Quick answer: The GLM 5.3 Coding Plan has three individual tiers. Lite is $18 per month with 2,000 five-hour credits and 10,000 weekly credits; Pro is $80 per month with 12,000 and 60,000 credits; Max is $168 per month with 28,000 and 140,000 credits. The current subscription page shows annual-effective monthly prices of $12.60, $56, and $117.60 respectively.
Every tier supports GLM-5.3, GLM-5-Turbo, and GLM-4.7. The right tier is therefore not determined by access to GLM 5.3; it is determined by how many weighted credits your busiest five-hour period and your full seven-day cycle consume.
This article was rebuilt from Z.AI's current Coding Plan documentation and subscription page on August 26, 2026. It replaces the earlier launch-period claim that the GLM 5.3 API and pricing were still unpublished and focuses this URL on subscription, credits, limits, supported tools, and tier selection.
GLM 5.3 Coding Plan prices and credit limits
| Plan | Monthly price | Annual-effective price | Five-hour credits | Weekly credits |
|---|---|---|---|---|
| Lite | $18/month | $12.60/month | 2,000 | 10,000 |
| Pro | $80/month | $56/month | 12,000 | 60,000 |
| Max | $168/month | $117.60/month | 28,000 | 140,000 |
The lower figures are effective monthly prices when paying annually, not month-to-month prices. The current subscription page positions Lite for lighter work on small repositories, Pro for more frequent work on mid-sized repositories, and Max for the heaviest individual usage.
All tiers support the same three core models listed in the current plan documentation:
GLM-5.3GLM-5-TurboGLM-4.7
Requests made to the older GLM-5.2 or GLM-5.1 model names are automatically routed to GLM-5.3. That matters when estimating usage: selecting an older flagship name is not a way to keep the old model or avoid GLM 5.3's credit multipliers.
How the five-hour and weekly limits work
Each plan is governed by both limits.
Five-hour credits control bursts
The five-hour allowance is dynamically refreshed. Z.AI says consumed credit quota resets five hours after consumption. It is not a fixed block such as 09:00-14:00 that resets for everyone at the same time.
This is the number to use when evaluating a long refactor, an agent session, or several large tasks run close together. A plan can have plenty of weekly capacity left and still pause heavy usage because the five-hour allowance has been consumed.
Weekly credits control sustained use
The weekly allowance is activated when the subscription starts and resets every seven days. It limits total usage across the cycle even when no individual five-hour period is unusually large.
A practical plan comparison therefore needs two estimates:
- Credits used during the busiest five-hour period.
- Credits used during the full seven-day cycle.
A tier is too small when either estimate regularly approaches its limit. Looking only at the weekly number can hide burst constraints; looking only at the five-hour number can hide sustained daily use.
How GLM 5.3 credits are calculated
Coding Plan credits are weighted token usage, not raw token counts. Z.AI publishes this formula:
credits = (
input tokens × input multiplier
+ cached input tokens × cached-input multiplier
+ output tokens × output multiplier
) / 10,000
The current multipliers are:
| Model | Input multiplier | Cached-input multiplier | Output multiplier |
|---|---|---|---|
| GLM-5.3 | 6.9 | 1.7 | 24 |
| GLM-5-Turbo | 5.7 | 1.5 | 21 |
| GLM-4.7 | 4.6 | 1.2 | 16 |
GLM 5.3 therefore consumes more plan credits per token than the other listed models. Output has the largest multiplier, while cached input is considerably cheaper than fresh input. The result is that two sessions with the same total token count can consume very different credits depending on output length and cache hits.
Worked credit example: a normal repository session
Suppose a session uses 20,000 fresh input tokens, 80,000 cached input tokens, and 4,000 output tokens:
(20,000 × 6.9 + 80,000 × 1.7 + 4,000 × 24) / 10,000
= (138,000 + 136,000 + 96,000) / 10,000
= 37 credits
That session consumes 37 credits during peak hours or 18.5 credits during off-peak hours.
Worked credit example: a large 1M-context task
Now consider 100,000 fresh input tokens, 900,000 cached input tokens, and 20,000 output tokens:
(100,000 × 6.9 + 900,000 × 1.7 + 20,000 × 24) / 10,000
= (690,000 + 1,530,000 + 480,000) / 10,000
= 270 credits
That task consumes 270 credits at peak or 135 credits off-peak.
These examples are planning models, not promises about a particular prompt. Record the fresh input, cached input, and output reported by your own sessions, then use the same formula to estimate the tier you need.
You can try GLM 5.3 on glm5.app with a realistic repository task before committing to a subscription, but use Z.AI's own usage statistics when measuring Coding Plan credits.
The Credit-Limit Mistake That Makes a Plan Feel Smaller
The most common mistake is reading “10,000 weekly credits” as “10,000 tokens” or “10,000 prompts.” It is neither. Credits are produced by weighted token usage, and each prompt may trigger multiple model calls inside an agent workflow.
A second mistake is comparing plans only by weekly allowance. Lite has 10,000 weekly credits, but a large burst is still bounded by its 2,000-credit five-hour limit. Pro and Max raise both limits, which is why the best tier depends on the shape of the workload rather than just the total number of coding hours.
A third mistake is ignoring the cache. With GLM 5.3, fresh input has a 6.9 multiplier while cached input has a 1.7 multiplier. Restarting sessions, constantly changing long prompt prefixes, or repeatedly rebuilding context can increase credit use. The plan will appear smaller when a workload has poor cache reuse and long outputs.
The fix is to calculate one or two representative tasks, then model both limits. Do not choose a tier from a generic “prompts per month” estimate when the official credit formula is available.
Peak and off-peak hours
Z.AI charges model usage at 50% of the standard credit rate during off-peak hours.
Peak hours are:
- Monday to Friday
- 14:00-18:00 Singapore Standard Time
- UTC+8
That is 06:00-10:00 UTC on weekdays. All other times, including the entire weekend, are off-peak under the current rule.
Convert 06:00-10:00 UTC into your own local time and account for daylight-saving changes where applicable. Avoid memorizing a local conversion that may shift seasonally. For large non-urgent refactors or batch agent work, moving execution outside the peak window halves the plan credits consumed by model usage.
This discount belongs to the Coding Plan credit system. It does not reduce the public GLM 5.3 API's per-token prices. The distinction is explained in the separate GLM 5.3 API pricing guide.
What Other Coding Plan Guides Leave Out
Many plan comparisons stop at the three prices. That misses the details that determine whether a tier actually fits:
- There are two independent limits to plan around. A workload must fit both its busiest five hours and the seven-day cycle.
- Credits are weighted. GLM 5.3 input, cached input, and output use multipliers of 6.9, 1.7, and 24.
- Off-peak applies after the standard credit calculation. It cuts model-credit consumption in half; it does not change the multipliers or API token prices.
- Model choice changes credit burn. GLM-5-Turbo and GLM-4.7 have lower published multipliers than GLM-5.3.
- Old flagship aliases route forward. Requests for GLM-5.2 and GLM-5.1 are routed to GLM-5.3.
- The plan is restricted to supported coding tools and products. It should not be treated as a cheap general-purpose API key for an unrelated application.
- Annual-effective prices require annual payment. They should not be compared directly with a flexible month-to-month commitment without considering the longer term.
The useful comparison is therefore not “Which plan has the lowest price?” It is “Which plan contains my peak five-hour task and my weekly workload, after realistic cache and off-peak assumptions?”
Which tier should you choose?
Choose Lite when both limits fit comfortably
Lite provides 2,000 credits per five-hour window and 10,000 per week. It can fit focused individual work when sessions are relatively contained, context is reused, and GLM 5.3 is reserved for tasks that need it.
Use the formula on your largest recent task. Lite is a reasonable starting point only when that task leaves enough five-hour headroom for follow-up calls and the full week remains well below 10,000 credits.
Choose Pro for larger bursts or sustained daily use
Pro raises the five-hour limit to 12,000 credits and the weekly limit to 60,000. It is the middle tier for people whose normal work regularly exceeds Lite rather than merely reaching it once.
The current subscription page also advertises faster generation, priority access to newer features, and a curated selection of MCP tools for Pro. Treat those as product benefits, while using the published credit limits—not marketing labels—to determine capacity.
Choose Max when Pro's measured limits are the bottleneck
Max provides 28,000 five-hour credits and 140,000 weekly credits. It is appropriate only when measured workloads need that additional headroom or when Max-specific service benefits matter.
Do not choose Max merely because a repository is large. A large repository with strong cache reuse and moderate output may consume fewer credits than a smaller project running repeated long agent loops. Usage shape is the better decision variable.
Monthly or annual billing?
The annual-effective prices shown on the subscription page are 30% below the listed monthly prices. Annual billing can make sense after a workload is stable and the tier has been validated. Monthly billing preserves flexibility while you learn actual credit use or while plan terms and models are changing quickly.
Supported tools, models, and endpoints
Z.AI's current Quick Start lists supported tools including Claude Code, Roo Code, Kilo Code, Cline, OpenCode, OpenClaw, Crush, Goose, Cursor, and other documented integrations. It says the plan is strictly limited to officially supported tools and products.
The Coding Plan supports these protocol base URLs:
| Protocol | Coding Plan base URL |
|---|---|
| Anthropic Messages | https://api.z.ai/api/anthropic |
| OpenAI Chat Completions | https://api.z.ai/api/coding/paas/v4 |
| OpenAI Responses | https://api.z.ai/api/v1 |
For Claude Code, Z.AI documents the glm-5.3[1m] alias for enabling the 1M context together with CLAUDE_CODE_AUTO_COMPACT_WINDOW set to 1000000. The /effort command can switch reasoning intensity, with max as the documented default.
The current plan overview also states that all plans support Vision Understanding, Web Search MCP, Web Reader MCP, and Zread MCP. Exact allowances and additional tier benefits can change, so confirm the current checkout and account pages before buying based on a specific tool entitlement.
For this site's own integration examples, see the glm5.app API documentation. The glm5.app service and Z.AI Coding Plan are separate commercial products with separate balances and terms.
Coding Plan vs pay-as-you-go API
Use the Coding Plan when a supported coding tool is acting as your development agent. Use the public API when an application, backend, customer-facing feature, or unsupported automation calls GLM 5.3.
A Coding Plan subscription does not turn its credits into public API balance. Conversely, public API spend does not increase the plan's five-hour or weekly allowance. Teams can use both, but the costs should be tracked separately.
Frequently asked questions
Is GLM 5.3 included in Lite, Pro, and Max?
Yes. The current official overview says all plans support GLM-5.3, GLM-5-Turbo, and GLM-4.7.
What are the current GLM Coding Plan prices?
The individual subscription page lists $18, $80, and $168 per month for Lite, Pro, and Max. It displays annual-effective monthly prices of $12.60, $56, and $117.60.
Do five-hour credits reset at a fixed time?
No. The five-hour allowance is dynamically refreshed, with consumed quota resetting five hours after consumption.
When does the weekly allowance reset?
Z.AI says it begins when the subscription is activated and resets every seven days.
How much does one prompt cost in credits?
There is no fixed amount. It depends on fresh input, cached input, output, the selected model's multipliers, tool calls, and whether model usage occurs during off-peak hours.
Does off-peak usage double my plan?
It halves model-credit consumption during qualifying hours, so the same published allowance can cover up to twice the modeled model usage when all of it occurs off-peak. Actual capacity still depends on token mix, cache hits, output, and tool usage.
Can I use a Coding Plan key in my own production app?
The official Quick Start says the plan is strictly limited to supported tools and products. Use the general pay-as-you-go API for other applications.
Which tier is best?
The smallest tier that safely contains both your busiest five-hour period and your seven-day usage is the rational choice. Measure first, then upgrade when a documented limit—not a guess—becomes the constraint.
Bottom line
Lite, Pro, and Max all provide GLM 5.3 access. What changes is capacity: 2,000/10,000, 12,000/60,000, or 28,000/140,000 five-hour/weekly credits. Apply the 6.9/1.7/24 GLM 5.3 multipliers to real token usage, account for the 50% off-peak rule, and choose the smallest tier that leaves operational headroom in both windows.
Test a representative task in GLM 5.3 on glm5.app, open the browser chat, and compare the result with the official Coding Plan usage statistics before locking into a longer billing period.
Sources
Facts and plan details were checked against these official Z.AI pages on August 26, 2026:
- GLM Coding Plan overview — Supported models, credit limits, reset rules, multipliers, MCP support, and off-peak treatment.
- GLM Coding Plan subscription page — Current individual-plan prices and tier benefits.
- GLM Coding Plan Quick Start — Supported tools, product-use restriction, protocols, and endpoints.
- How to switch Coding Plan models — GLM 5.3 configuration, 1M context alias, and effort controls.
- GLM 5.3 model guide — Current model availability, context, maximum output, and reasoning behavior.
