ZCode GLM Desktop: Install, Plans and Review
Quick answer: On 4 July 2026 the recording opened Z.ai’s official ZCode desktop app for GLM 5.2, already installed on Debian, walked the free and paid coding plans shown that day, then started a local site review from the app. Check live pricing on the product page. The numbers below are only what appeared on camera.
This is the desktop-app walkthrough, not a generic model bake-off. For picking a model on your own task, use How to choose an AI model for a task. For Claude model families named in the same recording, see Claude Opus vs Sonnet vs Haiku vs Fable: which to use. Another Linux desktop path is How to use ChatGPT on Linux: web and desktop options.
What this recording shows
ZCode is the official code editor from Z.ai for GLM 5.2. The product URL in the description is https://zcode.z.ai/. The host called it one of the easier ways to start with that model, and said GLM 5.2 was giving Claude Fable 5 a run across benchmarks. That is how it was framed on 4 July 2026, not a current ranking.
Before this app, the recording says the easy path was OpenCode: pick GLM 5.2 on OpenCode Go or OpenCode Zen. That earlier route is covered in OpenCode Go: pricing, models, limits and setup. This page stays on the desktop app shown that day.
1. Install ZCode on Linux
The site on camera listed macOS, Windows and Linux, and said you can start for free. The host was already on Debian with the app installed, so this recording does not show an installer command, package name or checksum. Download from the official ZCode page for your OS, then confirm the publisher before you sign in.
The marketing site was interactive in the recording, so you could click around before installing. Once the app was open, the working folder was a local Tube Analytics project. No other install steps were spoken.
2. Plans shown on 4 July 2026
The coding-plan screen showed a free tier plus three paid coding plans. As read on camera that day:
- Free: a daily credit grant, later named as 5 million GLM tokens per day.
- Light: $16 per month.
- Pro: $65 per month.
- Max: $144 per month.
The host stayed on the free plan because he did not expect to use GLM 5.2 that often. The meter in the app showed about 28% of then-current credits used. Treat every figure in this section as a 4 July 2026 screenshot, not a quote you can bill against today. Recheck zcode.z.ai before you pay.
A separate OpenCode Go usage view in the same recording ranked DeepSeek V4 Flash first, DeepSeek V4 Pro second, then MiniMax, with GLM 5.2 fourth on that list. Token prices on that OpenCode view were $1.40 per 1 million input tokens and $4.40 per 1 million output tokens, described as about 10 times cheaper than Claude Fable 5 at that moment. Another on-screen chart showed 48 cents per 1 million tokens for GLM 5.2 against $2.75 for Claude Fable 5. Both numbers were on camera. This page does not pick a current rate.
3. Desktop tools and a local review
Inside the app the host asked GLM 5.2 to suggest new features for Tube Analytics. A changes panel listed edited files and could group unstaged and staged edits. A sidebar opened a terminal, then opened a file in Nano. A browser pane could load the local site. The recording presents those panes as part of this desktop app, not as a claim about every other editor.
Feature suggestions spawned sub-agents, including a brainstorming skill, while planning documents in the repo were pulled in. Later, after plugins were added, the host stopped that task, opened a new one, and asked for a full code review and site audit of Tube Analytics, with ask-before-changes turned on. The video does not read out the audit findings. Do not treat the review as a finished report.
4. Modes, skills and usage
Model mode on camera was regular GLM 5.2 or turbo. Thought level was no thinking, high or max, with high recommended in that session. Agent access was planning mode, edit files automatically, ask before changes, or full access.
Existing skills were pulled in already on. The app could add MCP servers and extra plugins. Superpowers was one plugin the host added, then several others from the discover list. After that, available commands included code review, commit push and feature development. Folder indexing could be turned on from the same UI.
Usage shown in the app was 867,000 tokens across six sessions, on GLM 5.2 without turbo. For other cost-tracking setups recorded on this site, see 9Router Cost Tracking: Combos and Fallbacks and AI Token Usage Dashboard.
Bridgebench was the named benchmark board. GLM 5.2 on max mode was shown at 51 for intelligence next to Claude Fable 5 at 60, with GLM called considerably faster and cheaper on that chart, and number four in speed. Reasoning, refactoring and UI were the Bridgebench columns called out. Those are the scores on screen that day, not independent proof. The host also said Fable 5 was chewing through credits faster than other models he had used, so he had dropped to Opus 4.8 for most coding and kept Fable 5 for planning.
Watch the walkthrough
FAQ
Which operating systems did the recording name?
The 4 July 2026 site view listed macOS, Windows and Linux. The desktop app on camera was already installed on Debian. The recording did not show the installer command.
What ZCode plans were shown on 4 July 2026?
A free plan with daily credits, named as 5 million GLM tokens per day, plus Light at $16 a month, Pro at $65 a month and Max at $144 a month. Those figures are what appeared on that date, not a current price list.
Did the recording prove GLM 5.2 beats Claude Fable 5?
No. Charts on camera, including Bridgebench, were shown as vendor or third-party scores from that day. This page repeats the numbers as they appeared, not as proof of a ranking.
What local review did the recording run?
After adding plugins, the host started a full code review and site audit of a local Tube Analytics folder, with ask-before-changes on. The video did not report the audit findings.
Sources
- Video: GLM 5.2 Now Has An Official Desktop App! (4 July 2026)
- Z.ai, ZCode product page: https://zcode.z.ai/
You may also like
Build the next one with AI Income Lab
Step-by-step courses, reusable templates, and a community that answers when you get stuck.
Members get the Claude Code & Skills and Codex Tutorials & Courses lessons.
- Courses for builds like this one
- Copy-ready templates
- Ask questions while you build
- Weekly coaching on VIP
