Gemini AI Studio 的“上下文税”：一个 10 个词的提示让我花了 121 英镑
2 分•作者: daitandojo•6 个月前
我一直在通过 AI Studio 前端使用 Google 的 Gemini 1.5 Pro 来开发一个新平台。从技术上讲，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变革性的功能。然而，我最近发现了一种掠夺性的计费架构，我称之为“上下文税”。
如果你使用 AI Studio UI，你可能会在没有任何警告的情况下收到巨额账单。
事情是这样的，以及我在细则中发现的英国/欧盟隐私“小贴士”。
在 AI Studio 中，你从免费套餐开始。你上传你的代码库（比如 70 万个 token），然后免费工作直到达到每日配额。到那时，UI 会建议你添加一个 API 密钥来“继续对话”。
陷阱：大多数用户（包括我自己）都认为，在添加密钥后，他们将为增量使用付费（他们刚刚输入的 10-100 个 token）。
现实：AI Studio UI 默认不使用上下文缓存。相反，它会为每个“回合”重新提交整个 70 万个 token 的历史记录。关键是，即使该历史记录是“免费”建立的，你现在也要为它按付费套餐的费率“补税”。
消息 1：（70 万历史记录 + 10 个新 token）= 计费 700,010 个 token。
消息 2：（70 万历史记录 + 20 个新 token）= 计费 700,020 个 token。
在 10-15 个“回合”内，我被计费了 1.7 亿个 token。总额是多少？121.29 英镑。
坦白地说，AI Studio UI 在这方面具有欺骗性，并且随后与 Google 的沟通也无济于事。相反，他们更新了条款。
Token 计数器：页面顶部的计数器保持增量（例如，从 70 万到 700,010）。它从不警告：“你的下一条消息将花费 5.00 美元。”
批量计费滞后：Google 会批量处理这些费用并在数小时后报告。你无法实时看到“出血”。等你收到通知时，你已经亏了 100 英镑。
证据差距：在我的支持案例中，Google 自己的分项“证据”仅显示了 0.25 英镑的 SKU 级别使用量，但总小计为 121.29 英镑。他们无法解释账单的 99% 来自哪里。
一个额外的噱头：英国/欧盟隐私漏洞
在与账单斗争的同时，我深入研究了 Gemini API 附加服务条款（更新于 2025 年 12 月 18 日，就在我提交争议之后）。我发现了一个有趣的“小贴士”，适用于我们这些在英国或欧洲经济区的人。
根据新条款，Google 将“付费服务”状态与花钱分离开来。该条款改为：如果你有一个与你的项目关联的有效云计费帐户，你的 AI Studio 使用将被法律归类为“付费服务”，即使你正在使用免费配额。
好处：“付费服务”具有严格的非训练政策。如果你处于付费套餐中，Google 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帐户链接到你的项目，你将默认获得企业级隐私（无训练）的免费套餐使用。
我的建议：
链接你的帐户以获得隐私，但永远不要在 UI 中使用 API 密钥来扩展已达到免费配额限制的现有聊天。如果你达到了免费配额限制，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：如果你真的需要在商业上处理 100 万 token 的代码库，请完全避免使用 AI Studio UI。使用带有显式上下文缓存的 API，以避免为相同的 70 万个 token 支付 100 次费用。
我目前正在就此账单提出异议，理由是缺乏对交易的“知情同意”以及 12 月 18 日条款的追溯适用。
总结：对于长上下文开发者来说，Google AI Studio 的 UI 是一个“财务生物危害”。它会在你插入密钥的那一刻对你的免费历史记录进行补税，并且没有任何实时成本透明度。
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.