The GLM Coding Plan costs $18 a month for Lite, $80 for Pro and $168 for Max on monthly billing, and paying for a year cuts that by 30%, to $12.60, $56 and $117.60 a month. Each plan is a weekly pot of credits that you spend from 20+ supported agent tools, including Claude Code, Cline and OpenCode (apps where the AI reads your files and works through the steps for you), all running its GLM-5.3 model.
Our take: Lite at $18 is worth it as a second engine next to Claude, if you can do most of your heavy work outside Z.ai’s peak hours. Pro and Max are a harder sell, because they sit either side of Claude Max’s $100 price, and there are no refunds if the credits don’t suit you.
⚡ TL;DR: The GLM Coding Plan in One Minute
What It Is: A flat subscription from Z.ai (the company behind the GLM models) that runs GLM-5.3 in 20+ agent tools, including Claude Code.
Price: Lite $18, Pro $80, Max $168 a month; 20% off quarterly, 30% off yearly.
What You Get: Lite gets 10,000 credits a week and 2,000 per 5 hours; Pro six times that, Max 14 times. In our worked example, Lite’s week at peak covers roughly 34 to 46 typed requests in an agent tool, or twice that off-peak: enough for part-time use, not for full-time coding.
Our Take: Lite is the good deal. Size it by the worked maths below, not by the old “prompts per cycle” figures.
⚠️ The Catch: Weekday peak hours (2 pm to 6 pm Singapore time) cost double the off-peak rate, there are no refunds, and use outside supported tools can restrict your plan (more than three violations may get an account banned).
📑 Quick Navigation
GLM Coding Plan pricing: Lite, Pro and Max in October 2026
These are the individual plans on Z.ai’s subscribe page as read for this update, in US dollars. All three include GLM-5.3, GLM-5.3-Flash and, per Z.ai’s docs, the same MCP tools (add-ons that let the model see images, search the web and read pages); the subscribe page itself lists “a curated selection of MCP tools” under Pro.
| Plan | Monthly billing | Quarterly (20% off) | Yearly (30% off) | Credits: per 5 hours / per week |
|---|---|---|---|---|
| Lite | $18 | $14.40 a month | $12.60 a month | 2,000 / 10,000 |
| Pro | $80 | $64 a month | $56 a month | 12,000 / 60,000 |
| Max | $168 | $134.40 a month | $117.60 a month | 28,000 / 140,000 |
Over a year, Lite on yearly billing comes to $151.20 instead of $216, a $64.80 saving. Pro saves $288 and Max $604.80. Pro also lists faster generation speeds, and Max dedicated resources at peak times.
The 5-hour pot refills 5 hours after you spend it; the weekly pot starts on the day you subscribe and resets every 7 days. Run out and you wait: Z.ai says the plan doesn’t charge your account balance when the quota is used up.
Teams buy seats instead: a Standard Seat is $88 per seat a month ($79.20 on yearly billing) with 66,000 credits a week, a Premium Seat $188 with 155,000, and a team needs at least 2 seats.
What 10,000 credits a week buys: the worked maths
This is the part the subscribe page doesn’t explain. On Z.ai’s plan overview, a GLM-5.3 call costs (new input tokens x 6.9 + cached input tokens x 1.7 + output tokens x 24) / 10,000 credits. A token is a short chunk of text, a few characters long. Cached tokens are the parts of your project the tool resends on every call, which are much cheaper.
Two rules change the bill. Off-peak, every model call costs half. Peak hours are Monday to Friday, 2 pm to 6 pm Singapore time (UTC+8), so weekends are off-peak all day; in winter that window is 6 am to 10 am in London and 1 am to 5 am in New York.
Say each model call in Claude Code sends 50,000 tokens of context, 95% of them cache hits (47,500 cached, 2,500 new), and gets 2,000 tokens back. That call costs (2,500 x 6.9 + 47,500 x 1.7 + 2,000 x 24) / 10,000 = 14.6 credits at peak, or 7.3 off-peak.
| Plan | Calls per week, all at peak | Calls per week, all off-peak |
|---|---|---|
| Lite (10,000 credits) | about 685 | about 1,370 |
| Pro (60,000) | about 4,110 | about 8,219 |
| Max (140,000) | about 9,589 | about 19,178 |
Calls are not prompts. An agent tool calls the model many times for one request you type: Z.ai’s own notice on the old plans estimated 15 to 20 calls per prompt. At that rate, Lite’s 685 peak calls a week are roughly 34 to 46 typed requests, and the 5-hour pot (about 137 calls at peak) covers about 7 to 9 of them.
Is that good value? Developers who build GLM into their own code pay by the token on Z.ai’s API: $1.40 per million input tokens, $0.26 cached and $4.40 output. At those prices the same call costs about 2.5 cents. If you used the full quota at peak, four weeks of Lite would be worth about $67.56 of API calls, for $18. Z.ai’s own estimate at a 95% cache rate is 48 to 97 million GLM-5.3 tokens a week on Lite, depending on how much falls in peak hours. Our example assumes more output per call, so it lands lower, at about 36 million tokens a week at peak.
📊 Model Calls per Week in the Worked Example, Peak vs Off-Peak
Z.ai’s headline saving figure leans on that same off-peak discount, so read it with your own working hours in mind.
🔍 REALITY CHECK
The claim: “you can save up to 92% compared with pay-as-you-go calls to the GLM-5.3 standard API”
The fine print: That figure assumes every call lands off-peak. Z.ai’s own range runs from all-peak use at the low end to all-off-peak at the high end, and peak calls cost twice as much.
Verdict: If you code on weekday afternoons in Asia, plan around the low end of Z.ai’s estimates.
What changed in the August 2026 GLM coding model update, and the credits switch
Z.ai released GLM-5.3 on August 18, 2026, and reports a 50% gain over GLM-5.2 on Z.ai Code Bench, its own in-house benchmark (a scored set of coding tasks). It is built on the same base model as GLM-5.2, improved through further training. GLM-5.3-Flash, a smaller model that can read images, followed on August 26.
- Older models are gone from the plan. If an older setup still asks for GLM-5.2 or GLM-5.1, Z.ai routes it to GLM-5.3; GLM-4.7 requests go to GLM-5.3-Flash.
- Credits replaced prompts on July 30. The old plans counted prompts (about 80 per 5 hours on Lite) and are no longer sold to new users. Anyone who subscribed before July 30 keeps those terms until their current billing cycle ends, and plans discontinued on July 30 can still be renewed or upgraded.
- Keys split on October 16. For keys created from that date, a Coding Plan key uses plan credits and a standard API key uses standard billing, whatever tool you paste it into.
Does GLM-5.3 still have a 1M-token window?
Yes. GLM-5.3 keeps a 1M-token context window, which is how much text it can keep in view at once: enough for a large part of a codebase, and it writes up to 128K tokens in one reply. It takes text only, and reasoning is always on, with low, high and max effort levels. For screenshots or UI mockups, the plan’s Vision Understanding MCP tool does the looking.
How to use the GLM Coding Plan in Claude Code
Z.ai’s Claude Code guide covers the GLM and Claude Code setup in three steps:
- Subscribe, then create the key under API Keys, Personal Coding Plan. A standard API key won’t touch your plan credits and may bill your balance instead.
- In ~/.claude/settings.json, set ANTHROPIC_AUTH_TOKEN to that key, ANTHROPIC_BASE_URL to https://api.z.ai/api/anthropic, and API_TIMEOUT_MS to 3000000. To get the full window, map ANTHROPIC_DEFAULT_OPUS_MODEL and ANTHROPIC_DEFAULT_SONNET_MODEL to glm-5.3[1m] and ANTHROPIC_DEFAULT_HAIKU_MODEL to glm-5.3-flash[1m].
- Open a new terminal and run claude. Or skip the editing: npx @z_ai/coding-helper sets up supported tools for you.
Most other tools use the OpenAI-style endpoint https://api.z.ai/api/coding/paas/v4; Goose, like Claude Code, uses the Anthropic one. To switch back to your Claude login, remove those lines and restart Claude Code. Our guide to Claude Code Router shows how to send only some tasks to a cheaper model; check that any router sits on Z.ai’s supported-tools list before you use it with the plan, because the plan is limited to those tools. AI agents for developers compares the front-ends.
The fine print: refunds, cancelling and banned tools
- No refunds. Z.ai’s FAQ says subscriptions are non-refundable once bought, even if you haven’t used the plan. Start on monthly Lite before you pay for a year.
- Cancel early. Z.ai’s two pages disagree: the FAQ says at least 24 hours before your billing date, the usage policy says at least 3 days. Go with 3 days. The plan runs to the end of the period you paid for.
- Supported tools only. The plan is limited to officially supported tools. Account sharing is prohibited, using it in other tools can restrict your plan, and more than three violations may get the account banned.
- Upgrades. Moving up a tier applies at once, and the old plan’s remaining value goes to your account balance, pro rata; switching Lite monthly to Lite yearly only starts when the current plan ends.
- Concurrency. Z.ai suggests Lite for one project at a time, Pro for one or two and Max for two or more.
GLM Coding Plan vs Claude Pro and Claude Max
The fair comparison is with Claude’s own plans, since Z.ai’s docs lead with Claude Code among the supported tools.
| GLM Coding Plan | Claude Pro / Max / Ultra | |
|---|---|---|
| Monthly price | Lite $18, Pro $80, Max $168 | Pro $20 ($17 a month yearly), Max $100 (5x Pro usage), Ultra $200 (10x Pro weekly usage) |
| Model | GLM-5.3 and GLM-5.3-Flash | Claude models |
| Works in | 20+ supported agent tools, including Claude Code | Claude Code included |
| Usage meter | Credits per 5 hours and per week, half price off-peak | Usage limits per plan |
| Refunds | None once bought | See Anthropic’s terms |
Lite at $18 is $2 cheaper than Claude Pro, and the worked maths above shows it buys a lot of calls. Higher up, the gap disappears: GLM Pro at $80 sits just below Claude Max’s $100 price, and GLM Max at $168 above it. The two meters can’t be compared directly. If Claude models are what you want, price Claude Max first, or Claude Pro plus GLM Lite for overflow, which our guide to what Claude Code really costs lays out.
On speed and quality, the best outside measure we found is for the API, not the plan. Artificial Analysis, an independent firm that measures AI models, clocks GLM-5.3 (Max) through the API at 78 tokens per second, against an average of 75, and scores it 45 on its Intelligence Index against a median of 19 for comparable models. It also calls it “very verbose”. Coding Plan speed may differ by tier: Z.ai lists faster generation speeds as a Pro benefit.
💡 Key Takeaway: Buy GLM for the Lite price. From Pro up, the GLM Coding Plan lands either side of Claude Max’s $100 price, so the reason to pick it has to be the model or the tools, not the saving.
Is your code safe with Z.ai?
Z.ai’s privacy policy names a Singapore company, JINGSHENG HENGXING TECHNOLOGY PTE.LTD, as the operator, and says personal data is generally processed in Singapore. Its data processing addendum for business API customers says the content they send is processed in real time and not saved on Z.ai’s servers.
Lite lists “Default data privacy” (Pro and Max include all Lite benefits), and the team plan says data is not used for model training by default. The pages we read in October 2026 don’t say whether code sent on an individual plan is used for training, and they show no independent audit of these promises. For client code under a confidentiality agreement, ask the client before you route it through any third-party model.
The open weights are the other way out. GLM-5.3 is on Hugging Face with 753B parameters under Z.ai’s own GLM-5.3 license, not the MIT license GLM-5.2 used; GLM-5.3-Flash keeps MIT. At that size, self-hosting is a job for a company with a GPU cluster, not a laptop.
Is the GLM Coding Plan worth it? Who should pay
- You keep hitting Claude Pro’s limits: add GLM Lite at $18 and send routine work to it. Together that is $38 a month, less than half of Claude Max.
- You code evenings or weekends in Europe or the Americas: most of your hours fall outside Z.ai’s peak window, so your credits go twice as far.
- You run several agents at once all day: Pro or Max make sense only if you have priced Claude Max for the same work first.
- Skip it if you can’t risk a non-refundable plan on a meter you haven’t tried, or if your contracts rule out third-party models.
GLM Coding Plan FAQ
How much is the GLM Coding Plan?
Lite is $18 a month, Pro $80 and Max $168 on monthly billing, or $12.60, $56 and $117.60 a month on yearly billing. Team plans cost $88 per seat a month ($79.20 yearly), with at least 2 seats.
Is GLM Coding Lite enough?
For one project and part-time use, usually yes. Lite gives 10,000 credits a week and 2,000 per 5 hours, which is about 685 to 1,370 calls a week in our worked example, and Z.ai recommends it for one project at a time.
Can I use GLM in Claude Code?
Yes. Point ANTHROPIC_BASE_URL at https://api.z.ai/api/anthropic and use your Coding Plan key as ANTHROPIC_AUTH_TOKEN; Z.ai’s helper (npx @z_ai/coding-helper) can do it for you.
What happened to GLM-5.1, GLM-5.2 and GLM-4.6 on the coding plan?
The plan runs GLM-5.3 and GLM-5.3-Flash only. Requests for GLM-5.2 and GLM-5.1 go to GLM-5.3, and GLM-4.7 requests to GLM-5.3-Flash; Z.ai’s plan pages we read don’t mention GLM-4.6.
Why is my account balance charged when I have a coding plan?
The most likely cause is the key or the address. For keys created from October 16, 2026, only a Coding Plan key draws on plan credits; a standard key may bill pay-as-you-go even in a coding tool. An older key keeps working with your plan, but pointed at the standard address https://api.z.ai/api/paas/v4 it may still charge your balance while plan quota remains, so use /api/coding/paas/v4 or /api/anthropic.
Can I get a refund on the GLM Coding Plan?
No. Z.ai says subscriptions are non-refundable once purchased, even if unused, so cancel at least 3 days before renewal to be safe.
GLM Coding Plan review verdict
The GLM Coding Plan is a cheap way to add a second agent engine, but only at the bottom of the range. Lite at $18 buys far more API value than it costs if you keep most work off-peak; Pro and Max land either side of Claude Max’s $100 price. The credit system is more transparent than the old prompt counts, because Z.ai publishes the formula, but you have to do the maths yourself, and you pay before you know whether it fits.
✅ What We Liked
- ✓ Lite at $18 a month, $12.60 on yearly billing
- ✓ Z.ai publishes the credit formula and the off-peak discount
- ✓ Works in 20+ agent tools, including Claude Code
- ✓ GLM-5.3 and its 1M-token window on every tier (with the [1m] model setting)
- ✓ GLM-5.3’s API runs faster than average in Artificial Analysis’s measurements
❌ What Fell Short
- ✗ No refunds, even on unused time
- ✗ Weekday peak hours cost double
- ✗ Pro $80 and Max $168 sit either side of Claude Max’s $100 price
- ✗ Use outside supported tools can restrict the plan; repeated violations can mean a ban
- ✗ Two Z.ai pages give different cancellation deadlines
Start with one month of Lite at z.ai/subscribe, watch the weekly meter, and move up only after you’ve hit it a few times. For another budget model, read our Kimi AI review or DeepSeek review.
Founder of AI Tool Analysis. Reads every pricing page, help article and footnote so you don’t have to. Covering AI tools since 2025. See how we research →
Related Reading
- Is Claude Code Free? What You Really Pay
- Claude Code Router Review
- Every Claude Plan Explained
- Kimi AI Review
- DeepSeek Review
- Claude Code vs Cursor
- ChatGPT Codex Review
- Goose AI Review
Stay Ahead of the AI Curve
Don’t miss the next big update. Get AI Weekly delivered every Thursday at 9 AM EST with honest reviews, price drops, and features you can actually use.
- ✅ Sourced Reviews: Every price and limit checked against the vendor’s own pages, not rewritten from press releases
- ✅ Price Tracking: Know when tools drop prices or add free tiers
- ✅ Feature Launches: Major updates covered within days
- ✅ Comparison Updates: As the market shifts, we update our verdicts
- ✅ No Hype: Just the AI news that actually matters for your work
Free, unsubscribe anytime.
Last Updated: October 31, 2026
Facts checked: October 31, 2026 (prices, credits and policies against Z.ai’s subscribe, docs and policy pages)
Next Review Update: November 2026
Have a tool you want us to review? Suggest it here | Questions? Contact us