Zhipu GLM Coding Plan limits, measured
Zhipu publishes the GLM Coding Plan’s credit caps and per-model coefficients, but not how many tokens that is. The answer depends on the cache share of your requests, because cache hits cost about a quarter of uncached input. This page converts with the token mix of real coding traffic through Wokey: a Lite 5-hour window is about 6.1M tokens on GLM-5.3 and 30M on GLM-5.3-Flash. Outside weekdays 14:00–18:00 Beijing time credits are halved, so twice the tokens fit.
- Official credits and coefficients
- Real token mix
- Valued at API prices
- Lite · 5 hours · GLM-5.3
- 6.1M / 12M · peak / off-peak, coding-session mix
- Lite · 5 hours · GLM-5.3-Flash
- 30M / 60M · peak / off-peak, coding-session mix
- Weekly cap
- 5 windows · weekly credits are 5× the 5-hour credits; no monthly cap
- Cache hits
- about 25% · of the credits uncached input costs
Tokens in one 5-hour window
Official formula: credits = (input × input coefficient + cache hits × cache coefficient + output × output coefficient) ÷ 10,000. Coefficients (input / cache / output): 6.9 / 1.7 / 24 for GLM-5.3 and 2.3 / 0.56 / 8 for GLM-5.3-Flash. Each cell is “peak / off-peak”, and token counts include cache hits. “Coding session” uses the mix of real traffic.
| Model and usage | Credits per 1M tokens (peak) | Lite · 5 hours | Pro · 5 hours | Max · 5 hours |
|---|---|---|---|---|
| GLM-5.3 · coding session (86.5% cache, 5% output) | 326 | 6.1M / 12M | 37M / 74M | 86M / 170M |
| GLM-5.3 · almost no cache (10% output) | 861 | 2.3M / 4.6M | 14M / 28M | 33M / 65M |
| GLM-5.3-Flash · coding session (95.5% cache, 0.5% output) | 67 | 30M / 60M | 180M / 360M | 420M / 840M |
| GLM-5.3-Flash · almost no cache (10% output) | 287 | 7.0M / 14M | 42M / 84M | 98M / 200M |
A week is 5× these figures. For the same tokens GLM-5.3 costs 3× the credits of GLM-5.3-Flash. MCP calls (web search, page reader and so on) cost 1.2 credits each. Requests for GLM-5.2 or GLM-5.1 are switched to GLM-5.3, and GLM-5-Turbo or GLM-4.7 to GLM-5.3-Flash.
What a Lite month is worth
All of Lite’s weekly credits spent on one model, a month counted as 4.33 weeks, valued at that model’s official API reference price. Each cell is “all at peak / all off-peak”. Peak is only 20 hours a week, so real use sits closer to the off-peak side. The Wokey column is the same tokens paid per token through Wokey.
| Model and usage | Tokens per week | Month: official API price | Month: Wokey per token |
|---|---|---|---|
| GLM-5.3 · coding session (86.5% cache, 5% output) | 31M / 61M | $75 / $150 | $11 / $23 |
| GLM-5.3 · almost no cache (10% output) | 12M / 23M | $86 / $171 | $13 / $26 |
| GLM-5.3-Flash · coding session (95.5% cache, 0.5% output) | 150M / 300M | $24 / $48 | $12 / $24 |
| GLM-5.3-Flash · almost no cache (10% output) | 35M / 70M | $28 / $56 | $14 / $28 |
Lite costs ¥118 a month on monthly renewal, about $16 (at 1 USD ≈ 7.2 CNY). Pro has 6× Lite’s credits at about 4.6× the price, Max 14× at about 9.1×. A month of GLM-5.3 coding sessions that uses every weekly credit is worth about $75 (peak) to $150 (off-peak) at API prices.
Credit caps by plan (official)
There are only 5-hour and weekly caps, no monthly cap. The 5-hour credits refresh 5 hours after they are used; weekly credits refresh every 7 days from the order time. Prices are the monthly-renewal prices on Zhipu’s purchase page (checked 2026-11-14); quarterly billing is 20% off and yearly 30% off.
| Plan | Per month | Credits per 5 hours | Credits per week | Week ÷ 5 hours | Weekly credits per ¥1 |
|---|---|---|---|---|---|
| Lite | ¥118 | 2,000 | 10,000 | 5 | 85 |
| Pro | ¥538 | 12,000 | 60,000 | 5 | 112 |
| Max | ¥1,078 | 28,000 | 140,000 | 5 | 130 |
Per yuan, Max gives the most weekly credits (130), then Pro (112), then Lite (85). The ¥49 / ¥149 / ¥469 prices from before the move to credits on 2026-07-30 apply only to existing subscribers renewing at the old price.
Quota rules
- Peak is Monday to Friday 14:00–18:00 Beijing time, charged at 1×; all other hours are charged at 50%.
- When credits run out you wait for the next 5-hour period; other resource packs and account balance are not charged.
- Upgrading refreshes credits at the new plan’s level immediately; renewing does not reset them.
- The plan works only in officially supported tools (Claude Code, Codex, OpenCode, ZCode, WorkBuddy and others); API calls from elsewhere do not draw on it.
- OpenClaw is supported on secondary scheduling and can queue or be rate limited under load.
How to stretch the quota
- Avoid weekdays 14:00–18:00 Beijing time: the same work costs half the credits at any other hour.
- Keep the cache: cache hits cost about a quarter of uncached input. Avoid switching models or changing the system prompt, and do not restart long sessions needlessly.
- Use GLM-5.3-Flash for everyday work: the same tokens cost a third of GLM-5.3.
- Output is the most expensive: a GLM-5.3 output token costs about 3.5× an uncached input token, so asking for less long-form output saves the most.
- If you only hit the limit now and then, cover the overflow per token: GLM-5.3 through Wokey costs $0.21 input / $0.66 output per million tokens and $0.039 for cache reads, with no 5-hour window.
Method and limits
- Credit caps, coefficients and peak hours are from Zhipu’s documentation, unchanged.
- Token mixes come from real traffic sent through Wokey to 14 Coding Plan accounts from 2026-10-30 to 11-14. GLM-5.3-Flash is the median of 12 accounts (95.5% cache, 0.5% output); GLM-5.3 is the three accounts with the most traffic (95M tokens, 86.5% cache, 5% output).
- GLM-5.3’s mix varies widely by account: some are over 60% output with almost no cache, which is not coding use. The “almost no cache” row covers that case; it costs about 2.6× the credits per million tokens of a coding session.
- Whether real deductions match the official formula is not answered here: account owners also use their accounts outside Wokey, and the data cannot separate that use.
- Thinking tokens count as output. Zhipu can change coefficients, caps or peak hours at any time; the date on this page is the last check.
FAQ
How much quota does the GLM Coding Plan include?
It is counted in credits: Lite 2,000 per 5 hours and 10,000 a week; Pro 12,000 and 60,000; Max 28,000 and 140,000. There is no monthly cap. At the mix of real coding sessions, a Lite 5-hour window is about 6.1M tokens at peak to 12M off-peak on GLM-5.3, and 30M to 60M on GLM-5.3-Flash.
How are GLM Coding Plan credits calculated?
Credits = (input tokens × input coefficient + cached tokens × cache coefficient + output tokens × output coefficient) ÷ 10,000. The coefficients are 6.9 / 1.7 / 24 for GLM-5.3 and 2.3 / 0.56 / 8 for GLM-5.3-Flash. Weekdays 14:00–18:00 Beijing time are charged at 1× and all other hours at 50%. MCP calls cost 1.2 credits each.
Why does the GLM Coding Plan run out so fast?
Usually one of three reasons. Using it at peak hours, when credits are charged at twice the off-peak rate. Using GLM-5.3, which costs 3× the credits of GLM-5.3-Flash for the same tokens. Or missing the cache or generating a lot of output: with almost no cache GLM-5.3 costs about 861 credits per million tokens, 2.6× a normal coding session (about 326).
When does the GLM Coding Plan quota refresh?
The 5-hour credits refresh 5 hours after they are used; weekly credits refresh every 7 days from the order time. Weekly credits are 5× the 5-hour credits, so a week holds at most 5 full 5-hour windows. Renewing does not reset credits; upgrading refreshes them immediately.
Lite, Pro or Max?
Pro has 6× Lite’s credits and Max 14×, at ¥118, ¥538 and ¥1,078 a month. Weekly credits per yuan are highest on Max (130), then Pro (112), then Lite (85). A Lite window is only about 6.1M tokens of GLM-5.3 at peak, so GLM-5.3 as a daily driver needs at least Pro.
Do cache hits count toward the GLM Coding Plan quota?
Yes, at about 25% of uncached input (GLM-5.3: input coefficient 6.9, cache coefficient 1.7). Volcengine Ark’s Coding Plan gives cache hits no discount, and Kimi Code barely counts them.
Is the GLM Coding Plan worth it, or should I pay per token?
With GLM-5.3 coding sessions that use every weekly credit, a Lite month (¥118, about $16) is worth about $75 (all at peak) to $150 (all off-peak) at API prices, above the fee. If you do not use the credits up, or only use it occasionally, paying per token is cheaper.