GLM Coding Plan: run GLM-5.3 inside Claude Code for less
A screenshot-verified walkthrough of z.ai's budget coding subscription: what the Lite, Pro and Max tiers cost in 2026, how to connect them to Claude Code with an API key and a few environment variables, where the usage limits bite, and whether the quality holds up.
TL;DR — the short version
- The GLM Coding Plan is z.ai's subscription that runs the GLM-5.3 and GLM-5.3-Flash models inside Claude Code and 20+ other coding tools. Plans start at 18 dollars a month, or 12.6 billed yearly.
- Setup takes about five minutes: subscribe, create a z.ai API key, then either run npx @z_ai/coding-helper or paste a small env block into ~/.claude/settings.json.
- Every plan pairs a 5-hour rolling window with a weekly credit budget — Lite ships 10,000 credits per week, Pro multiplies that by 6 and Max by 14. Off-peak requests cost half the credits.
- Verdict: a credible budget alternative to Claude's 100-200 dollar plans for everyday coding. Pair it with the free DeepSeek route on this site and most coding work needs no frontier subscription at all.
Claude Code for $10! (GLM Coding Plan)
Channel:Samuel Gregory12:49
GLM Coding Plan × Claude Code (official integration guide)
Official documentation:Z.AI Developer Documentationdocs.z.ai
Screenshots come from two screen-only reviews (credited below) plus the official z.ai pricing page and documentation, captured September 19, 2026. z.ai reprices this plan often — always confirm current tiers before subscribing.
Frame credits: "Using GLM-5.1 in Claude Code is Ridiculously EASY" by Agentic Hub and "GLM 4.6: The $3/Month Claude Code Killer? (Full Setup & Test)" by Eric Tech, both on YouTube and linked per step. Pricing and docs screenshots are from z.ai's official website.
From subscription to first GLM prompt, step by step
1 · Pick a plan and get your API key
- 1
Compare the Lite, Pro and Max tiers
Open z.ai's subscription page and pick a billing cycle: monthly, quarterly (20 percent off) or yearly (30 percent off). Lite targets light daily use with 10,000 weekly credits, Pro multiplies usage by six for mid-sized repos, and Max by fourteen for power users. The screenshot shows the official page as of September 2026.
Official GLM Coding Plan pricing tiers on z.ai — captured September 2026 - 2
Subscribe and confirm the plan is active
After checkout, the z.ai console lists your plan under API Management → Subscription with its auto-renew date and the tools it covers. The GLM Coding Plan is not pay-as-you-go API credit — this subscription card is what Claude Code will authenticate against.
The Subscription tab confirms the active GLM Coding plan and its renewal dateWatch at 3:38 - 3
Create a dedicated API key
Go to API Management → API Keys and click Create a new API key. Give it a recognizable name such as claude-code so its usage is easy to audit, then confirm. You must be subscribed to a plan for the key to carry the coding-quota benefits.
Name the key before confirming — one key per tool keeps usage auditableWatch at 3:46 - 4
Copy the key once, store it safely
The new key appears in the API keys table with its masked secret, creation time and last-used columns. Copy the full value now — you will paste it into Claude Code's settings in a moment, and the console only shows the masked form afterwards.
The API keys table after creation — copy the secret before closing the dialogWatch at 1:32
2 · Point Claude Code at the GLM endpoint
- 5
Find Claude Code's settings folder
Claude Code keeps its configuration in a .claude folder inside your user profile — C:\Users\<you>\.claude on Windows or ~/.claude on macOS and Linux. The file you need is settings.json; create it if it does not exist yet.
The .claude folder inside the user profile holds settings.jsonWatch at 0:24 - 6
Paste the GLM env block into settings.json
Add an env object with ANTHROPIC_BASE_URL set to https://api.z.ai/api/anthropic and ANTHROPIC_AUTH_TOKEN set to your z.ai key, plus API_TIMEOUT_MS of 3000000. This redirects Claude Code's Anthropic-protocol requests to z.ai, where your subscription quota is consumed instead of Anthropic credits.
The two required env vars: ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKENWatch at 0:40 - 7
Map Opus, Sonnet and Haiku to GLM models
Claude Code internally requests opus, sonnet and haiku slots. Map them with ANTHROPIC_DEFAULT_OPUS_MODEL, ANTHROPIC_DEFAULT_SONNET_MODEL and ANTHROPIC_DEFAULT_HAIKU_MODEL — the official docs currently suggest glm-5.3 for Opus and Sonnet and glm-5.3-flash[1m] for Haiku, with CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC set to 1.
Model-mapping env vars route Claude Code's internal model slots to GLMWatch at 2:50 - 8
Follow the official doc when model versions change
z.ai keeps a "Switching Models in Claude Code" guide that lists the exact model IDs of the day and the default configuration. When GLM ships a new flagship, you only update the three model-mapping values rather than redoing the whole setup.
z.ai's docs list the current model IDs for Claude Code's default configurationWatch at 1:00 - 9
Shortcut: let the Coding Tool Helper configure everything
Prefer not to hand-edit JSON? Run npx @z_ai/coding-helper (needs Node.js 18+). The wizard asks for your language, plan, API key and target tools, then writes the same settings and can register MCP servers for you. Claude Code, Codex, OpenCode, Crush and Factory Droid are supported.
Official npx @z_ai/coding-helper wizard — the automated alternative to manual editing - 10
Launch Claude Code against the GLM endpoint
Open a NEW terminal (old windows keep stale env), cd into your project, and start claude. For a one-off session you can also inject the variables inline: ANTHROPIC_BASE_URL=https://api.z.ai/api/anthropic ANTHROPIC_AUTH_TOKEN=your-key claude — nothing gets written to disk that way.
Inline env vars launch a one-off GLM session without touching settings.jsonWatch at 3:30
3 · First run, real tasks and usage checks
- 11
Confirm GLM is active on the welcome banner
Claude Code's start screen shows which model is in use and the billing mode — a GLM build with API Usage Billing means your env block won. If Claude Code asks "Do you want to use this API key", answer yes.
The banner names the active GLM build and the API Usage Billing modeWatch at 2:08 - 12
Send a cheap smoke-test prompt
Type hi and check that the reply arrives in Claude Code's usual formatting. Tools, slash commands and permission prompts behave exactly as on the official plan — only the model behind them has changed.
First response from GLM inside Claude Code — same interface, different engineWatch at 5:06 - 13
Run a real task end to end
Now give it real work — scaffold an interface, run /init on a repo, refactor a module. Long-horizon agentic coding is exactly what GLM-5.3 is tuned for, and weekly credits are what you are spending, so watch your first big task closely to gauge the burn rate.
A real request running on GLM — note the working indicator while it codesWatch at 2:24 - 14
Use plan mode before big builds
For larger features, let the model plan first: it produces a phased checklist (foundation, core components, interface, testing) that you can approve or trim before a single credit is spent on code. Planning passes are far cheaper than regenerating a wrong implementation.
GLM's phased implementation plan rendered inside Claude CodeWatch at 6:02 - 15
Know where usage shows up — and where it doesn't
Anthropic's own console shows zero usage because your requests never touch Anthropic billing. Quota lives in the z.ai console, metered in credits against your plan's weekly budget with a 5-hour rolling window on top. If a session dies mid-task, you probably hit the 5-hour wall, not a bug.
Anthropic Console stays empty on the GLM plan — quota is metered by z.aiWatch at 4:24
GLM Coding Plan tiers at a glance (September 2026)
| Plan | List price | Usage allowance | Best for |
|---|---|---|---|
| Lite | $18 / month ($12.6 on yearly) | 10,000 credits per week | Light daily use, learning, small repos |
| Pro | $80 / month ($56 on yearly) | 6x Lite usage, priority access to new models, faster generation | Day-to-day development on mid-sized repos |
| Max | $168 / month ($117.6 on yearly) | 14x Lite usage, dedicated resources at peak times | Advanced users on mid-to-large repos |
Quarterly billing saves 20 percent and yearly 30 percent versus list price. Every tier can run GLM-5.3 and GLM-5.3-Flash, and requests for older GLM versions are auto-routed to the newest one. Prices change often — verify on z.ai.