双子座人工智能工作室的“上下文税”：一个10个单词的提示让我花费了121英镑
我一直通过AI Studio前端使用谷歌的Gemini 1.5 Pro来开发一个新平台。技术上讲，超过100万的上下文窗口对于“拼接”一个55,000行的代码库来说是一个颠覆性的改变。然而，我最近发现了一种掠夺性的计费结构，我称之为“上下文税”。
如果你使用AI Studio的用户界面，你可能会在没有任何警告的情况下面临巨额账单。
事情是这样的，我在细则中发现了一个关于英国/欧盟隐私的“专业提示”。
在AI Studio中，你一开始是免费的。你上传你的代码库（比如，70万tokens），并在达到每日配额之前免费工作。此时，用户界面会建议添加一个API密钥以“继续对话”。
陷阱：大多数用户（包括我自己）认为在添加密钥后，他们只会为增量使用（刚输入的10-100个tokens）付费。
现实：AI Studio的用户界面默认不使用上下文缓存。相反，它会在每一次“回合”中重新提交整个70万tokens的历史记录。关键是，尽管这段历史是“免费”积累的，但在每条后续消息中，你现在会按照付费层的费率“补税”。
消息1：（70万历史 + 10个新tokens）= 收费70,010个tokens。
消息2：（70万历史 + 20个新tokens）= 收费70,020个tokens。
在10-15个“回合”内，我被收费了170,000,000个tokens。总计？121.29英镑。
坦率地说，AI Studio的用户界面在这方面具有误导性，与谷歌的后续沟通也没有帮助。相反，他们更新了条款。
计费计数器：页面顶部的计数器保持增量（例如，从700,000到700,010）。它从不警告：“你下一条消息将花费5.00美元。”
批量计费延迟：谷歌将这些费用进行批量处理，并在数小时后报告。你无法实时看到“流血”。当你收到通知时，你已经亏损了100英镑。
证据缺口：在我的支持案例中，谷歌自己列出的“证据”仅显示了0.25英镑的SKU级别使用，但总小计为121.29英镑。他们无法解释账单的99%来自何处。
一个额外的伎俩：英国/欧盟隐私漏洞
在与这笔账单作斗争时，我深入研究了Gemini API附加服务条款（更新于2025年12月18日，正好在我提交争议后）。我发现了一个对我们在英国或欧洲经济区的人来说很有趣的“专业提示”。
根据新条款，谷歌将“付费服务”状态与花费金钱解耦。条款改为：如果你有一个与项目关联的活跃云计费账户，你的AI Studio使用在法律上被归类为“付费服务”，即使你在使用免费配额。
好处：“付费服务”有严格的非训练政策。如果你在付费层，谷歌不会使用你的提示来改进其模型。
漏洞：如果你在英国/欧洲经济区并将计费账户链接到你的项目，你的免费层使用默认获得企业级隐私（不进行训练）。
我的建议：
链接你的账户以保护隐私，但绝不要在用户界面中使用API密钥来延续达到免费配额限制的现有聊天。如果你达到免费配额限制，请停止。不要“升级”现有会话。如果必须继续，请开始一个新的会话，以避免对你的历史记录征收“上下文税”。
通过API使用上下文缓存：如果你确实需要在商业上处理一个100万tokens的代码库，完全避免使用AI Studio的用户界面。使用API并明确上下文缓存，以避免为同样的70万tokens支付100次费用。
我目前正在就缺乏“知情同意”进行争议，并对12月18日条款的追溯适用提出异议。
总结：谷歌AI Studio的用户界面对长上下文开发者来说是一个“财务生物危害”。一旦你插入密钥，它就会对你的免费历史进行补税，且没有实时费用透明度。
查看原文
I’ve been utilizing Google’s Gemini 1.5 Pro via the AI Studio front-end to develop a new platform. The 1M+ context window is, technically speaking, a game-changer for "stitching" together a 55,000-line codebase. However, I recently discovered a predatory billing architecture that I’m calling the "Context Tax."<p>If you use the AI Studio UI, you might be walking into a massive bill without a single warning.<p>Here is how it happened, and the UK/EU privacy "pro-tip" I found in the fine print.<p>In AI Studio, you start on the Free Tier. You upload your codebase (say, 700k tokens) and work for free until you hit the daily quota. At that point, the UI suggests adding an API key to "continue the conversation."<p>The Trap: Most users (myself included) assume that after adding the key, they will be billed for incremental usage (the 10-100 tokens they just typed).<p>The Reality: The AI Studio UI does not use Context Caching by default. Instead, it re-submits the entire 700k token history for every single "turn." Crucially, even though that history was built up for "free," you are now "back-taxed" for it at the Paid Tier rate on every subsequent message.<p>Message 1: (700k history + 10 new tokens) = Billed for 700,010 tokens.
Message 2: (700k history + 20 new tokens) = Billed for 700,020 tokens.<p>Within 10-15 "turns," I was billed for 170,000,000 tokens. The total? £121.29.<p>The AI Studio UI is, frankly, deceptive about this, and subsequent communication with Google was not helpful. Instead, they updated the terms.<p>The Token Counter: The counter at the top of the page remains incremental (e.g., it goes from 700,000 to 700,010). It never warns: "Your next message will cost $5.00."<p>Batch Billing Lag: Google batches these charges and reports them hours later. You don't see the "bleeding" in real-time. By the time you get the notification, you’re already £100 in the hole.
The Evidence Gap: In my support case, Google’s own itemized "evidence" showed only £0.25 of SKU-level usage, but a total subtotal of £121.29.
They couldn't explain where 99% of the bill came from.<p>A Bonus Gimmick: The UK/EU Privacy Loophole
While fighting this bill, I dug into the Gemini API Additional Terms of Service (Updated Dec 18, 2025, just after I submitted my dispute). I found a fascinating "Pro-tip" for those of us in the UK or EEA.<p>Under the new terms, Google has decoupled "Paid Service" status from spending money. The Clause instead reads: If you have an active Cloud Billing account linked to your project, your AI Studio usage is legally classified as a "Paid Service," even when you are using the free quota.
The Benefit: "Paid Services" have a strict non-training policy. Google does not use your prompts to improve its models if you are in the Paid Tier.<p>The Loophole: If you are in the UK/EEA and link a billing account to your project, you get Enterprise-level privacy (No training) on your free-tier usage by default.<p>My Recommendation:<p>Link your account for privacy, but NEVER use the API key in the UI to extend an existing chat which reaches the free quota limit. If you hit the free quota limit, stop. Do not "upgrade" the existing session. If you must continue, start a fresh session to avoid the "Context Tax" on your history.<p>Use Context Caching via API: If you actually need to work on a 1M token codebase commercially, avoid the AI Studio UI entirely. Use the API with explicit context caching to avoid paying for the same 700k tokens 100 times over.<p>I am currently disputing the bill based on the lack of "Informed Consent" for the transaction and the retroactive application of the Dec 18 terms.<p>TL;DR: Google AI Studio's UI is a "Financial Biohazard" for long-context developers. It back-taxes your free history the moment you plug in a key, with zero real-time cost transparency.