GLM Coding Plan: Prices, Credits and Limits

Z.ai's GLM Coding Plan has three individual tiers: Lite at $18 a month, Pro at $80 and Max at $168. Quarterly billing is 20% off and yearly 30% off. Limits are credits per five hours and per week, and team seats start at $88. Prices checked on 31 October 2026 from Z.ai's subscribe page and docs.

Prices checked on 31 October 2026

Every figure comes from z.ai/subscribe and docs.z.ai. The subscribe page opens on the yearly view, so we switched it to monthly, quarterly and yearly, and to the Team tab. Prices are in US dollars.

Billed totals work out as follows, by simple multiplication of the page prices: Lite is $43.20 a quarter or $151.20 a year, Pro is $192 a quarter or $672 a year, and Max is $403.20 a quarter or $1,411.20 a year. The page shows the per-month figure and does not print the charge in the same place, so confirm the total at checkout.

For other coding tools in the same price band, see the AI coding agent pricing comparison. The model behind the plan is listed at GLM 5.3 and GLM 5.3 Flash.

What the plan is

The GLM Coding Plan is a subscription for using Z.ai's GLM models inside coding agents, not a separate editor. You subscribe, take an API key and plug it into a supported tool. Z.ai's page lists ZCode (its own IDE), AutoClaw, Claude Code, Codex, WorkBuddy, TraeWork, CodeBuddy, TraeCode, OpenCode, Cursor and OpenClaw, and says the plan supports 20+ agent tools. Its docs also name Cline.

Usage is metered in credits, not tokens or messages. Model calls and tool calls both spend credits, so it behaves like the shared usage limits on other subscriptions, with the multipliers published.

The plan is "strictly limited to use within officially supported tools and products", in the words of Z.ai's FAQ. The usage policy adds that use in unsupported tools may restrict benefits, and that benefits are for the subscriber only: account sharing and multi-user access are prohibited.

Lite, Pro and Max

All three tiers include GLM-5.3 and GLM-5.3-Flash, access to the latest models, support for 20+ agent tools, default data privacy and four MCP tools: Vision Understanding, Web Search, Web Reader and Zread.

• Lite is for lightweight iteration on a small repo. Z.ai calls it "rolling access" to the latest flagship models.
• Pro is 6x Lite usage and adds priority access to the latest models, a curated selection of MCP tools and faster generation speeds.
• Max is 14x Lite usage and adds first access to the latest models and dedicated resources during peak times.

The jump from Lite to Pro costs 4.4 times the price for 6 times the usage, and from Lite to Max it costs 9.3 times the price for 14 times the usage. Looked at as credits per dollar on monthly billing, Lite buys 555 weekly credits per dollar, Pro 750 and Max 833. The larger tiers are cheaper per credit, and the gap widens slightly on discounted billing.

How the limits work

There are two limits, both in credits.

• Five-hour credits. Lite 2,000, Pro 12,000, Max 28,000. Once used up, you wait for the next five-hour cycle. Z.ai's docs say these credits refresh five hours after consumption.
• Weekly credits. Lite 10,000, Pro 60,000, Max 140,000. They start when you subscribe and reset every seven days. The FAQ says the cycle counts from when you place your order.

A quota reset card restores weekly credits and also the five-hour quota, and the cycle restarts from the moment the card is used. Z.ai's pages do not say how many reset cards you get or how you earn them.

Z.ai says rate and concurrency limits depend on tier and are adjusted dynamically with available resources. It gives no numeric concurrency caps. Its recommendation is one project at a time on Lite, one or two on Pro and two or more on Max.

How credits are spent

Credit use is a formula. For model calls, credits equal input tokens times the input multiplier, plus cached input tokens times the cached multiplier, plus output tokens times the output multiplier, all divided by 10,000.

MCP tools cost credits per call. Web Search, Web Reader and Zread are 1.2 each, and Z.ai counts visual understanding under the GLM-5.3-Flash multipliers.

Peak and off-peak

Model usage in off-peak hours is charged at 50% of the standard credit rate. Peak hours are Monday to Friday, 14:00 to 18:00 Singapore Standard Time (UTC+8). Z.ai's token estimates give a low and high end for each plan, where the low end assumes all usage is in peak hours and the high end assumes all of it is off-peak.

Z.ai's estimated weekly allowance on GLM-5.3 at a 95% cache hit rate is 48 to 97 million tokens for Lite, 290 to 580 million for Pro and 676 to 1,352 million for Max. On GLM-5.3-Flash at the same rate it is 146 to 292 million, 877 to 1,755 million and 2,047 to 4,095 million. These are Z.ai's estimates, and they depend on how much of your input is cached.

Worked examples from the listed multipliers

These apply Z.ai's published formula to made-up token counts. They are arithmetic, not measured usage.

One agent session on GLM-5.3. Say 100,000 uncached input tokens, 500,000 cached input tokens and 50,000 output tokens. Credits are (100,000 x 6.9 + 500,000 x 1.7 + 50,000 x 24) divided by 10,000, which is (690,000 + 850,000 + 1,200,000) / 10,000, or 274 credits.

• Lite's five-hour limit of 2,000 covers 7 such sessions, and its weekly 10,000 covers about 36 at peak rates.
• Off-peak the same session is 137 credits, so Lite's week covers about 72.
• Pro's weekly 60,000 covers about 218 at peak.

The same session on Flash. (100,000 x 2.3 + 500,000 x 0.56 + 50,000 x 8) / 10,000 is (230,000 + 280,000 + 400,000) / 10,000, or 91 credits. That is about a third of the GLM-5.3 cost.

Plan against API rates. Z.ai's API list prices per million tokens are $1.40 input, $0.26 cached and $4.40 output for GLM-5.3. The same 274-credit session costs $0.14 + $0.13 + $0.22, or about $0.49 on the API. Lite's 10,000 weekly credits are about 36.5 of those sessions, or about $17.90 a week at API rates, against $18 a month for the plan. On this mix, the plan is far cheaper than the API if you use the weekly allowance. If you use a tenth of it, the API is cheaper.

Annual saving. Pro yearly is $56 a month, or $672 a year, against $960 on monthly billing. That saves $288, or 30%. Lite saves $64.80 a year. Max saves $604.80.

Team of three. Three Standard seats at $88 are $264 a month. On annual billing at $79.20 each, that is $237.60 a month, or $2,851.20 a year, and 198,000 weekly credits across the team in total. The credits are per seat, not pooled, per Z.ai's seat quota table.

Team plan

Team seats are billed per seat per month:

Z.ai's team docs add these terms:

• Minimum: two seats, no stated maximum. Mixing Standard and Premium seats is not supported, and upgrading Standard to Premium is not supported yet.
• Admin: the primary administrator does not hold a seat by default but can assign one to their own account.
• One seat per member. Seats cannot be shared.
• Adding and removing: seats can be added mid-subscription with fees prorated for the rest of the cycle. They cannot be reduced directly. A reduction needs a new purchase after the cycle ends.
• Quota exceeded: without overage the model is unavailable until the next reset. An admin can enable on-demand overage in advance, and the seat is then billed for actual overage. The seat returns to its included quota at the next reset.
• Overage price: a 10% discount on model API list price, described as a limited-time offer.
• Spending limits: per-member limits can be set.
• Team Plan Key: members use a dedicated key, separate from other platform API keys.

The Team plan page on z.ai lists a Team and an Individual plan as separate products, and its FAQ asks whether both can be active at once. We could not read the answer there. The docs list the primary administrator and seat answers above.

Billing, renewal and cancellation

• Auto-renewal. Subscriptions renew at the end of each billing cycle. Charges are applied in the order bonus balance, cash balance, then the linked payment method.
• Cancel. Z.ai's FAQ says to cancel at least 24 hours before your next billing date. The usage policy says at least 3 days. The two pages disagree, so cancel three days ahead to be safe. The current period stays valid after you cancel.
• Refunds. None. Both pages say subscriptions are non-refundable once purchased, even for unused time.
• Violations. Rule violations can lead to rate limiting, account freezing or restrictions. More than three violations may lead to a ban, and flagged accounts can appeal.
• Invites. The subscribe page offers an Invite Friends program that gives up to 20% back as credits, with no limit and no expiry. Z.ai does not say on the page what the 20% is a percentage of.

What changed recently

Z.ai's release notes show GLM-5.3 released on 18 August 2026 and GLM-5.3-Flash on 26 August 2026. Both are listed among the plan's models, and the release notes entry for GLM-5.3 reports a 50% gain over GLM-5.2 on Z.ai Code Bench, per Z.ai. The notes do not list a pricing or plan change in the entries we read.

Per Z.ai's docs, requests for GLM-5.2 or GLM-5.1 are now routed to GLM-5.3, and GLM-4.7 requests go to GLM-5.3-Flash. If you were choosing a model by name, the plan now chooses for you. Z.ai's pricing page does not give dates for the off-peak discount, so we cannot say when it started.

How it compares with peer tools

GLM is cheap on list price. Z.ai's API charges $1.40 input and $4.40 output per million tokens for GLM-5.3, and $0.15 and $0.50 for Flash. Our LLM API prices page sets those beside the other vendors.

On subscriptions, the entry tier is $18 a month for Lite on monthly billing, or $12.60 on yearly. Claude Pro is $17 a month annual ($20 monthly) and Cursor Plus is $8, per our AI coding agent pricing comparison. The big difference is what you get: the GLM plan gives GLM models only, while those plans give their own vendors' models. See Claude Code pricing, Claude Max plan and Cursor pricing for the details.

Agents that can run on the plan include Claude Code, OpenCode, Cline and Kilo Code per Z.ai's tool list (Kilo Code is named in its IDE recommendations). If you are weighing a router instead, see OpenRouter alternatives.

Our take

The GLM Coding Plan is a good fit if you already know you want GLM models and use an agent daily. The credit formula is published, the multipliers are clear and the plan is cheap against API rates for steady use. The Flash model costs a third as many credits in our example, which makes it the obvious choice for routine edits.

The trade-offs are practical. Credits are weekly and per five hours, so a heavy day can stall you until the next window. Peak hours are a Singapore afternoon, which is the European morning (06:00 to 10:00 UTC) and the US night, so US users mostly work off-peak while European users overlap the peak. Subscriptions do not refund, and the cancel deadline differs between two Z.ai pages.

Start on Lite. If you hit the five-hour limit several times a day, step up to Pro, which has the best credits-per-dollar jump relative to the price. We have not run the plan for this page, so this advice comes from the published numbers.

How to keep costs down

1. Use GLM-5.3-Flash for routine work. It costs about a third of the credits in our example.
2. Work off-peak. Model usage outside Monday to Friday 14:00 to 18:00 Singapore time is charged at half rate.
3. Keep context cached. Cached input is 1.7 against 6.9 for uncached on GLM-5.3, so stable context is cheaper.
4. Pay yearly if you will stay. It is 30% off, and the plan is non-refundable either way.
5. Switch off MCP tools you do not use. Each call costs 1.2 credits.
6. Do not run parallel agents on Lite. Z.ai recommends one project at a time.
7. Use only supported tools. Unsupported tools may restrict benefits.
8. Set team spending limits before enabling on-demand overage.

Checklist before you pay

• Check that your agent tool is on Z.ai's supported list.
• Pick a billing period and note the total charge, not the per-month price.
• Cancel at least three days before renewal if you do not want to continue.
• For teams, decide on seat type, because mixing is not supported.
• Set per-member limits if you enable overage.
• Note your weekly reset time from your order date.
• Do not share the account.

Figures we could not confirm

Z.ai does not publish a price for individual overage, since individual plans have no stated overage option, only quota resets and waiting. It does not say how reset cards are earned or how many you get. The cancel deadline appears as 24 hours in the FAQ and 3 days in the usage policy. The Team plan page asks whether Team and Individual plans can run together, but we could not read the answer. We found no numeric concurrency limits and no dated pricing changes.