All quota data in this article comes from screenshots and official announcement pages shared by users in Zhipu's official Feishu group "GLM Coding User Exchange Group 1⃣ | Official," collected between June 22 and July 5, 2026. Estimated figures are clearly marked; numbers not directly verified by screenshots are not presented as facts.
Why This Article?
Zhipu's GLM Coding Plan official page describes subscription quotas in "prompt counts" — for example, "approximately 80 prompts per 5 hours" for the Lite tier. But this phrasing is frustratingly vague: how many tokens does one prompt actually consume? Does a request with 100K tokens of context count the same as one with 1,000 tokens?
The answer: the plan actually charges by token consumption, not prompt count. The "prompt count" on the official page is just a fuzzy estimate for easier understanding. The real limits are token-based — both the 5-hour cap and the weekly cap are token ceilings. The "3x consumption during peak hours" doesn't mean "3 prompts count as 1," but rather that token consumption is multiplied by 3.
This article breaks down the real token-based billing rules: how many tokens each tier actually provides, how tight the weekly limits are, and whether the subscription plans are actually more cost-effective than pay-per-token API billing — so you can make an informed decision before subscribing, without being confused by the vague "prompt count" terminology.
Background: Why Is Everyone Asking About Plan Quotas?
Zhipu's GLM Coding Plan is one of China's earliest AI coding subscription services, powered by the flagship GLM-5.2 model (1M context window, MIT open weights), supporting Claude Code, Cursor, Cline, Roo Code, and 20+ mainstream coding tools.
In February 2026, Zhipu significantly repriced the plans: China-region Lite/Pro/Max went from ¥40/¥200/¥400 to ¥49/¥149/¥469, with quotas cut by roughly one-third. The international version saw even steeper hikes, going from early pricing of $6/$30/$80 through multiple increases to $18/$72/$160. Meanwhile, GLM-5.2's thinking token consumption soared compared to GLM-5.1, with group members universally reporting "doubled costs" — making "how much can the plan actually be used" the most pressing question for developers.
I. Current Offerings: New Plan Tier Pricing
1.1 China Region Pricing (bigmodel.cn)
The following data is from the official purchase page screenshot dated June 29, 2026, with annual subscription at 20% off:
| Tier | Original (Monthly) | Annual (Discounted) | Key Features |
|---|---|---|---|
| Lite | ¥49/mo | ¥39.2/mo | Basic usage quota, suitable for small repos and lightweight iteration |
| Pro Most Popular | ¥149/mo | ¥119.2/mo | 5x Lite usage quota, faster generation speed |
| Max Maximum Volume | ¥469/mo | ¥375.2/mo | 20x Lite usage quota, peak-hour dedicated resource priority |
1.2 International Region Pricing (z.ai)
The following data is from the z.ai/subscribe subscription page screenshot shared by a group member who received an email from Zhipu's English site, with annual subscription at approximately 30% off:
| Tier | Original (Monthly) | Annual (Discounted) | Features |
|---|---|---|---|
| Lite | $18/mo | $12.6/mo | Basic usage quota |
| Pro Popular | $72/mo | $50.4/mo | 5x Lite usage quota, priority access to latest flagship models |
| Max Max Usage | $160/mo | $112/mo | 20x Lite usage quota, first access to latest flagship models |
Cross-Region Price Gap: For the same configuration, international pricing is over 2x the China region (Max: $160 ≈ ¥1,152 vs. China ¥469). Group members commented: "Requires PayPal to buy" and "Way more expensive." Overseas developers are even researching how to register Alipay to purchase China-region plans — a curious "Western tax" in the AI era.
1.3 Team Plan Pricing
| Tier | Monthly (Annual Discounted) | 5-Hour Quota | Weekly Quota |
|---|---|---|---|
| Standard | ¥598/mo (¥538.20) | 60M tokens | 300M tokens |
| Premium | ¥1,198/mo (¥1,078.20) | 160M tokens | 800M tokens |
Team plan quotas are shared across the entire team, not per individual. Group members confirmed: "5 hours max 160M tokens — that's definitely the total for everyone."
II. Real Quota Data: The 5-Hour / Weekly / Monthly Triple Limit
The new plans use a three-tier quota mechanism, which is the biggest difference from the old plans:
| Dimension | Old Plan | New Plan |
|---|---|---|
| 5-hour quota | Yes (only limit) | Yes |
| Weekly quota | None | Yes ← New restriction |
| MCP monthly quota | Unclear | Yes |
| Peak-hour 3x consumption | Applies | Applies |
| Availability | Discontinued | Available (requires rush purchase) |
2.1 5-Hour Quota (Confirmed + Estimated)
| Tier | Off-Peak | Peak (3x consumption) | Data Source |
|---|---|---|---|
| Lite | 30M tokens | 10M tokens | Directly confirmed by screenshot |
| Pro | ~150M tokens (estimated) | ~50M tokens (estimated) | Calculated at 5x Lite |
| Max | ~600M tokens (estimated) | ~200M tokens (estimated) | Calculated at 20x Lite |
Lite data from group member screenshot (June 24, 2026): "Lite new plan 5-hour quota can only push out 10M tokens — less than 1/3 of the original." During peak hours, GLM-5.2 consumes 3x the quota per call; off-peak hours consume normally.
2.2 Weekly Quota
| Tier | Off-Peak | Peak | Data Source |
|---|---|---|---|
| Lite | Unclear | — | Screenshot only shows percentage |
| Pro | ~250M tokens (estimated) | — | Calculated at 5x Lite |
| Max | ~1 billion tokens | 300–400M tokens | Confirmed by group members |
Max weekly limit directly confirmed by group members (July 2, 2026): "Max 1 billion per week" and "Off-peak 1 billion, peak 3x consumption, peak only 300–400M."
Weekly quota = 5h quota × 5: Team plan data shows the weekly quota is exactly 5x the 5-hour quota (Standard: 60M × 5 = 300M; Premium: 160M × 5 = 800M). If individual plans follow the same ratio, Lite weekly ≈ 150M and Pro ≈ 750M. However, Max was confirmed at 1 billion (not 600M × 5 = 3 billion), suggesting individual plans may use a different weekly/5h ratio, or the 5h estimate is too high.
2.3 Usage Statistics UI Evidence
Screenshots of usage statistics pages shared by group members clearly show the difference between old and new plans:
New Plan Page (3 cards):
- 5-hour usage: 90% used, resets at 08:19
- Weekly usage: 32% used, resets July 2, 2026 09:59
- MCP monthly usage: 1% used, resets July 25, 2026 09:59
Old Plan Page (2 cards):
- 5-hour usage: 100% used, resets at 20:30
- MCP monthly usage: 0% used, resets July 21, 2026 18:00
- (No "Weekly usage" card)
III. GLM-5.2 Thinking Token Surge: The Hidden Cost Crisis
This is the most concentrated pain point in group discussions. GLM-5.2's thinking mode consumes significantly more tokens than GLM-5.1, directly impacting the usable duration of plans:
- July 4, 2026 group feedback: "Now it casually thinks in thousands of tokens — is this really reasonable?" — "Just say two sentences and it thinks over 10K tokens"
- Same thread: "Latency increased dramatically, costs skyrocketed, doubled" / "Can't cover the costs, oh my god"
- "Only the old plan is affordable; the new plan can't handle it"
- June 28, 2026: "Feels like 5.2 consumes way more tokens than 5.1; old plan Lite is completely insufficient"
One heavy user shared (June 30, 2026): "Used nearly 5 billion this month. I can't live without GLM anymore. The only issue is that peak-hour speed isn't fast enough."
IV. Old Plan vs. New Plan: How Big Is the Gap?
The old plan is discontinued, but comparison data from group discussions is staggering:
| Feature | Old Plan | New Plan |
|---|---|---|
| Weekly limit | None | Yes |
| Lite 5h quota | ~30M (off-peak) | 30M (off-peak) / 10M (peak) |
| Pro/Max 5h quota | ~40M | Estimated 150M / 600M |
| Overall capacity feel | "Old Max is nearly 10x new Max" | — |
A Xianyu (second-hand marketplace) listing explicitly states: "Zhipu GLM discontinued old Max plan annual account... no weekly usage limit, quota is nearly 10x the new Max plan."
Group member testimony (July 2, 2026): Old Lite "80M tokens per day, nearly 600M per week" — while new Lite weekly limit is estimated at only 50–150M, a 4–12x gap.
A GitHub blogger's humorous annotation feels particularly apt here: "It seems at Zhipu you only work 4 days a week, 5 hours a day." — The new plan's weekly limit equals five 5-hour windows, while the old plan had no such restriction.
V. Plan vs. API Billing: Which Is More Cost-Effective?
This is the core question of this article. We compare using GLM-5.2's official API pricing:
5.1 API Pricing Baseline
| Billing Item | Price (¥/M tokens) |
|---|---|
| Input | ¥8 |
| Output | ¥28 |
| Cache Hit | ¥2 |
5.2 Real Input-Output Ratio: 198:1
In coding scenarios, input vastly exceeds output — but you might not realize how extreme the gap is. We analyzed real call data from SpanAgent's production environment (Hong Kong server) for the GLM-5.2 model:
| Metric | Value |
|---|---|
| Total Requests Analyzed | 5,729 |
| Total Input Tokens | 540,887,462 (541M) |
| Total Output Tokens | 2,725,565 (2.73M) |
| Input:Output Ratio | 198.4 : 1 |
This means that in real AI coding workflows, generating 1 token of output requires nearly 200 tokens of input (codebase, conversation history, system prompts, etc.). At this ratio, the blended API cost is only ~¥8.10/M tokens — nearly approaching the pure input price, since output is a negligible fraction.
Why is the ratio so extreme? Because agentic coding tools (Claude Code, Cline, etc.) send the complete code context, file contents, and conversation history as input with every call, while output is only a small completion or response from the model. The longer the context, the more input bloat — GLM-5.2 supports a 1M context window, and single requests carrying hundreds of thousands of input tokens are the norm in practice.
5.3 Monthly Cost Comparison
Using the confirmed Max weekly limit of 1 billion tokens as the baseline, estimating monthly quotas at Max=20x Lite, Pro=5x Lite (4.33 weeks/month), and the real blended cost of ¥8.10/M tokens:
| Tier | Weekly Quota | Monthly Quota | Plan Annual Price | Same Budget via API | Plan / API |
|---|---|---|---|---|---|
| Lite | ~50M | ~220M | ¥39.2/mo | 5M | 45x |
| Pro | ~250M | ~1.08B | ¥119.2/mo | 15M | 74x |
| Max | ~1B | ~4.33B | ¥375.2/mo | 46M | 93x |
The gap is crushing. Take the Max plan: at ¥375.2/month (annual), spending the same amount on direct API calls would buy only ~46M tokens — while the Max plan provides a monthly quota of 4.33 billion tokens. The plan delivers 93x more tokens than the same budget on API.
5.4 Purchasing Power Per Yuan: The Real Truth
| Option | Tokens per ¥1 | vs. API |
|---|---|---|
| Direct API billing | ~123K tokens/¥1 | Baseline |
| Lite Plan | ~5.52M tokens/¥1 | 45x |
| Pro Plan | ~9.08M tokens/¥1 | 74x |
| Max Plan | ~11.54M tokens/¥1 | 93x |
The conclusion is clear: At the real input-output ratio (198:1), plans crush direct API billing across all tiers. The reason is simple: API pay-per-use charges ¥8.10 per million tokens, while plans spread the fixed monthly fee across billions of tokens in quota, driving unit costs far below what API can compete with. The higher the tier, the larger the monthly quota, and the stronger the purchasing power — Max plan delivers 93x more tokens per yuan than API.
Why is the gap so enormous? Because the plan's effective token unit price is far below the official API rate. Max plan at ¥375.2/month with 4.33B tokens monthly quota works out to ~¥0.087/M tokens — while the API blended cost is ¥8.10/M tokens, a 93x difference. The plan is essentially Zhipu amortizing compute costs at scale through subscription pricing, then passing tokens to developers at a unit price far below API rates.
5.5 What's the Catch?
Plans aren't without costs — they're just hidden in the restrictions:
- Weekly quota: New plans have a weekly total cap (Max confirmed at 1B/week). If you burn through it early in the week, you wait for reset. API has no weekly limit — call anytime.
- Peak-hour 3x consumption: During weekday peak hours, each call consumes 3x quota, effectively reducing usable tokens to 1/3. Max's 1B/week becomes only 300–400M during peak.
- 5-hour window limit: Each 5-hour refresh cycle means Lite only has 10M tokens during peak hours — deep tasks may not complete.
- Rush purchase required: Group members report "every day at 10 AM they open for purchase, but there's actually no stock." Max tier is especially scarce.
- GLM-5.2 thinking token surge: 5.2's thinking consumption increased dramatically over 5.1, with members reporting "doubled costs" — this accelerates 5h and weekly quota depletion.
5.6 How to Choose? The Bottom Line
| Your Situation | Recommended Option | Reason |
|---|---|---|
| Light usage, occasional scripting | API pay-per-use | Fixed monthly fee wasted if rarely used; API is cheaper for sporadic use |
| Daily AI coding 2–4 hours | Pro Plan | 74x API purchasing power; 5h quota sufficient for daily iteration |
| Full-time AI coding, agentic workflows | Max Plan | 93x API purchasing power; peak-hour dedicated resources |
| Highly variable usage, sometimes heavy sometimes idle | API pay-per-use | Plan wasted when idle; API only charges when used |
| Team of 3–5 sharing | Team Premium | 160M/5h shared quota; per-person cost lower than individual Max |
| Maximum cost-efficiency with stable usage | Max Plan Annual | 11.54M tokens/¥1 — 93x API |
Important note: The above comparison is based on "off-peak" weekly limits. Peak-hour 3x consumption means actual usable tokens are 1/3 of off-peak. But even so, plan purchasing power remains 30x+ API (Max at peak ≈ 31x), still significantly better than API.
VI. Purchasing Tips & Notes
- Rush purchase required: Group members report "every day at 10 AM they open for purchase, but there's actually no stock" (July 3, 2026). Set an alarm and be ready, especially for Max tier.
- Try monthly before annual: Start with monthly billing to confirm your actual usage pattern, then lock in the 20% annual discount. If you rarely hit the 5h limit, your tier may be too high; if you frequently max out the weekly quota, consider upgrading.
- Watch GLM-5.2 thinking mode: Thinking token surge is the biggest cost variable right now. If your tool supports disabling or reducing thinking effort, it can significantly extend plan usability.
- International users should consider China region: International pricing is 2x+ the China region for the same quotas. If you have Alipay/WeChat Pay access, China-region plans offer far better value.
Conclusion
The three-tier quota mechanism (5h + weekly + monthly) of GLM Coding Plan is essentially Zhipu's resource allocation strategy under compute supply constraints. The old plan's "no weekly limit" era is gone; new plan users must adapt to a "rationed usage" model.
From a cost-efficiency perspective, plans significantly outperform direct API billing across all tiers — at the real 198:1 input-output ratio, Max plan purchasing power is 93x API. But this doesn't mean blindly choosing a plan: light users waste the fixed monthly fee, and variable-usage scenarios may benefit from API's flexibility. The key is understanding your own usage pattern and choosing the right option.
Try GLM-5.2 via SpanAgent
If you're looking for a reliable way to access GLM-5.2 without committing to a subscription plan, SpanAgent offers GLM-5.2 through its API aggregation service. Here's why it's worth considering:
- Pay-per-request billing: Only pay for what you actually use — no wasted monthly fees during idle periods. Compared to official plan pricing, SpanAgent's per-request model is more cost-effective for variable workloads.
- No expiration anxiety: Plans expire — your subscription fees are gone if you don't use them. With SpanAgent, your balance stays until you use it. Top up anytime, use anytime.
- No rush purchases: No need to set alarms at 10 AM to fight for limited stock. SpanAgent is always available.
- Smart routing: Automatically selects the best available endpoint for GLM-5.2, ensuring stable performance even during peak hours.
- Full API compatibility: Works with Claude Code, Cursor, Cline, and all OpenAI-compatible tools.
Whether you're a light user who can't justify a subscription, or a heavy user looking for a backup when your plan quota runs out, SpanAgent provides a flexible, no-strings-attached way to call GLM-5.2. Visit spanagent.xyz to get started.
Data Notes & Disclaimer
All quota figures in this article come from the official plan page and group member screenshots. However, readers should note: some users have reported that the token usage shown on the official plan page is higher than what they measure through their client-side tools. This discrepancy may stem from:
- Official statistics may include hidden tokens (system prompts, tool call overhead) that clients don't display
- Thinking tokens (reasoning tokens) may be counted differently on the server vs. client side
- Whether cache-hit tokens count toward quota consumption has not been officially clarified
This means: the "monthly quota" figures in this article are based on the official counting method; your actual usable tokens may be slightly less than what the official page shows. However, this doesn't change the core conclusion — even with a discount applied, the plan's per-yuan token purchasing power still significantly outperforms direct API billing (Max plan at 50% of stated quota is still 46x API). We recommend readers compare official usage statistics with client-side statistics after subscribing to get a clear picture.
---
Data sources: Zhipu official Feishu group "GLM Coding User Exchange Group 1⃣ | Official" (June 22 – July 5, 2026), with 353 total messages analyzed, including 160 messages containing quota/pricing keywords and 32 supporting screenshots. API pricing source: Zhipu BigModel Open Platform. Estimated figures are marked; numbers not directly verified by screenshots are for reference only.