# Gemini 2.5 Model Bug Causing Massive Bills, Google Support Unresponsive to Core Issue **URL:** \ **Category:** Gemini API\ **Tags:** gemini-flash-2-5, billing\ **Created:** [July 1, 2025, 7:04am UTC](https://discuss.ai.google.dev/t/gemini-2-5-model-bug-causing-massive-bills-google-support-unresponsive-to-core-issue/91841 "2025-07-01T07:04:34Z")\ **Posts on this page:** 11\ **Page:** 1
**Author:** ![jbang](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/jbang/32/27914_2.png) [@jbang](https://discuss.ai.google.dev/u/jbang)\ **Post date:** [July 1, 2025, 7:04am UTC](https://discuss.ai.google.dev/t/gemini-2-5-model-bug-causing-massive-bills-google-support-unresponsive-to-core-issue/91841/1 "2025-07-01T07:04:34Z")
Hi everyone, I’m writing here out of sheer frustration and a desperate need for this critical issue to be properly addressed by Google. I’ve encountered a severe bug with the Gemini 2.5 Flash model that has resulted in an exorbitant bill, and despite providing clear evidence, Google Cloud Billing Support seems to be missing the point entirely. **The Problem: Gemini 2.5 “Infinite Loop” Output** My Google Cloud bill for June suddenly skyrocketed. After investigating, I pinpointed the cause: the **Gemini 2.5 Flash** model is occasionally entering what appears to be an infinite loop, generating an incredibly large volume of **invalid, repetitive, and useless text**. For instance, the model frequently outputs tens of thousands of characters filled with meaningless content like`



...` right up to the maximum token limit (65536 tokens). This isn’t productive output; it’s a clear **model bug** that’s directly inflating my token usage and, consequently, my bill. **My Attempt to Get Support & Google’s Response** I promptly contacted Google Cloud Billing Support, explaining the situation in detail. I even provided a specific API request log from **June 27, 2025, around 00:03 AM (UTC+8)**, clearly showing an instance of this runaway output. Here’s a summary of their initial reply: They acknowledged my charges, breaking down my input and output tokens. They confirmed these charges are “consistent with pricing” for Gemini 2.5 Flash. They suggested using Vertex AI Context Caching to reduce costs. **Why This Response Is Unacceptable** While I appreciate their quick response, their answer completely **missed the fundamental issue**. I’m not questioning the pricing of Gemini 2.5 Flash, nor am I unaware of cost-saving features like context caching. **My problem is not a misunderstanding of billing; it’s a critical bug in their model that is generating massive amounts of garbage output, for which I am being charged.** I’ve clarified this in my follow-up email, emphasizing that: This is a **model bug** , not a user error or a billing inquiry. The model is producing **invalid output up to its token limit.** **These costs, resulting from a bug, should not be my responsibility.** I specifically requested they **investigate the provided log sample** and my account’s usage for these anomalies. It feels like my concerns about a defective product are being met with advice on how to use the product more cheaply, rather than an acknowledgment of the defect itself. **My Plea** Google, you need to **urgently address this Gemini 2.5 model bug**. Charging users for a model’s faulty behavior that generates endless, useless tokens is simply unacceptable and unfair. I urge the Google Cloud team, especially those responsible for the Gemini models, to seriously investigate this issue. Please review the specific logs I’ve provided and all related usage data. We need a proper solution, which includes a refund for these unjust charges caused by the model’s malfunction. Has anyone else experienced similar issues with Gemini 2.5 generating excessive, invalid output? Your shared experiences might help highlight the severity of this problem. Thank you for your time and any support or advice you can offer. > Attachment link (API request log): [https://drive.google.com/file/d/1ZXcMVmoneE3BPgwcCdyby6jXt7CQ-og1/view?usp=sharing](https://drive.google.com/file/d/1ZXcMVmoneE3BPgwcCdyby6jXt7CQ-og1/view?usp=sharing) ![Snipaste_2025-07-01_14-44-32](https://d3qe71uytubmmx.cloudfront.net/original/3X/0/6/06cf7a3aff9734e5ea0fd3084bd5607486c350ed.png) ---
**Author:** ![Krish\_Varnakavi1](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/krish_varnakavi1/32/56962_2.png) [@Krish\_Varnakavi1](https://discuss.ai.google.dev/u/Krish_Varnakavi1)\ **Post date:** [July 1, 2025, 4:33pm UTC](https://discuss.ai.google.dev/t/gemini-2-5-model-bug-causing-massive-bills-google-support-unresponsive-to-core-issue/91841/3 "2025-07-01T16:33:21Z")
Hi @jbang, Welcome to the Google AI Forum! 🎊 🎊 Thanks for using Gemini models for your use-cases.. I understand your frustration about invalid output tokens causing high bills.. Can you share your Project ID to investigate further on token issues. ---
**Author:** ![Joe1](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/joe1/32/32121_2.png) [@Joe1](https://discuss.ai.google.dev/u/Joe1)\ **Post date:** [July 1, 2025, 6:16pm UTC](https://discuss.ai.google.dev/t/gemini-2-5-model-bug-causing-massive-bills-google-support-unresponsive-to-core-issue/91841/4 "2025-07-01T18:16:03Z")
This problem is not unique to gemini. It affects most LLM’s, especially smaller models. Gemini does support a [frequencyPenalty](https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/inference#frequencyPenalty) parameter, but the docs imply that its only supported in Gemini Flash 2.0. ---
**Author:** ![jbang](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/jbang/32/27914_2.png) [@jbang](https://discuss.ai.google.dev/u/jbang)\ **Post date:** [July 2, 2025, 12:58am UTC](https://discuss.ai.google.dev/t/gemini-2-5-model-bug-causing-massive-bills-google-support-unresponsive-to-core-issue/91841/5 "2025-07-02T00:58:31Z")
@Krish_Varnakavi1 Google Cloud Project ID: jbang-370306 ---
**Author:** ![Eric\_Creative\_Tech](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/eric_creative_tech/32/60273_2.png) [@Eric\_Creative\_Tech](https://discuss.ai.google.dev/u/Eric_Creative_Tech)\ **Post date:** [August 5, 2025, 9:04pm UTC](https://discuss.ai.google.dev/t/gemini-2-5-model-bug-causing-massive-bills-google-support-unresponsive-to-core-issue/91841/7 "2025-08-05T21:04:48Z")
I have this problem too. I started using it significantly less when I got an email saying I was 80% to $100 and still got a $600 bill. I’m unemployed so when I pulled $600 of the last $1100 of my 401k so I can eat and this broken bill swooped in and stole it from me and I cannot eat for a week and a half, I get really mad and decided to post my complaints all over LinkedIn on some important people related to the project. I don’t want to literally starve because of a known billing problem. ---
**Author:** ![Justine\_Chang](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/justine_chang/32/35929_2.png) [@Justine\_Chang](https://discuss.ai.google.dev/u/Justine_Chang)\ **Post date:** [August 26, 2025, 6:10am UTC](https://discuss.ai.google.dev/t/gemini-2-5-model-bug-causing-massive-bills-google-support-unresponsive-to-core-issue/91841/9 "2025-08-26T06:10:12Z")
I would like to add that I’m facing this as well. It keeps on thinking and thinking and thinking until it runs out of tokens, and I get charged for it. ---
**Author:** ![Chris\_Frederick](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/chris_frederick/32/62125_2.png) [@Chris\_Frederick](https://discuss.ai.google.dev/u/Chris_Frederick)\ **Post date:** [August 27, 2025, 3:28pm UTC](https://discuss.ai.google.dev/t/gemini-2-5-model-bug-causing-massive-bills-google-support-unresponsive-to-core-issue/91841/11 "2025-08-27T15:28:15Z")
I am also seeing this; it’s rather frequent. Sometimes retrying will be succesful; though often even trying 2 or 3 more times won’t fix it. This bug impacts reliability and is stopping me from using Gemini in a production setting. I am going to have to consider other vendors models if a fix or solution is not identified. ---
**Author:** ![Ryan\_L](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/ryan_l/32/37104_2.png) [@Ryan\_L](https://discuss.ai.google.dev/u/Ryan_L)\ **Post date:** [August 31, 2025, 12:09am UTC](https://discuss.ai.google.dev/t/gemini-2-5-model-bug-causing-massive-bills-google-support-unresponsive-to-core-issue/91841/13 "2025-08-31T00:09:48Z")
@Krish_Varnakavi1 I’ve just been hit by a huge bill too. I can see in my logs (between 26-29 Aug) that using flash 2.5 it looped for 39567 seconds (10 hours and 59 mins) on multiple calls!! It’s cost me 100’s of $s. My project id is gen-lang-client-0467589595 Can you tell me what I need to do to get this refunded? ---
**Author:** ![Krish\_Varnakavi1](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/krish_varnakavi1/32/56962_2.png) [@Krish\_Varnakavi1](https://discuss.ai.google.dev/u/Krish_Varnakavi1)\ **Post date:** [September 6, 2025, 12:04am UTC](https://discuss.ai.google.dev/t/gemini-2-5-model-bug-causing-massive-bills-google-support-unresponsive-to-core-issue/91841/15 "2025-09-06T00:04:08Z")
Thanks for the Project ID @Ryan_L. I have forwarded to appropriate team and following up.. Will let you know once I hear back from them. For refunds, please raise a request with [Billing Support](https://cloud.google.com/support/docs/get-billing-support). ---
**Author:** ![Tomohisa\_Ikuta](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/tomohisa_ikuta/32/62997_2.png) [@Tomohisa\_Ikuta](https://discuss.ai.google.dev/u/Tomohisa_Ikuta)\ **Post date:** [September 6, 2025, 10:42am UTC](https://discuss.ai.google.dev/t/gemini-2-5-model-bug-causing-massive-bills-google-support-unresponsive-to-core-issue/91841/16 "2025-09-06T10:42:26Z")
I also experienced a Gemini API billing bug, and I was suddenly charged $70,000. The official website states that refunds will be made within 48 hours, but the charge still remains in the admin panel. > **[Google AI Studio](https://aistudio.google.com/status)** > > The fastest path from prompt to production with Gemini I contacted billing support, but have not received a reply. I would appreciate it if someone from the technical team could look into this and provide a solution. Google Cloud Project ID: xroid-459200 Billing Support Case: 62587733 ![image](https://d3qe71uytubmmx.cloudfront.net/original/3X/1/5/155306c86df1da3c25ca1a3583a3600aee10db85.png) ---
**Author:** ![Jack\_Kirby](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/jack_kirby/32/35039_2.png) [@Jack\_Kirby](https://discuss.ai.google.dev/u/Jack_Kirby)\ **Post date:** [January 29, 2026, 7:33pm UTC](https://discuss.ai.google.dev/t/gemini-2-5-model-bug-causing-massive-bills-google-support-unresponsive-to-core-issue/91841/18 "2026-01-29T19:33:59Z")
I am also experiencing this issue. I contacted support and they are investigating and will let me know in 3-5 business days. my heart honestly drops and i get this intense worry that the costs will just keep skyrocketing.