Zhipu Invite-a-Friend 20 Million Tokens Tested: GLM-5.2 General Quota Is Only 2 Million

Recently, I came across a referral campaign on the official Zhipu BigModel page.

The main headline of the campaign pop-up read:

GLM models fully upgraded. Invite friends to register with real name and get Tokens

The pop-up also stated:

Claim up to a total of 200 million Tokens in resource packages

The most prominent content on the poster was:

Invite you to experience Zhipu GLM-5.2, the new flagship
New users register to get 20 million Tokens

The campaign page is located in the official Zhipu BigModel section. Because the page relies on JavaScript, you must log in via a browser to view the specific campaign pop-up and resource package details. (Zhipu AI)

After seeing this text, I naturally understood it to mean that after inviting a friend to register with their real name, the friend would receive a 20 million Tokens new user resource package primarily for GLM-5.2.

So, I sent the invitation link to a friend.

Unexpectedly, after my friend completed the registration and checked the actual resource packages credited to their account, they found the situation was quite different from what we had initially understood.

1. The First Impression of the Campaign Poster

Visually, the most prominent information on the campaign poster included:

• GLM models fully upgraded
• Zhipu GLM-5.2, the new flagship
• New users register to get 20 million Tokens
• Claim up to a total of 200 million Tokens

When an ordinary user sees this poster, it is easy to associate “GLM-5.2 new flagship” with “20 million Tokens.”

At least, that is how I understood it at the time.

I did not expect that these 20 million Tokens were actually a combination of resource packages for several different models, and that the portion truly usable as general model quota accounted for only a very small fraction.

2. The Resource Packages Actually Credited to My Friend

After completing registration, my friend went to the “Finance — Resource Package Management” page on Zhipu BigModel and saw a set of new user exclusive resource packages.

The three main resource packages related to Tokens were:

Resource Package	Quantity	Scope of Use
General Model Inference Resource Package	2 million Tokens	Applicable to basic models billed by Tokens
GLM-4.6V Model Resource Package	6 million Tokens	Only applicable to GLM-4.6V
GLM-4.5-Air Model Resource Package	12 million Tokens	Only applicable to GLM-4.5-Air

The three parts add up exactly to:

2 million + 6 million + 12 million = 20 million Tokens

In terms of total numbers, Zhipu did issue 20 million Tokens.

However, not all of these 20 million Tokens can be freely used for GLM-5.2.

Only 2 million Tokens actually belong to the general model inference resource package. The remaining 18 million Tokens are tied to GLM-4.6V and GLM-4.5-Air respectively.

This means the general quota accounts for only:

2 million ÷ 20 million = 10%

of the total Token count.

3. The Issue Is Not Whether 20 Million Were Issued

Strictly by the numbers, Zhipu did not issue fewer Tokens.

The three resource packages my friend received did indeed total 20 million Tokens.

The real point of contention is how the campaign poster guides user expectations.

The visual focus of the poster is “GLM-5.2 new flagship,” immediately followed by the emphasis on “new users register to get 20 million Tokens.” But upon receipt, only 2 million belong to the general model quota, while the remaining 18 million are restricted to GLM-4.6V and GLM-4.5-Air respectively.

Therefore, a more accurate description of this campaign would be:

New users can receive a combined resource package totaling 20 million Tokens upon registration, which includes 2 million general model Tokens, 6 million GLM-4.6V Tokens, and 12 million GLM-4.5-Air Tokens.

If the campaign poster had displayed this from the start, users would likely not have misunderstood.

However, the current promotional approach makes it easy for users to see GLM-5.2 first and then assume the 20 million Tokens are available quota for GLM-5.2.

This is why I think it feels somewhat like a “word game.”

4. Why My Friend Complained to Me

After checking the resource packages, my friend directly told me:

It feels like they’re just tossing a beggar a coin.

The reason he complained to me is also very simple.

I was the one who sent him the invitation link, and the main point I told him at the time was that he could get 20 million Tokens after registering, and that the campaign poster featured GLM-5.2.

Only after completing the real-name registration did my friend realize that the general model quota he actually cared about was only 2 million, and the other 18 million were not the GLM-5.2 general quota he wanted to use.

From my friend’s perspective, he might feel that he registered for “20 million GLM-5.2 Tokens,” only to end up with a tenth of that in general quota.

Although I did not intentionally mislead him, this experience still left me feeling a bit awkward.

I originally just wanted to share what looked like a good new user campaign with a friend, but instead of feeling like he got a good deal, my friend thought I had helped the platform run a poorly executed promotion.

5. Do 2 Million General Tokens Have Value?

Objectively speaking, 2 million general Tokens are not entirely without value.

For users who just want to test APIs, try out new models, or run a small number of requests, 2 million Tokens can still last for a while.

The 6 million GLM-4.6V Tokens and 12 million GLM-4.5-Air Tokens are not necessarily useless either. If a user happens to need a vision model or is willing to use GLM-4.5-Air, these exclusive quotas can still be useful.

The problem is whether the user actually needs these specific models.

If the user came specifically for GLM-5.2, then the other 18 million exclusive Tokens, despite their large quantity, may be of very limited practical value to them.

Therefore, when evaluating a resource package, you shouldn’t just look at the total Token count; you should also consider the following aspects:

1. Whether it can be used for the model you actually need;
2. Whether it is general quota or model-specific quota;
3. Whether there is an expiration date;
4. Whether there are limits on request counts, concurrency, and billing rules;
5. When multiple resource packages exist simultaneously, what order the system uses to deduct them.

12 million Tokens that cannot be used for the target model may not be more valuable to a user than 2 million general Tokens.

6. “Up to 200 Million” Also Needs to Be Understood Separately

The campaign pop-up also featured another very eye-catching slogan:

Claim up to a total of 200 million Tokens in resource packages

The keywords here are “up to” and “total.”

This does not mean that every newly registered user can directly receive 200 million Tokens.

Based on how the campaign page presents it, what my friend actually received after registering was a combined new user resource package totaling 20 million Tokens; the “up to 200 million” seems more like the cumulative upper limit obtainable across the entire invitation campaign.

Therefore, when seeing such promotions, you need to distinguish between several concepts:

• 200 million Tokens is the maximum cumulative quantity displayed by the campaign;
• 20 million Tokens is the total amount of the combined resource package a new user receives after registering;
• 2 million Tokens is the actual general model inference quota shown in my friend’s account;
• The remaining 18 million Tokens are restricted to GLM-4.6V and GLM-4.5-Air.

If you only look at the largest number on the poster without checking the resource package details, it is easy to conflate these figures.

7. Before Sharing Referral Campaigns in the Future, It Is Best to Confirm Resource Package Details First

After this experience, I probably won’t just look at the promotional poster when I see similar referral campaigns in the future.

A safer approach is to first find a user who has already completed registration, confirm the actual resource packages credited, and then explain to friends:

• How many Tokens are credited in total;
• How many are allocated to each model;
• How much the actual general quota is;
• Whether real-name verification is required;
• When the resource packages expire.

Especially when sending an invitation link to a friend, it is best to clarify the restrictions in advance.

For example, in this case, I could have directly told my friend:

You can get a total of 20 million Tokens after registering, but only 2 million of those are general model quota. Another 6 million are restricted to GLM-4.6V, and 12 million are restricted to GLM-4.5-Air.

If my friend understands the full picture before deciding whether to register, there won’t be a significant psychological gap.

8. My Final Thoughts

This campaign did not fail to issue 20 million Tokens entirely.

Based on my friend’s resource package dashboard, Zhipu did indeed make up the 20 million Tokens through three resource packages.

However, by displaying “GLM-5.2 new flagship” and “new users register to get 20 million Tokens” together, the campaign poster easily leads users to believe that these 20 million Tokens can be mainly or entirely used for GLM-5.2.

The actual situation is:

• The general model quota is only 2 million Tokens;
• The GLM-4.6V exclusive quota is 6 million Tokens;
• The GLM-4.5-Air exclusive quota is 12 million Tokens.

So, I won’t say that Zhipu failed to issue enough 20 million Tokens.

But I do think this promotional method exploits the discrepancy between the “total amount” and the “actual quota usable for the target model.”

There is nothing wrong with the numbers themselves, but the understanding users form is likely completely different from the actual credited results.

For the platform, this promotional method might improve registration conversion rates; but for the inviter and the invitee, if the actual resource packages fall too far short of expectations, the ultimate loss might not just be a registration experience, but also the friend’s trust in the inviter.