GLM Coding Plan Pricing in 2026: What $18, $80 and $168 Buy, and What Renewal Costs

If you’ve read three articles about this plan, you’ve probably seen three prices. Ten dollars a month. Thirty. Seventy-two for Pro. All three are still on page one of the search results. Only the last ever appeared on an official page we could check, and it’s out of date too.

The plan changed twice in 2026. In April, Z.ai retired its plans without weekly limits. On July 30 it replaced prompt counts with credits, and Pro and Max now cost $80 and $168 where the April notice listed $72 and $160. Most guides you’ll find were written before that. They describe GLM-5.2, prompt quotas and a product that is no longer sold.

So this guide does one thing. It reports what Z.ai’s own subscribe page, documentation and terms said on October 1, 2026, including four rules that are easy to miss until they cost you money.

How much does the GLM Coding Plan cost?

The GLM Coding Plan has three individual tiers. Billed monthly they cost $18, $80 and $168. Paying quarterly takes 20% off and paying yearly takes 30% off, which brings Lite down to $12.60 a month. The subscribe page shows only per-month equivalents, so the totals you’re charged up front are in the last two columns.

There is also a team plan at $88 or $188 per seat a month, with a two-seat minimum. We found no first-month promotional price on the page. The one documented discount for newcomers is 10% off a first order placed through a referral link.

What do you get on each GLM Coding Plan tier?

Every tier runs the same two models, GLM-5.3 and GLM-5.3-Flash, in the same 16 supported coding agents, which include Claude Code, Cursor, Cline, OpenCode and Codex. What you pay for is quota and priority. Each tier has a five-hour allowance and a weekly one, and Pro and Max are promised faster generation and resources at busy times.

Two things follow from that table. Bigger tiers are cheaper per credit: Pro gives six times Lite’s quota for 4.4 times the price. And on every tier the weekly allowance equals exactly five full five-hour windows. If you drain a window every working day, you finish the week’s credits on Friday afternoon.

How far does a credit go?

A credit is a weighted count of tokens. Z.ai’s formula multiplies input, cached-input and output tokens by a factor for the model, adds them up and divides by 10,000. On GLM-5.3 the factors are 6.9, 1.7 and 24. On GLM-5.3-Flash they are 2.3, 0.56 and 8, which makes Flash a third of the price for the same work.

In round numbers, by our arithmetic: on GLM-5.3 at the standard rate, a million fresh input tokens costs 690 credits, a million cached input tokens costs 170, and a million output tokens costs 2,400. Coding agents reread the same files constantly, so most of what they send is cached. Z.ai’s own estimate, assuming 95% of input is cached, is that Lite covers 48 to 97 million GLM-5.3 tokens a week, or 146 to 292 million on Flash.

The low end of each range is all peak-hour use and the high end is all off-peak. That spread is the single biggest lever you have, and it depends on where you live.

When are peak hours in your time zone?

Peak hours are Monday to Friday, 14:00 to 18:00 Singapore time. Outside them, model usage costs half the credits. For a developer in the United States that’s good news: the peak window falls in the middle of the night. In Europe it lands on the working morning.

Two temporary promotions were running when we checked, and both end on October 7, 2026: all-day off-peak rates, and a night-time campaign for GLM-5.3-Flash. Don’t size your plan on them.

Is the GLM Coding Plan cheaper than paying per token?

Yes, if you use most of your quota. Z.ai sells the same GLM-5.3 through its pay-as-you-go API at $1.40 per million input tokens, $0.26 cached and $4.40 output. Set against the credit formula, one credit is worth roughly 0.15 to 0.2 cents of API usage at the standard rate. Lite’s 43,000 or so credits a month would therefore cost about $65 to $90 through the API, and twice that if you work off-peak. You pay $18.

Z.ai’s headline claim is a saving of up to 92% against its API when all usage is off-peak. Our numbers land close to that if we also assume yearly billing. Both calculations assume you empty the quota every week. A plan you use at a quarter of its allowance is no bargain. If your usage is light or uneven, the per-token API has no weekly cap and no commitment.

What will you pay at renewal?

Whatever the price is on the day you’re charged. Z.ai’s subscription terms say auto-renewal uses the price and promotion in force on the charge date, not the price of your original purchase. That clause matters more here than on most subscriptions, because the list price has moved.

Z.ai’s April 21, 2026 notice listed Pro at $72 and Max at $160 and mentioned an earlier increase without dating it. Today they’re $80 and $168. Lite stayed at $18. Subscribers posting on GitHub and Hacker News recall paying $15 and $30 for Pro earlier in the year, which we couldn’t confirm on an official page. The term discounts moved the other way: quarterly went from about 10% to 20%, yearly from about 20% to 30%.

A yearly plan locks today’s price for twelve months, which is the one protection against another increase. It also locks you in with no refund. Which risk is larger depends on whether you have already used the plan for a month.

Can you get a refund or cancel?

You can cancel, and you can’t get a refund. Z.ai’s FAQ and usage policy both say a plan is non-refundable once purchased, even if you never use it. Cancelling stops the next charge, and the period you paid for stays valid to its end.

Payment is by bank card or PayPal, processed through Stripe. One more rule belongs here because it can cost you the plan: it may only be used inside the supported tools. The usage policy says other access can lead to restricted benefits, and repeated violations to a ban. Account sharing and resale are prohibited.

Where does your code go?

To a Singapore company, under terms that are less protective for individuals than for teams. The plan is sold by Jingsheng Hengxing Technology Pte. Ltd., and Z.ai’s privacy policy says data is generally processed in Singapore and may be transferred to group affiliates overseas. Z.ai is the international brand of Zhipu AI, a Beijing company that listed in Hong Kong in January 2026.

On training, the Terms of Use say Z.ai may use individual users’ content to improve its services, and that enterprise and API customers’ content is excluded unless they agree. Team plans are excluded by default. The individual plan cards advertise “Default data privacy” without defining it, and we found no opt-out switch described for individual subscribers. If that matters for your code, ask Z.ai in writing or buy the team plan.

Two more facts belong in a buying decision. In September 2026, Silicon UK reported that Z.ai’s own ZCode desktop app had been uploading workspace snapshots, including Git history, by default. Z.ai apologised and removed the feature. The report concerned ZCode, not the plan used through Claude Code or another agent.

And in January 2025 the US Commerce Department added Beijing Zhipu Huazhang Technology and nine related companies to its Entity List. The rule restricts exports of US-controlled items to those companies. It doesn’t mention buying a subscription, and the Singapore seller isn’t among the names. Lawyers at Sheppard Mullin wrote in July 2026 that sending export-controlled technical data into the hosted API would need a licence, and that the risk is real but untested. If your code is export-controlled or regulated, get advice before using this or any foreign-hosted model. The rule itself says nothing about ordinary subscribers, and we aren’t lawyers.

What do subscribers report?

Reports are mixed, and most of them are about the limits. On Hacker News in September 2026, one subscriber said they had pushed about 900 million tokens, 95% of them cached, through the $18 plan on GLM-5.3-Flash. The same day, in a different thread, another said the plan had been a decent deal but they kept running into its limits. Since they only code in spurts on personal projects, paying per token for GLM-5.3-Flash through OpenRouter now works out better for them.

Reliability is the other recurring theme. In June 2026 a paying subscriber logged hundreds of rate-limit errors over several days in a GitHub issue, and Pro and Max users replied with the same experience. On Hacker News, a Lite subscriber described speeds dipping below 60 tokens a second. Those reports predate GLM-5.3 and the credits system, so treat them as history, not as today’s service level. They are also the reason to start monthly.

Which tier should you choose?

• Lite if you work on one project at a time and want to find out whether GLM-5.3 is good enough for your code. Five full sessions a week is plenty for evenings and weekends.
• Pro if coding agents are your daily tool. Six times the quota for 4.4 times the price, with faster generation. Yearly billing brings it to $56 a month.
• Max if you run several agents in parallel all day. Z.ai publishes no numeric concurrency limit, so confirm that it handles your setup during a monthly period first.
• The API instead if your usage is occasional. No weekly cap, no renewal, and Flash costs $0.15 per million input tokens.

How does it compare with other coding plans?

At entry level the plans cluster within a few dollars of each other, so price settles little. What differs is the model behind the plan and how openly the quota is stated. Z.ai publishes credit numbers. Anthropic and OpenAI publish multipliers and ranges.

The comparison most readers want is with Anthropic’s own plans, since the GLM plan runs inside Claude Code. We cover prices, independent benchmark results and data terms side by side in GLM Coding Plan vs Claude Code.