop from 3x to 2x and off-peak from 1x to 0.67x, roughly a 1.5x effective allowance. The base allowance itself is never stated in tokens.
Developers ask why a lab that open-weights its models keeps the harness closed, and a Chinese-owned agent with full system permissions drew explicit trust warnings.
What ZCode is: a GLM-5.2 desktop agent, not a command line
ZCode is a desktop application from Z.ai, the Zhipu team behind the open-weight GLM models, that wires GLM-5.2 into a plan, code, review and deploy loop. It runs on macOS, Windows and Linux, with Linux included from day one in beta. It is not a terminal tool: the interface is a GUI built around "Goals" the agent plans, executes and verifies, and it can be driven remotely through WeChat, Feishu or Telegram bots.
Three GLM Coding plans back it, and subscribers get a usage discount when they run GLM-5.2 inside ZCode:
| Plan | Price / month | Usage vs Lite |
|---|---|---|
| Lite | $16.20 | base allowance |
| Pro | $64.80 | 5x |
| Max | $144 | 20x |
The in-app discount rescales quota: peak hours (14:00-18:00) move from 3x to 2x, and the remaining twenty off-peak hours from 1x to 0.67x, roughly a 1.5x effective allowance. What that base allowance is in tokens is never published.
The thread's verdict: an exact copy, a moat, and a trust wall
The first thing developers noticed was the resemblance to a rival's app rather than to Claude Code.
"It's basically an exact copy of Codex." (a reader who compared the two interfaces)
The bigger question was why a company that publishes its model weights keeps the harness shut. A Rust developer tied it to the fight over model distillation.
"the harness is part of the the moat, it doesn't seem super surprising that the other side of that would try to also make it harder for them to tell how well they're doing" (a Rust developer)
Then came the objection that has nothing to do with code quality.
"I don't find a closed-source Chinese agent system trustworthy … you are just handing over your entire system to a Chinese-owned server." (a commenter citing China's National Intelligence Law)
Reliability drew its own complaints, with one user reporting the same failure on most prompts.
"Cannot connect to API: write EPIPE" (a user who hit the error on four of five prompts)
For all the skepticism, the plans keep selling out, which is its own signal.
"their coding plan is always selled out." (a developer in China)
Get breakdowns like this every day
The top Hacker News stories, unpacked — by email, Telegram, or RSS.
A frog-gut microbe wiped out tumors in every treated mouse; read the fine print
A single IV dose of Ewingella americana, a bacterium pulled from amphibian guts, cleared colorectal tumors in 100% of treated mice and beat both chemo and a checkpoint drug. The write-up that put it on Hacker News comes from an outlet the thread calls a crank blog.