Social media claim says ‘unlimited token’ trial exposed AI cost and management strain

An unverified post says a team of more than 20 tested intensive AI use under a US$1 million budget, reporting that cost, review work and organizational design became key constraints.

Published 8/11/2026, 8:57:27 AMLast updated 8/11/2026, 8:57:27 AMEditorial publication

The claim originates from a social-media poster who says they lead a team of more than 20 at a Silicon Valley AI training or inference company and received a US$1 million budget for an “unlimited token” experiment. The account cites a peak individual daily model cost of US$7,000, a team median of roughly US$2,000 per person per day, and increased fatigue from parallel sessions, reviewing AI output and frequent context switching.

Available material provides only broad context on growing enterprise AI usage and budget controls; it does not independently confirm the organization, invoices, model names, methodology, sample, or productivity findings. There is also a clear tension between “unlimited” use and a fixed US$1 million budget. At the stated median spend, a team of 20 or more could exhaust that budget quickly, so the account should be treated as anecdotal rather than generalizable research.

Source evidence 6

Source review does not guarantee permanent accuracy. Consider the original source, observation time, and later corrections.

Token调用量激增：从技术到商业的全方位重构 - U深研unifuncs.com · supporting

### 相关链接 Token太贵中国开源模型霸榜 ### 核心信息（时间标注：2026年2月，即1个月前） OpenRouter数据显示，2026年2月中国AI模型调用量三周大涨127%，首次超越美国，全球前五中占据四席，合计占比85.7%。一年前份额尚不足2%。最新的2月16日至22日的周榜单中，平台调用量排名前五的模型中，有四款来自中国厂商，包括MiniMax的M2.5、月之暗面的Kimi K2.5、智谱的GLM-5、DeepSeek的V3.2。这四款模型合计贡献了Top 5总调用量的85.7%。 ## Token消耗的底层逻辑变了 OpenRouter与a16z联合发布的《2025 AI使用报告》，覆盖超100万亿Token的匿名元数据。里面有一组数据极其关键——编程任务的Token占比从2025年初的11%飙到了50%以上，成为平台最大的单一使用品类。同时，Agent驱动的工作流（模型自主执行多步骤任务）产生的输出Token已超过平台总输出的一半。过去的问答式AI，一轮对话消耗几百到几千Token，用户不再提问，Token就不会再消耗。Agent模式下，机器可以持续在后台跑流程。 OpenRouter自身的数据：2月9日当周，平台处理了13万亿Token，1月第一周是6.4万亿，一个月翻了一倍。平台最新周Token总量12.1万亿，是一年前的12.7倍。 AI的使用模式从“对话型”切换到了“流程型”，Token消耗从“按次”变成了“按流量”。成本敏感度被急剧放大。 ## Agent模型组合背后的机会极端假设Agent全天候运行，每天吞掉数十亿Token。在这种负载下，价格差距就是生死线。 [...] hk.finance.yahoo.com www.esmchina.com www.moomoo.com m.bbtnews.com.cn www.hstong.com m.leiphone.com www.eet-china.com www.cww.net.cn www.beijingprice.cn www.idc.com introl.com unifuncs.com www.stashaway.hk www.ebrun.com m.huxiu.com view.inews.qq.com www.hstong.com www.