URGENT: Huge cost cache increase issue

[junkx]

5 days after filling the billing support this is my response:

Ignoring all the details I gave (mutliple devs affected the same day, no increasae in usage, model used, SKUs report, huge increase of text output tokens from nowhere etc..)

For the love of god, can we please have a clear answer to this issue

[Jonathon_Habersberge]

I received a similar response unfortunately. Hope someone is looking into it.

[junkx]

Has anyone gotten an answer to our problem?

[junkx]

The only response i got form billing support is:

“Gemini and other generative AI models process input and output at a granularity called a token. Gemini API usage is primarily billed based on the amount of data processed, measured in tokens for both input and output. Different models are billed at different rates.”

And some generic things like go to AI studio dashboard to get your usage cost. lol

Waited a week for that answer. Crazy stuff…

Still paying 4x the price as before

[Liz2k]

Hey, Google Engeneering Team.

Critical system issue, previously discussed, has reoccurred. I need your attention.

Project ID: gen-lang-client-0787405973
SKU: E181-DFF8-56CF

1. Context of the previous incident (March 15-18, case ticket 69155729):

In mid-March, a “zombie cache” issue appeared in my project. The `cachedContents` API returned an empty list (`{}`), meaning I had zero active caches. Yet, the billing system was charging me ~$45/day for “cached text storage token hours.” Since I couldn’t delete what I couldn’t see, I completely disabled the Gemini API.

After I submitted the previous support ticket and waited a few days, your system seemed to self-correct. The historical billing graphs for those dates automatically cleaned themselves up and recalculated properly. This proves that the issue was a glitch on Google’s backend, not a user error.

SKU became from Mega hours (40$ per day) → 32K (for all period) which is correct.

2. The issue has returned (April 5 - April 7):

Unfortunately, this exact same bug happened again on April 5. A new set of “zombie caches” appeared.

I did everything right on my end: I verified via the API that my cache list was completely empty. However, the billing system once again showed that I was accumulating massive storage hours for empty space. This phantom cache consumed about $18 before I was forced to use the nuclear option again: I completely disabled the Gemini API on April 7, but it still growing after I did Gemini API Disable on account.

I have disabled Gemini API on 7 Apr (at 2pm UTC+4)

3. Why this is a critical Production Blocker:

Your previous agent mentioned: “The billing model is designed to charge for the time you use the storage.” I completely agree with this rule. However, it is fundamentally broken when my storage is completely empty, yet the billing system charges me hourly for “ghost” data every day every hour.

I am currently testing my application. I am now terrified to launch this project into production. With real users, the volume will be 10x higher. If a “zombie cache” gets stuck on your side again, it could silently drain $100+ per day. Furthermore, completely disabling the Gemini API to fix a Google backend glitch is not a viable workaround for a live production app with active users.

Or give me ability somehow to find and Delete those caches.

[Liz2k]

As I predicted, today the phantom billing hours for April 6th disappeared from the graph (reverted to normal), and tomorrow I expect April 7th to correct itself as well.

Please check attachments:

1. Current Issue 5-7 April - $23.

   

2. Current Issue 5-7 April (After 1 day) - $18

   

3. Current Issue 5-7 April (After 2 days) - $9

   

4. Tomorrow it will delete all fake cache. Which is great

However, please look closely at these facts:

1. Phantom constant growth: The billing graph suddenly started showing continuous cache usage for March 15-18 (at $45/day) and April 5-7 (at $9/day). It just started growing continuously every single day for absolutely no reason.

2. The API shows 0 caches: I used the Gemini API to delete all caches. The API returned a response confirming that NO caches exist (every hour every day 0 caches). Yet, the billing graph kept growing. I physically cannot understand what this cache is or how to delete it, because the system itself says it doesn’t exist.

3. Only the “Disable API” button works: On March 18 and April 8, the only action that stopped this phantom growth was completely disabling the Gemini API.

4. The 3-day retroactive rollback: During the 3 days after disabling the API, the billing system slowly deletes these phantom “cache hours” from the graph day by day.

My question is: how did the “actual usage” even exceed a million hours for a cache that the API says doesn’t exist?

I am very glad that the billing is slowly correcting itself back to the normal, real numbers. But I need to know why your system is suddenly generating $45/day or $9/day of fake “actual usage” out of nowhere.