Frequent coding and repository iteration
Test GLM Coding Plan
Subscription-style coding capacity
GLM coding plan
GLM Coding Plan: should you use it for your coding workload?
GLM Coding Plan is a subscription route for supported AI coding tools, not general API balance. It can fit frequent coding and execution work, but that does not mean replacing Claude everywhere. Modroam evaluates it by routing real tasks, not by publishing a universal model ranking.
Quick decision
Route the workload before choosing the plan
These are conditional test routes, not claims that GLM is always cheaper or stronger than Claude.
Architecture, design or final review
Keep a trusted primary model
The task is quality-sensitive
Bulk edits or repeated execution
Route to GLM after a small-task test
The workload is cost-sensitive
Non-coding API workload
Use the general API
Coding Plan has a supported-tool boundary
What the plan is
Coding subscription, with a firm route boundary
The GLM Coding Plan packages usage for supported coding environments. It uses a dedicated Coding endpoint and does not turn subscription credits into general API balance.
Z.AI currently documents Claude Code, Cline, OpenCode and OpenClaw among its supported tools. The supported list is broader and can change, so verify the live tool-integration page before connecting another client.
Freshness boundary · Checked August 11, 2026
Use the live subscription page for current commercial tiers and model positioning. Use the integration documentation for endpoints and setup. When those pages update at different speeds, do not assume an older model label changes the current commercial offer.
Current plans and limits
Three tiers, each with 5-hour and weekly credits
Checked August 11, 2026. Prices are regular monthly list prices shown by Z.AI; term discounts or promotions shown at checkout are separate from these figures.
Lite
$18/month
- 5-hour credits
- 2,000
- Weekly credits
- 10,000
- Recommended concurrency
- One project at a time
Pro
$72/month
- 5-hour credits
- 12,000
- Weekly credits
- 60,000
- Recommended concurrency
- 1–2 simultaneous projects
Max
$160/month
- 5-hour credits
- 28,000
- Weekly credits
- 140,000
- Recommended concurrency
- 2+ simultaneous projects
Credits are usage units, not prompt counts or general API dollars. Z.AI says 5-hour credits refresh dynamically five hours after consumption, weekly credits reset every seven days, and concurrency can change with resource availability.
Coding Plan vs general API
Do not send every workload through the subscription route
Billing model
Coding Plan: Recurring subscription
General API: Pay for metered API usage
Allowed usage
Coding Plan: Officially supported coding tools and product environments
General API: Programmatic workloads allowed by API terms
Endpoint
Coding Plan: Dedicated Coding or Anthropic-compatible endpoint
General API: General API endpoint
Best fit
Coding Plan: Frequent interactive coding and repository work
General API: Backend, batch and custom automation
Choose API when
Coding Plan: The task is outside the plan policy
General API: You need raw usage billing or server-side control
Using it with Claude Code
Keep the client; point it at the plan
Claude Code remains the client, while Z.AI supplies the API key, Anthropic-compatible endpoint and backend GLM mapping. A Claude label in the interface does not prove an Anthropic model is running.
{
"env": {
"ANTHROPIC_AUTH_TOKEN": "your_zai_api_key",
"ANTHROPIC_BASE_URL": "https://api.z.ai/api/anthropic"
}
}The official one-command shell installer is for macOS/Linux, not Windows. Z.AI currently documents the same manual Claude settings file across macOS, Linux and Windows; Windows also needs Git for Windows for Claude Code. Model mappings can update, so avoid hard-coding an old label unless your test requires it.
Follow the current official Claude Code setupWorkload routing
Do not migrate everything at once
This is a test framework, not a Modroam benchmark result. Validate quality, corrections, quota and total task cost on your own repository.
Architecture and planning
Keep a high-trust primary model
Use judgment quality as the first filter
Implementation and repetitive coding
GLM candidate
Start with one bounded repository task
Bulk edits and mechanical execution
GLM candidate
Measure correction rate as well as spend
Background agent loops
Test before routing
Validate reliability and quota behavior first
Final review and production-critical changes
High-trust review layer
Do not remove independent review
Who should skip it
A coding subscription is not the default for every user
- You only write a few lines of code occasionally.
- Most of your workload is not coding.
- You need general API automation outside the Coding Plan policy.
- You have not yet tested GLM on a representative repository task.
Access request
Need GLM access?
Tell us whether you need platform access, an API route, a coding plan or setup support. We’ll check availability, scope and a test path.