Z.ai's GLM Coding Plan is $18, $80 or $168 a month: how much coding does each tier buy?

GLM Coding Plan is Z.ai's subscription for running its GLM models inside coding agents such as Claude Code, Codex, Cursor, OpenCode and Cline. You point the tool at Z.ai's endpoint, pay a flat price, and your work is counted against a credit allowance instead of being billed per token. Every tier gets the same two models, GLM-5.3 and the lighter GLM-5.3-Flash.

So the tiers differ in how much you can use, not in what you can use. Pro gives you six times Lite's credits for about four and a half times the price; Max gives fourteen times for a little over nine.

Three tiers

Scroll sideways to see every column.

The plan page opens on yearly billing, which is 30% off. There is also a quarterly option at 20% off ($14.40, $64 and $134.40 a month). Pro adds faster generation and a curated set of MCP tools, and Max gets dedicated resources during peak times.

Two allowances, running side by side

Credits used inside a 5-hour window come back five hours later. The weekly count starts on the day you subscribe and resets every seven days. On every tier the weekly figure is exactly five 5-hour windows, so a few long sessions in a row can take a big bite out of the week. When you hit either limit, the tool waits for the reset. Nothing is charged to your account balance.

What a credit buys

Each model has a multiplier for input, cached input and output, and credits = tokens × multiplier ÷ 10,000. For GLM-5.3 the multipliers are 6.9, 1.7 and 24, so a million output tokens costs 2,400 credits. GLM-5.3-Flash costs about a third as much (2.3, 0.56 and 8).

That sounds tight next to Lite's 2,000 credits per window, but a coding agent mostly resends the same files and conversation, and that is billed as cached input. Z.ai's own estimate, at a 96% cache hit rate, is 50 to 99 million GLM-5.3 tokens a week on Lite, 297 to 595 million on Pro and 694 to 1,387 million on Max. On Flash the figures are about three times higher.

When you work changes the count

The low end of each range assumes all your work falls in peak hours, and the high end assumes none of it does, because off-peak use is counted at half rate. Peak hours are Monday to Friday, 14:00 to 18:00 Singapore time, which is 06:00 to 10:00 UTC. On the US East Coast that is 2 to 6 a.m. during daylight saving time, and on the West Coast 11 p.m. to 3 a.m. An ordinary American working day never touches it. In the UK and the rest of Europe the window covers the morning, so a long agent run started after lunch goes twice as far.

Yearly, quarterly or monthly

Plans renew automatically. To stop, cancel on the Subscription page at least three days before the next billing date, and the plan runs until the end of the period. Paid periods are not refunded, even if you have not used them up, so Z.ai recommends picking the term that matches how much you will use it.

Starting on monthly costs you very little flexibility. If you switch the same tier from monthly to yearly, the year is added after the month you already paid for (Z.ai's example: Lite monthly, then Lite yearly, gives 13 months). Moving up a tier takes effect at once, and what is left of the old plan goes to your account balance on a pro-rated basis.

Where the plan works

Only in the coding tools Z.ai supports, set up with its plan endpoint. If you want to call GLM from your own app or scripts, that goes through the pay-as-you-go API, where GLM-5.3 is $1.40 per million input tokens, $0.26 cached and $4.40 output (API pricing). A subscription is for one person. For a team there is a separate Team plan with central seat management and a usage dashboard.

Z.ai says it generally provides its services from Singapore and processes personal data there. Its data processing addendum for the API says the text you send is processed in real time and not saved on its servers.

Which tier

If you don't use a coding agent, or you want GLM inside your own product, the API fits better than the plan. If you work on one project for a few hours a day, Lite is enough, especially if most of your hours fall outside the peak window. If the agent runs most of the day or you keep two projects going, take Pro. Max is for several projects at once, or for running subagents in parallel. Not sure? Start Lite on monthly billing, check the weekly usage on the subscription page after a week or two, and then move to yearly or up a tier.

Sources: Z.ai GLM Coding Plan page (prices, billing periods, what each tier adds, supported tools, cancelling), plan overview (credit allowances, multipliers, peak hours, token estimates), usage policy (projects at once, renewal, refunds, one person per subscription), FAQ (what happens when credits run out, upgrades, use outside supported tools), API pricing and privacy policy (where data is processed, data processing addendum), checked October 9, 2026.

← Back to the comparison