有没有其他人遇到过阿里云AI API的欺诈性账单问题？
我想分享一个我在阿里云上遇到的令人担忧的案例，并询问是否有人遇到过类似的问题。
我在一个大约3200行代码的项目中测试了他们的Qwen LLM API（代码补全 + 聊天）。在2小时内，他们的系统声称我消耗了1000万个令牌。
为了让大家更好地理解这个情况：
即使将整个3200行代码重复作为上下文发送（每次调用最多约15万个令牌），也需要大约67次调用才能达到1000万个令牌。
我的日志显示的请求量远远没有达到这个水平。
然而，他们的账单立即扣除了约4.19美元，并留下了一个“待处理”的7.18美元未付款项。
当我向支持团队提出这个问题时，他们的回应是：
“这可能是因为像VSCode这样的插件在长上下文中自动补全”或者“每次调用都携带历史上下文，因此令牌消耗得更快。”
这个解释连基本的合理性检查都无法通过：
携带上下文并不会神奇地将使用量乘以100倍。
我使用过的任何LLM API（OpenAI、Anthropic、Mistral、Groq）都没有出现令牌消耗与实际使用如此脱节的情况。
将责任归咎于VSCode插件是一种推卸责任，而不是根本原因分析。
他们没有承担责任，而是试图用200美元的优惠券来让我沉默，而这些优惠券仅限于阿里云使用。我拒绝了。此后，我要求：
退还他们已经扣除的4.19美元。
清除那张虚假的7.18美元“未付款”账单。
删除我的账户并确认我的代码/日志被删除（出于隐私考虑）。
他们拒绝提供（2）和（3）。
这引发了两个更大的问题：
信任：如果他们的账单如此不透明，任何开发者如何能够安全地大规模使用他们的API？
隐私：如果他们乐于对客户的令牌使用进行误导，那么他们在后台对上传的代码和提示做了什么？
我已经开始在各个平台（Reddit、LinkedIn）上记录和发布证据。下一步：联系记者。
是否还有其他人遇到过阿里云的类似账单或支持失败？
我很想听听那些详细测试过他们API账单的工程师的意见。
——
警告开发者：在使用阿里云进行AI工作负载时要极其谨慎。
不透明的账单 + 无能的支持 = 危险的组合。
查看原文
I want to share an alarming case I’ve been dealing with on Alibaba Cloud, and ask if others have faced similar issues.<p>I ran tests against their Qwen LLM API (code completion + chat) on a project that’s ~3,200 lines of code.
Over a period of 2 hours, their system claimed I had consumed 10 million tokens.<p>To put this into perspective:<p>Even if the entire 3,200 lines were repeatedly sent as context (≈150k tokens max per call), it would take ~67 calls to reach 10M tokens.<p>My logs show nowhere near that volume of requests.<p>Yet their billing immediately deducted ~$4.19 and left me with a “pending” $7.18 outstanding charge.<p>When I raised this with support, their response was:<p>“This may be because of plugins like VSCode auto-completing with long contexts”
or
“Every call carries historical context, so tokens add up faster.”<p>That explanation doesn’t pass even basic sanity checks:<p>Carrying context doesn’t magically multiply usage by 100x.<p>No LLM API I’ve used (OpenAI, Anthropic, Mistral, Groq) has ever produced token consumption so detached from real usage.<p>Suggesting that VSCode plugins are responsible is a cop-out, not a root cause analysis.<p>Instead of accountability, they tried to silence me with $200 in coupons valid only for Alibaba Cloud. I refused. I’ve since demanded:<p>Refund of the $4.19 they already charged.<p>Clearing of the fraudulent $7.18 “outstanding” bill.<p>Deletion of my account and confirmation my code/logs are erased (privacy concerns).<p>They refuse to provide (2) and (3).<p>This raises two much bigger issues:<p>Trust: If their billing is this opaque, how can any developer safely use their API at scale?<p>Privacy: If they’re happy to gaslight customers on token usage, what are they doing with uploaded code and prompts behind the scenes?<p>I’ve started documenting and publishing evidence across platforms (Reddit, LinkedIn). Next step: reaching out to journalists.<p>Has anyone else faced similar billing or support failures with Alibaba Cloud?
Would love to hear from engineers who’ve tested their API billing in detail.<p>—<p>Warning to devs: be extremely cautious using Alibaba Cloud for AI workloads.
Opaque billing + impotent support = dangerous combination.