
TL;DR
ZCode is Z.ai's free desktop agentic development environment built around GLM-5.2. Here is the developer setup, pricing breakdown, and how it compares to Claude Code and Cursor.
| Resource | Link |
|---|---|
| ZCode Documentation | zcode.z.ai/en/docs/welcome |
| GLM Coding Plan Pricing | z.ai/subscribe |
| GLM-5 GitHub Repository | github.com/zai-org/GLM-5 |
| GLM-5.2 on Hugging Face | huggingface.co/THUDM/GLM-5.2 |
| Z.ai API Documentation | platform.z.ai/docs |
Last updated: July 7, 2026
ZCode launched publicly the week of July 1, 2026, positioning itself as the official harness for GLM-5.2 - Zhipu AI's open-weights coding model that scores 62.1% on SWE-bench Pro. The app is free. You pay for the AI models you connect.
Z.ai calls ZCode an "Agentic Development Environment" - an ADE rather than an IDE. Where a Cursor or VS Code fork puts the editor first and bolts an agent onto it, ZCode puts the agent conversation at the center and arranges everything the agent touches around it: a file manager, a terminal, a Git panel, and a live browser preview, all in one Electron app.
ZCode is a desktop application that bundles:
The design philosophy is that the agent maintains context across files, terminal output, browser state, execution modes, and Git state simultaneously - reducing mid-task context breaks that plague other tools.
Download the desktop app for your platform:
After installation:
New users get a 5-day free trial: 3M GLM-5.2 tokens/day plus 2M GLM-5-turbo tokens/day (5M total daily).
ZCode itself is free. The models cost money. Z.ai offers the GLM Coding Plan as a flat-fee subscription:
| Plan | Monthly | Annual | Prompts/5hr | Prompts/Week | MCP Calls/Month |
|---|---|---|---|---|---|
| Lite | $18 | $151.20 | ~80 | ~400 | 100 |
| Pro | $72 | $604.80 | ~400 | ~2,000 | 1,000 |
| Max | $160 | $1,344 | ~1,600 | ~8,000 | 4,000 |
Through September 2026, there is a 30% promo that drops Lite to $12.60/month, Pro to $50.40/month, and Max to $112/month.
The GLM Coding Plan works with ZCode and 20+ other clients: Claude Code, Cline, Roo Code, OpenClaw, and others that support custom model providers.
GLM-5.2 and GLM-5-turbo normally consume quota at:
A limited-time promotion through September 2026 drops off-peak to 1x consumption.
During ZCode's campaign (through July 31, 2026), GLM-5.2 usage via the Coding Plan is metered at a 0.67 factor - effectively about 1.5x the usable quota.
If you prefer API pricing over subscriptions:
| Model | Input (per MTok) | Output (per MTok) |
|---|---|---|
| GLM-5.2 | $1.40 | $4.40 |
| GLM-5-turbo | $0.28 | $0.84 |
These are competitive rates - GLM-5.2 output is roughly 10x cheaper than Claude Fable 5.
Newsletter
Get the weekly deep dive
Tutorials on Claude Code, AI agents, and dev tools, delivered free every week.
From the archive
Jul 6, 2026 • 8 min read
Jul 6, 2026 • 8 min read
Jul 6, 2026 • 7 min read
Jul 6, 2026 • 5 min read
Set a verifiable session objective with /goal - the agent keeps iterating until the goal is verified complete. This is the "agentic" part of the agentic development environment.
/goal "Add user authentication with email/password and OAuth to the Next.js app"
The agent will plan, implement, test, and iterate until the goal is done or it hits a blocker.
Subagents are stored as plain Markdown files at ~/.zcode/agents/. They can be invoked:
@name in chatThis is similar to Claude Code's skills system but with automatic routing.
ZCode lets you modify prior messages without restarting tasks. If the agent went in the wrong direction, you can edit your original prompt and resume from there without losing context.
SSH and Docker container support enables agent operations in target environments. You can point ZCode at a remote machine or container and the agent executes there - useful for testing in production-like environments or working with large codebases you do not want to clone locally.
ZCode supports remote control via:
You can start a long-running task from your desk, then check on it from your phone.
The model powering ZCode scores:
GLM-5.2 is a 744B parameter mixture-of-experts model with 40B active parameters. It uses IndexShare architecture that reduces per-token computation by 2.9x at 1M context - making long-horizon tasks more efficient.
The model is Apache-2.0 licensed with no regional restrictions.
| Feature | ZCode | Claude Code | Cursor |
|---|---|---|---|
| Primary Model | GLM-5.2 | Claude models | Multiple |
| App Type | Standalone Electron | CLI + Extensions | VS Code Fork |
| Goal Mode | Yes | Via skills | Via Composer |
| Custom Subagents | Yes | Yes | No |
| MCP Support | Yes | Yes | Limited |
| Mobile App | Yes | No | iOS Beta |
| Open-Source Model | Yes (Apache-2.0) | No | No |
| Monthly Cost | $18-160 | $20 (subscription) | $20 |
| Edit History | Yes | No | No |
ZCode's unique advantage is the combination of goal-directed execution with an open-weights model at competitive pricing. The disadvantage is that GLM-5.2, while strong, is not quite at Claude Opus 4.8 levels on the hardest tasks.
ZCode runs on Z.ai's infrastructure, which operates under Chinese data law. Every GLM-5.2 API call routes through servers subject to PRC jurisdiction. For most development work this is fine. For code involving regulated data, sensitive IP, or compliance requirements, consider whether this matters for your use case.
This is not unique to ZCode - it applies to any tool using GLM models via Z.ai's API.
Good fit:
Not the best fit:
/goal to set your first objectiveIf the trial works for your use case, the Lite plan at $18/month (or $12.60/month with the current promo) is the next step.
The app is free. The AI models cost money. New users get a 5-day free trial with 5M tokens/day. After that, you need a GLM Coding Plan ($18-160/month) or pay-as-you-go API access.
ZCode is designed as the official harness for GLM models. It does not support Claude, GPT, or other providers. If you want multi-model support, look at Claude Code or Cursor.
GLM-5.2 scores 62.1% on SWE-bench Pro vs Sonnet 5's 63.2%. They are in the same ballpark. GLM-5.2 is open-weights (Apache-2.0) and cheaper at $1.40/$4.40 per MTok vs Sonnet 5's $2/$10 introductory rate.
ZCode lets you modify prior messages in a conversation without starting over. If the agent went down a wrong path, you can edit your original prompt and the agent continues from there with full context.
No. ZCode requires internet access to call the GLM API. There is no local model option within ZCode itself, though GLM-5.2 can be self-hosted separately via vLLM or SGLang.
Goal Mode sets a verifiable objective for the session. The agent keeps iterating - planning, implementing, testing, fixing - until the goal is complete or it hits a blocker that requires human input.
Yes. The GLM Coding Plan works with 20+ clients including Claude Code, Cline, Roo Code, and others that support custom model providers. You are not locked to ZCode.
Z.ai operates under Chinese data law. All API calls route through PRC-jurisdiction servers. This matters for regulated industries and sensitive code. It does not matter for most development work.
Read next
Z.ai shipped GLM-5.2 in mid-June with a usable 1M-token context window, two thinking-effort levels, and MIT open weights now released. Here is the setup guide for Claude Code, pricing breakdown, and what to test before the benchmarks arrive.
8 min readTerminal agent, IDE agent, local-plus-cloud agent. Three architectures compared - how to decide which fits your workflow, or why you should use all three.
8 min readFrom terminal agents to cloud IDEs - these are the AI coding tools worth using for TypeScript development in 2026.
8 min readTechnical content at the intersection of AI and development. Building with AI agents, Claude Code, and modern dev tools - then showing you exactly how it works.
Anthropic's agentic coding CLI. Runs in your terminal, edits files autonomously, spawns sub-agents, and maintains memory...
View ToolThe original AI coding assistant. 77M+ developers. Inline completions in VS Code and JetBrains. Copilot Workspace genera...
View ToolAI-powered terminal assistant from AWS. Natural language chat, command autocompletion, code generation. Agentic mode rea...
View ToolOpen-source AI agent built in Rust, now governed by the Agentic AI Foundation at the Linux Foundation. Desktop app, CLI,...
View ToolConfigure Claude Code for maximum productivity -- CLAUDE.md, sub-agents, MCP servers, and autonomous workflows.
AI AgentsStep-by-step guide to building an MCP server in TypeScript - from project setup to tool definitions, resource handling, testing, and deployment.
AI AgentsSet up Codex Chronicle on macOS, manage permissions, and understand privacy, security, and troubleshooting.
Getting Started
Check out Zed here! https://zed.dev In this video, we dive into Zed, a robust open source code editor that has recently introduced the Agent Client Protocol. This new open standard allows...

In this episode, we explore the newly released GPT-5 Codex by OpenAI, a specialized version of GPT-5 designed for agentic coding tasks. Codex offers advanced features, including enhanced code...

To try everything Brilliant has to offer—free—for a full 30 days, visit https://brilliant.org/DevelopersDigest/ . You’ll also get 20% off an annual premium subscription. Exploring Windsurf...

Z.ai shipped GLM-5.2 in mid-June with a usable 1M-token context window, two thinking-effort levels, and MIT open weights...

Terminal agent, IDE agent, local-plus-cloud agent. Three architectures compared - how to decide which fits your workflow...

From terminal agents to cloud IDEs - these are the AI coding tools worth using for TypeScript development in 2026.

Complete pricing breakdown for every major AI coding tool. Claude Code, Cursor, Copilot, Windsurf, Codex, Augment, and m...

A controlled study of 660 Claude Code trials shows clean codebases reduce token usage by 7-8% and file revisitations by...

A new SonarSource study finds clean code doesn't boost agent pass rates - but it cuts token usage by 8% and file revisit...

New tutorials, open-source projects, and deep dives on coding agents - delivered weekly.