[Google Gemini API's Sky-High 'Phantom Billing' Bug: Charges Persist After Cache Deletion, Zero Output Still Billed] According to monitoring by Beating, Google's AI developer forum has recently seen multiple urgent complaints about the Gemini API's billing system spiraling out of control. Several developers have reported facing massive abnormal charges due to underlying system bugs during normal usage—for instance, one individual was charged nearly 27,000 RMB within just 12 hours. Currently, Google's billing and technical teams are passing the buck to each other, with no official statement on a fix or a quick refund channel released. Investigations reveal two core bugs responsible for the developers' exorbitant bills: 1. The 'Phantom Cache' bug: After the context cache created via the API expires or is deleted, the front-end management list shows it as cleared, but Google's backend billing continues to 'idle charge' at a rate of thousands of RMB per hour. 2. The 'Thought Loop' trap: When tools like internet search are enabled, the model's 'thinking budget limit' fails, causing the model to enter an infinite reasoning loop while handling simple tasks. This burns through up to 64,000 tokens before timing out and crashing. Even if the final result is 'zero output' (no useful response returned), Google still charges the full amount, with thinking costs skyrocketing by 1,500 times. Due to a severe 32- to 72-hour delay in Google's cloud billing system and the lack of an automatic limit-breaking mechanism, developers often find themselves charged massive amounts before receiving any alerts. With official customer support deflecting responsibility and no direct responses on the forums, some affected developers have announced they will completely abandon using Gemini's context cache and inference models in production environments to mitigate financial risks. [Original Link]