Recently, I have been focusing on translating the Chinese historical articles on my blog into English, while also developing the Go Tour Chinese translation project. As the translation tasks progressed, the Zhipu resource pack I previously purchased quickly neared depletion.
I originally wanted to purchase another moderately sized resource pack to continue completing the remaining translation tasks. However, upon entering Zhipu’s resource pack page, I found that my account could not purchase the 20 million Tokens GLM-5.2 special offer pack currently priced at 39.9 yuan. The main options available to me were reduced to:
- 100 million Tokens;
- Price: 189.9 yuan;
- Validity period: 3 months.
The average unit price for 100 million Tokens is indeed very low, but for me, the real issue is not the unit price, but whether I can consume enough quota within three months.
If the daily Token usage drops significantly after the historical article translations are completed, purchasing such a large resource pack will likely result in a massive amount of quota expiring.
On the other hand, I had previously invited a friend to register a Zhipu account. My friend’s new account could purchase the 20 million Tokens, 39.9 yuan GLM-5.2 special offer pack. Since my friend does not plan to continue using the Zhipu platform, I began considering whether I could purchase this resource pack through his account with his consent.
This article combines my order records, actual resource pack consumption, Zhipu API pricing, and subsequent translation plans to analyze:
Between the 20 million Tokens special offer pack, the 100 million Tokens premium pack, and API pay-as-you-go billing, which one is truly more suitable for my current usage?
The resource packs and API prices discussed in this article are based on the page content displayed on August 6, 2026, and are subject to change in the future.
1. I previously purchased the GLM-5.1 10 million Tokens trial pack
First, I need to clarify a point of confusion.
What I previously purchased was not the 20 million Tokens GLM-5.2 special offer pack currently shown on the page.
According to the order details, what I purchased on June 13, 2026, was:
- Product name: GLM-5.1 70% off trial pack
- Resource pack capacity: 10 million Tokens
- Unit price: 29.9 yuan
- Quantity purchased: 1
- Creation time: June 13, 2026, 00:32:45
- Actual payment: 29.9 yuan
- Payment method: Account balance
From the order information, it is confirmed that I purchased the 10 million Tokens, 29.9 yuan GLM-5.1 70% off trial pack, not the 20 million Tokens GLM-5.2 special offer pack currently priced at 39.9 yuan.
Later, following the release of GLM-5.2, I noticed on the resource pack management page that the applicable scope of this resource pack now includes both GLM-5.1 and GLM-5.2.
Based on this change, it appears that after GLM-5.2 was released, Zhipu expanded the applicable models for the existing resource pack, allowing the unused Tokens to continue being used for GLM-5.2.
Therefore, it is necessary to distinguish between the following two different resource packs:
| Item | Resource pack I previously purchased | Resource pack currently unavailable for purchase |
|---|---|---|
| Product name | GLM-5.1 70% off trial pack | GLM-5.2 special offer pack |
| Tokens | 10 million | 20 million |
| Price | 29.9 yuan | 39.9 yuan |
| Purchase time | June 13, 2026 | Not yet purchased |
| Currently applicable models | Page shows support for GLM-5.1, GLM-5.2 | GLM-5.2 |
2. How much is left in the 10 million Tokens resource pack
To determine how much quota I still need, I recorded the resource pack balance on several occasions.
| Record time | Current balance |
|---|---|
| August 2, 2026, 09:44 | 3,602,778 Tokens |
| August 3, 2026, 10:58 | 2,914,063 Tokens |
| August 6, 2026, 11:56 | 1,241,576 Tokens |
From August 3 at 10:58 to August 6 at 11:56, a total of approximately 73 hours, the consumption was:
2,914,063 − 1,241,576 = 1,672,487 Tokens
Calculated over the last few days, the average daily consumption is approximately:
550,000 Tokens
At the current usage rate, the remaining 1,241,576 Tokens will last for approximately:
1,241,576 ÷ 550,000 ≈ 2.26 days
In other words, if I maintain the current translation intensity, this resource pack will last for about another two days.
However, this rate cannot be directly used to predict demand over the next three months.
The reason is that I am currently in the intensive phase of translating the blog’s historical articles into English. Consuming several hundred thousand Tokens per day is a temporary peak and not the normal level for the long-term operation of the website.
3. Why has Token consumption been so fast recently
Currently, there are three main types of tasks using GLM-5.2:
- Re-translating the blog’s historical Chinese articles into English;
- Translating newly published Chinese blog posts;
- Developing and testing the Go Tour Chinese translation workflow.
Previously, I estimated that there were still about 900 historical blog articles that needed to be re-translated into English. I have been continuously advancing this work, so the actual remaining number is now lower than that estimate.
Go Tour has about 110 pages that need to be translated and calibrated.
The current high consumption is fundamentally caused by a one-off historical task. Once all the historical articles have been processed, the daily tasks will likely only consist of:
- English translation of newly published blog posts;
- A few retries after translation failures;
- Minor corrections to published articles;
- Go Tour upstream content updates;
- Adjustments to prompts, glossaries, and validation rules.
By then, the average daily Token usage will likely drop significantly.
Therefore, I cannot simply extrapolate the continuous demand for the next three months based on the recent peak rate of about 550,000 Tokens per day.
4. The two GLM-5.2 resource packs currently visible
In the Zhipu trial flash sale section, two GLM-5.2 resource packs are currently visible:
| Resource pack | Tokens | Price | Validity period |
|---|---|---|---|
| GLM-5.2 special offer pack | 20 million | 39.9 yuan | 3 months |
| GLM-5.2 premium pack | 100 million | 189.9 yuan | 3 months |
In terms of capacity, 20 million Tokens is clearly more aligned with my actual needs.
However, when I clicked to purchase using my own account, the page immediately prompted:
We are sorry, your number of purchases has reached the limit.
The problem is that I have never purchased this current 20 million Tokens GLM-5.2 special offer pack.
What I bought previously was the 10 million Tokens GLM-5.1 70% off trial pack.
Later, I tested this with my friend’s newly registered account, and his account could still purchase the 20 million Tokens, 39.9 yuan GLM-5.2 special offer pack without any issues.
Considering the actual situation of both accounts, this purchase restriction is likely not calculated separately for a specific product, but is instead related to one of the following conditions:
- Whether the user is a new user;
- Whether the user has previously purchased a trial resource pack;
- Whether the user has already used their one-time special offer purchase eligibility;
- Whether the account meets the current campaign conditions.
At least based on the actual test results from my account and my friend’s account:
My old account can no longer purchase the 20 million Tokens special offer pack, while my friend’s new account still has the eligibility to purchase it.
5. What is the deal with the 20 million Tokens gifted for friend registration
There is another easily confusing “20 million Tokens” here.
Previously, I noticed that Zhipu had launched a campaign inviting new users to register. The campaign page prominently advertised:
Invite friends to register, and new users will receive 20 million Tokens.
I sent the invitation link to my friend, and he completed the registration.
When we first saw “20 million Tokens,” we both assumed it was a general-purpose resource that could primarily be used for GLM-5.2. However, after my friend checked the actual credited amount, he found that the 20 million Tokens had been split into several different resource packs.
Among them, the quota that could actually be used for general models like GLM-5.2 was only about 2 million Tokens. The rest of the quota was restricted to specific models like GLM-4.6V and GLM-4.5-Air.
In other words:
The promotion stated a total of 20 million Tokens, but the portion that can actually be used for GLM-5.2 is only about 2 million Tokens.
I have already documented this experience in detail in a previous article:
“Zhipu Invite a Friend for 20 Million Tokens Tested: Only 2 Million General Quota for GLM-5.2”
After seeing the actual credited amount, my friend felt that this “20 million Tokens” was not very meaningful to him.
He originally had no clear need to use the GLM API, and since the quota that could actually be used for GLM-5.2 was far lower than expected, he later decided to stop using the Zhipu platform.
Because my friend no longer plans to continue using this account, I have the opportunity, with his consent, to consider purchasing the 20 million Tokens, 39.9 yuan GLM-5.2 paid special offer pack through his account.
Here, it is necessary to distinguish between three different resources:
- The 10 million Tokens GLM-5.1 70% off trial pack I previously purchased, priced at 29.9 yuan;
- The 20 million Tokens combined resource my friend received for free upon registration, of which only about 2 million can be used for GLM-5.2;
- The 20 million Tokens GLM-5.2 paid special offer pack that my friend’s new account can purchase, priced at 39.9 yuan.
What I am now considering using is the third type of paid resource pack, not the combined resource my friend received for free upon registration.
6. How much would it cost to pay directly based on API usage
The Zhipu pricing page shows that the current API prices for GLM-5.2 are:
- Input: 8 yuan / million Tokens;
- Output: 28 yuan / million Tokens;
- Cache hit: 2 yuan / million Tokens.
Translation tasks involve both input and output.
The input content typically includes:
- System prompts;
- Translation requirements;
- Glossaries;
- Protected markers;
- Chinese titles;
- Chinese summaries;
- Chinese body text;
- Gutenberg or Go Tour page structures.
The output is mainly the translation generated by the model.
Since the prices for input and output are different, when using 20 million Tokens, the actual cost for pay-as-you-go billing depends on the proportion of output Tokens.
Estimating based on a total of 20 million Tokens for both input and output, without any cache hits:
| Output proportion | Input Tokens | Output Tokens | API pay-as-you-go cost | Resource pack price | Savings |
|---|---|---|---|---|---|
| 20% | 16 million | 4 million | 240 yuan | 39.9 yuan | 200.1 yuan |
| 30% | 14 million | 6 million | 280 yuan | 39.9 yuan | 240.1 yuan |
| 40% | 12 million | 8 million | 320 yuan | 39.9 yuan | 280.1 yuan |
| 50% | 10 million | 10 million | 360 yuan | 39.9 yuan | 320.1 yuan |
Therefore, for tasks like blog and tutorial translation that require generating a large amount of output text, purchasing the 39.9 yuan, 20 million Tokens resource pack can save approximately:
200–320 yuan
The exact amount depends on the input-to-output ratio and whether there are any cache hits.
7. Calculating savings based on a real API call
In a recent Go Tour translation call, the data I recorded was:
- Input Tokens: 1,669
- Output Tokens: 501
- Total: 2,170
- Cache Tokens: 0
Among these, the output Tokens accounted for approximately:
501 ÷ 2,170 ≈ 23.09%
If the subsequent 20 million Tokens maintain a similar input-to-output ratio, they would roughly consist of:
- Input: 15.3825 million Tokens;
- Output: 4.6175 million Tokens.
Calculating directly based on API prices:
The input cost is approximately:
15.3825 × 8 = 123.06 yuan
The output cost is approximately:
4.6175 × 28 = 129.29 yuan
The total is approximately:
123.06 + 129.29 = 252.35 yuan
Purchasing the resource pack only costs 39.9 yuan, so the savings would be:
252.35 − 39.9 = 212.45 yuan
This equates to a saving of about 84%.
This result only represents the input-to-output ratio observed in a single actual call. Variations in article length, prompt structure, and the number of failed retries will cause fluctuations, but it still serves as a useful reference sample for the current translation tasks.
8. Cache hits affect the savings amount
The cache hit price for GLM-5.2 is only 2 yuan / million Tokens, far lower than the standard input price of 8 yuan / million Tokens.
If a large number of repeated prompts hit the cache later on, the actual cost of direct pay-as-you-go billing will decrease, and the savings of the resource pack compared to pay-as-you-go billing will also shrink.
However, in the actual call I recently recorded:
- Cache Tokens: 0;
- No cache hits occurred.
Therefore, under the current translation workflow, I cannot assume for now that a large amount of input will be billed at the cache price.
At this stage, estimating using the standard input price is still closer to the actual situation I have observed.
9. Do all 100 million Tokens need to be used to be worth it
The 100 million Tokens resource pack is priced at 189.9 yuan, making the average cost per million Tokens:
189.9 ÷ 100 = 1.899 yuan
This unit price is much lower than API pay-as-you-go billing.
Therefore, from a pure cash cost perspective, you do not need to use all 100 million Tokens for it to be more cost-effective than pay-as-you-go billing.
Calculating based on different output proportions:
| Output proportion | API comprehensive unit price | Pay-as-you-go usage corresponding to 189.9 yuan |
|---|---|---|
| 20% | 12 yuan / million Tokens | Approx. 15.83 million Tokens |
| 30% | 14 yuan / million Tokens | Approx. 13.56 million Tokens |
| 40% | 16 yuan / million Tokens | Approx. 11.87 million Tokens |
| 50% | 18 yuan / million Tokens | Approx. 10.55 million Tokens |
Based on the approximately 23.09% output proportion from the previous real call, the comprehensive API price is approximately:
12.62 yuan / million Tokens
189.9 yuan is roughly equivalent to:
189.9 ÷ 12.62 ≈ 15.05 million Tokens
In other words:
If I can actually use about 15 million Tokens within three months, the total expenditure on the 100 million Tokens resource pack may already be lower than direct pay-as-you-go billing.
So, “not being able to finish 100 million Tokens” does not necessarily mean that purchasing it will result in an absolute loss.
However, this still does not mean I should buy it right now.
10. Why 100 million Tokens is still too much for me
Calculated at the recent peak rate of about 550,000 Tokens per day:
- 20 million Tokens would last for about 36 days;
- 100 million Tokens would take about 182 days to completely use up.
But the 100 million resource pack only has a validity period of 3 months, or about 90 days.
Even if the current peak rate is maintained for the next 90 days, I would only use:
550,000 × 90 ≈ 49.5 million Tokens
Theoretically, about half of the quota could still expire.
More importantly, the current daily consumption of 550,000 Tokens is a temporary peak caused by the intensive translation of historical articles.
Once the historical article translations are finished, daily usage may drop rapidly. By then, the quota actually used within three months could be far lower than 49.5 million Tokens.
Although from the perspective of “comparing with pay-as-you-go billing,” it might break even after using about 15 million Tokens, compared to the 20 million Tokens special offer pack available on my friend’s account, the 100 million pack still has clear issues:
- It requires a one-time payment of 189.9 yuan;
- The 20 million pack on my friend’s account only costs 39.9 yuan;
- There are not enough clearly defined tasks to consume 100 million Tokens within three months;
- A large amount of quota may expire;
- It easily leads to looking for unnecessary translation tasks just to avoid waste.
Although the average unit price of the resource pack is lower, it does not mean that a larger pack is always more suitable.
11. Usage will drop significantly after historical article translations are completed
Right now, the primary task is to re-translate the blog’s historical Chinese articles into English.
This is a one-off project with a clear end date.
Once all the historical articles have been processed, I will not continue translating dozens of old articles every day. Daily consumption will likely only include new articles and a few maintenance tasks.
Assuming a new blog post is published every day or every few days, even if a single article requires several thousand to tens of thousands of Tokens, the overall consumption will be much lower than the current phase of intensively processing historical articles.
Therefore, the risk of purchasing 100 million Tokens based on short-term peaks right now is:
Shortly after purchasing the resource pack, the historical translation tasks may already be finished, leaving the remaining two months without sufficient actual demand.
In this situation, although pay-as-you-go billing has a higher unit price per million Tokens, it does not have the issues of an expiration date and a large amount of balance expiring, making it potentially more suitable for subsequent low-frequency usage.
12. Should I add a third language to the blog
I have also considered whether to add another language to the blog in the future.
Currently, English traffic accounts for less than 15% of the website’s total traffic. Given that the English site has not yet generated high traffic, adding another language still carries significant uncertainty in terms of actual returns.
Based on my current rough estimate, even if a third language is added, the growth in advertising revenue would likely be within about 30%. This percentage is just a personal estimate and not a verified result.
However, adding a blog language requires work that goes beyond just calling the translation API.
It also requires handling:
- Over two thousand historical articles;
- Article titles, summaries, and body text;
- Category and tag systems;
- Multilingual article associations;
- New subdomains;
- CDN and caching;
- Sitemaps;
- Search engine indexing;
- Long-term synchronization of new articles;
- Manual corrections after translation failures;
- Compatibility after WordPress, theme, and plugin updates.
Therefore, I should not decide to add a language to the blog simply because I have a large number of Tokens on hand.
The correct order should be:
First confirm that adding a new language has value in itself, and then purchase resource packs based on clearly defined tasks.
Rather than first purchasing 100 million Tokens and then, to prevent the resource pack from expiring, creating a massive long-term task for myself in reverse.
13. Is Go Tour suitable for adding other languages
Compared to the entire blog, Go Tour is more suitable for multilingual experiments.
The main reasons include:
- There are only about 110 pages;
- The content structure is relatively uniform;
- The original English content is relatively stable;
- A page directory and translation status have already been established;
- A glossary has already started being built;
- Structure protection and validation workflows are already in place;
- Subsequent maintenance costs are significantly lower than those of over two thousand blog posts.
However, even if a complete language is added to Go Tour, the Token consumption is still very small relative to 100 million Tokens.
Therefore, Go Tour multilingual support can be a project development direction, but it cannot be the sole reason for purchasing the 100 million resource pack.
14. New languages present a “chicken or egg” problem
When choosing the next language for Go Tour, I cannot simply rely on the current traffic sources in Analytics or Search Console.
If the website currently only has Chinese pages, users searching in Japanese, Spanish, Korean, or other languages will naturally have a hard time finding this site through search engines.
Without pages in a corresponding language, it is difficult to generate impressions and traffic in that language.
This creates an obvious cold start problem:
Should I wait to see traffic in a certain language before deciding to translate, or should I provide that language first to have a chance at gaining the corresponding traffic?
Therefore, I cannot treat “visitors already appearing in a certain language” as a prerequisite for adding a new language.
A more feasible approach is:
- Select candidate languages based on the size of the external developer community and Go language usage;
- Check the quantity and quality of existing Go tutorials in that language;
- First translate 15–25 representative pages;
- Submit them for search engine indexing;
- Observe the impressions, search terms, and clicks generated by these pages themselves;
- Then decide whether to complete all approximately 110 pages.
This approach both breaks the cold start problem and avoids bearing the development and maintenance costs of a full language version right from the beginning.
However, even if I complete one or two new Go Tour languages in the future, it will still not be enough to naturally consume 100 million Tokens.
So, new languages should be driven by project value, not by resource pack balance.
15. What to note when purchasing the 20 million resource pack through a friend’s account
My friend has explicitly stated that he does not plan to continue using the Zhipu platform. Therefore, his new account and special offer pack purchase eligibility could potentially be used for my historical translation project.
From an economic perspective, this plan is very suitable:
- Capacity is 20 million Tokens;
- Price is only 39.9 yuan;
- At the current peak rate, it can be used for about 36 days;
- Compared to pay-as-you-go billing, it is expected to save about 200–320 yuan;
- The capacity is sufficient to continue processing a large number of historical articles;
- It will not leave as much idle quota as the 100 million pack.
However, the resource pack still belongs to my friend’s account, and actual API calls must use the API Key created under his account.
This means:
- Call logs are saved in my friend’s account;
- The resource pack balance is managed by my friend’s account;
- Translation content will be called through my friend’s account;
- If account verification, order, or resource pack issues arise, my friend’s cooperation is needed;
- If the API Key is leaked, the resource pack could be consumed by a third party;
- Whether this usage arrangement is allowed should be subject to platform rules or customer support responses.
If I ultimately adopt this plan, the safer approach is:
- Do not share my friend’s login password;
- Have my friend create a separate API Key specifically for this translation project;
- Store the API Key only in the local
.envfile; - Ensure that
.envis already ignored by Git; - Do not write the key into code, logs, commit records, or chat content;
- Immediately delete the corresponding API Key once the resource pack is used up or the collaboration ends.
16. My final choice
Taking into account the tasks that are currently clearly defined, I plan to proceed in the following order.
Phase 1: Continue using my remaining resource pack
My own resource pack still has about 1.24 million Tokens left.
At the current rate, it is expected to last for about two more days. Continuing to use the existing quota first can avoid resource waste and does not require immediately switching accounts and API Keys.
Phase 2: Consider purchasing the 20 million Tokens paid pack through my friend’s account
Of the 20 million Tokens my friend received for free upon registration, only about 2 million can actually be used for GLM-5.2. Since my friend does not plan to continue using the Zhipu platform, the GLM-5.2 special offer pack purchase eligibility on his account might be just right for my translation project.
It must be emphasized that what I am considering purchasing is:
The 20 million Tokens, 39.9 yuan GLM-5.2 paid special offer pack.
It is not the combined resource pack my friend received for free upon registration, nor is it the 10 million Tokens GLM-5.1 70% off trial pack I previously purchased.
If I confirm that there are no issues with the account and API Key usage arrangement, these 20 million Tokens will primarily be used to:
- Complete the English translation of the remaining historical blog articles;
- Complete the Go Tour Chinese translation;
- Continue calibrating prompts, glossaries, and validation workflows.
Phase 3: Switch to pay-as-you-go billing after historical translations are finished
Once all the historical articles have been processed, the daily Token usage will drop significantly.
If only a few new articles need to be translated at that point, direct billing based on actual API usage may be more flexible, and I would not have to worry about the resource pack expiring after three months.
Phase 4: Decide whether to add languages based on project value
A third language for the blog will not be fully launched for now.
For Go Tour, once the Chinese version is stable, I can select a candidate language for a small-scale pilot. I will wait until the pilot pages actually generate search impressions and traffic data before deciding whether to complete all pages.
I will not add a language just to consume Tokens, nor will I assume that a new language definitely has no demand just because there is no corresponding language traffic right now.
17. Summary
Through my order records, resource pack balance, and actual API call data, I ultimately reached the following conclusions:
- What I purchased in June 2026 was the 29.9 yuan, 10 million Tokens “GLM-5.1 70% off trial pack”;
- This resource pack was likely later expanded to support both GLM-5.1 and GLM-5.2;
- The GLM-5.2 special offer pack currently priced at 39.9 yuan with a capacity of 20 million Tokens is a different product;
- My old account cannot purchase the 20 million special offer pack, but my friend’s new account still can;
- Of the “20 million Tokens” my friend received for free upon registration, only about 2 million can actually be used for GLM-5.2;
- Because my friend does not plan to continue using the Zhipu platform, I have the opportunity to consider purchasing the paid special offer pack through his account;
- Compared to API pay-as-you-go billing, the 39.9 yuan, 20 million Tokens paid pack is expected to save about 200–320 yuan;
- Calculated based on the input-to-output ratio of a single real call, the savings are about 212 yuan;
- The 100 million Tokens do not need to be fully consumed to potentially be more cost-effective than pay-as-you-go billing;
- However, the current daily consumption of about 550,000 Tokens is only a short-term peak caused by the intensive translation of historical articles;
- After the historical articles are completed, daily usage will likely drop significantly;
- Therefore, 100 million Tokens is still too much for the tasks I have confirmed for the next three months;
- A third language for the blog involves a lot of work and should not be started just to consume a resource pack;
- Go Tour is more suitable for piloting a new language with a small number of pages to verify whether it is worth further investment.
A lower resource pack unit price does not mean that purchasing a larger plan is always a better deal.
What truly needs to be considered is always:
Within the resource pack’s validity period, how many tasks are actually clearly defined, able to be completed, and genuinely valuable.