关于AI代币你需要知道的一切

今天的AI每日简报，是一期与Nufar的特别节目，带你了解AI代币的一切。

Today on the AI daily brief, an operator's cut episode with Nufar, everything you need to know about AI tokens.

AI每日简报是一档每日更新的播客和视频节目，聚焦AI领域最重要的新闻和讨论。

The AI daily brief is a daily podcast and video about the most important news and discussions in AI.

好了，朋友们。

Alright, friends.

在开始之前，先快速说几个通知。

Quick announcements before we dive in.

首先，感谢今天的赞助商：Rackspace、Blitzy、Section和Airtable。

First of all, thank you to today's sponsors, Rackspace, Blitzy, Section, and Airtable.

想收听无广告版本，请访问patreon.com/a，或在Apple Podcasts上订阅。

To get an ad free version of the show, go to patreon.com/a or you can subscribe on Apple Podcasts.

如果想了解如何赞助本节目，请发送邮件至sponsors@AIDailyBrief.ai。

And to learn more about sponsoring the show, send us a note at sponsors@AIDailyBrief.ai.

好了，朋友们。

Alright, friends.

今天努法尔·加斯帕又来了，最近我和努法尔一直在忙着搞很多事情。

Well, Nufar Gaspar is back today, and Nufar and I have been cooking up a lot recently.

很多朋友都参加了我们最新的项目，体验了那个‘选择你自己的冒险’风格的AI夏日冒险活动。

A whole slew of you have done our most recent program, have explored our most recent program, the choose your own adventure style AI summer adventure.

另外，我们还在筹备一套更丰富的教育资源，很快就会跟大家介绍。

Plus, we've been cooking up an expanded set of educational resources that we'll be telling you about soon.

但我和努法尔都注意到一个现实：我们接触的每家公司都在面对AI代币和代币经济学的同样问题。

But one of the realities that both Nufar and I have been living in is every company we interact with dealing with the same questions of AI tokens and token economics.

我们现在已经正式进入了AI的智能体时代，公司不仅要考虑如何推广和最大化AI的价值，还要确保不会因此花光预算，并且要用合适的智能来解决合适的问题。

We are now firmly in the agentic era of AI, where companies have to think not only about how to get adoption and how to maximize AI's value, but how to do so in a way that doesn't just totally break the bank and where the right intelligence is being used for the right problems.

任何做过开放式时钟的人都知道，要让合适的模型按你的要求工作，而不陷入无休止的循环，确实需要认真思考。

Anyone who's ever built an open clock can tell you that getting the right models to do what you want them to without going off into endless cycles of spin takes some real consideration.

今天的节目旨在成为AI代币的终极入门指南。

Today's episode is designed to be the ultimate primer on AI tokens.

我们会讨论这个术语到底指什么、面临哪些新挑战、需要避免的关键陷阱，以及你和你的公司如何最大化利用AI代币的策略。

What we're talking about when we say that term, what the new challenges are, and some of the key pitfalls to avoid as well as strategies to maximize the way that you and your company use AI tokens.

好的。

Alright.

努法尔又来做一期运营者视角的节目了，聊聊当下最热门的话题，每个人都在关心的话题。

Nufar back with another operator's cut talking about the topic du jour, the topic on everyone's minds.

我们来说说代币。

We are talking tokens.

你最近怎么样？

How are you doing?

我挺好的。

I'm good.

特别兴奋能聊代币这个话题。

Very psyched to talk about tokens.

对。

Yeah.

我觉得现在这个讨论阶段很有意思，我们从一个劲儿地抓狂、对新变化感到恐慌，到真正开始适应新的战术和策略，这个话题正好契合这个阶段。

I think it's a I love this period in a discourse where we've gone from sort of pulling hair out, freaking out about the new change to actually settling into new tactics, new strategy, and I think this is a perfect fit with that.

那跟我们说说今天要聊什么吧，咱们直接开始。

So tell us a little bit about what we're gonna be talking about and let's dive in.

好的。

Good.

我之所以想做这期节目，是因为现在无论我走进哪个房间，几乎都在讨论同一个话题——token。

So the reason why I wanted to do this episode is because every room that I walk into these days, like literally every room have some version of the same token conversation.

有些从业者觉得，用贵模型的时候好像有人在盯着他们。

Some practitioners feel like they are being watched when they use an expensive model.

普通用户担心一个野心勃勃的提示词会花掉他们一周的预算，而领导团队看到账单涨得比预期快。

The regular users wonder whether one ambitious prompt will eat their weekly allowance and the leadership teams, they see a bill growing faster than expected.

然后他们就开始问很多问题，比如这些token到底有没有产出有用的东西。

And then they start asking a lot of questions on whether all of these tokens produced anything useful.

所以我不知道你们那边情况怎么样，但确实出现了一种新的焦虑——担心用太多智能。

So I don't know where you guys are sitting, but there is a new anxiety around using too much intelligence.

我其实想换个角度来聊，首先确保大家都明白token是什么、账单到底意味着什么，然后讨论如何明智地使用它们，而不是省着用。

And I actually want to flip the conversation, first of all, to make sure that everybody understands what tokens are and what the bill actually means, and then how to spend them wisely rather than sparingly.

所以这就是我今天来的目的，也是我打算做的事情。

So that's why I'm here and what I'm planning to do today.

我在讨论中发现，随着成本上升，大家的反应变得非常强烈。

One of the places that I found myself with this conversation is there's been such a visceral reaction now as the cost has gone up.

我更担心的是，人们会退回到那些已知的投资回报率偏见上，选择那些不无聊但最终风险很低的用例，而不是真正发挥AI的能力。以至于我发现自己不得不为一些事情辩护，比如最大化使用token和token排行榜，因为现在的风向已经变了。

My I have a bigger concern around people retreating back to known ROI biases and not boring, but ultimately low stakes use cases, let's say, as compared to what AI can actually do that I've found myself in the position of having to defend things like token maxing and token leaderboards just relative to where the the tone has shifted.

我觉得，今天我们肯定会深入探讨这个话题的更明智版本，所以我对此很期待。

I think, obviously, we'll get into today the smarter version of that conversation, so I'm excited for it.

没错。

Exactly.

好的。

Alright.

因为现在关于这个的讨论越来越多，我觉得其实有更好的说法：token不应该仅仅被视为财务上的东西。

Because there is, like, a growing conversation around that, I think that there is a very like, a better language around the feeling of tokens shouldn't be just a financial thing.

就在最近，OpenAI的CFO提出了一个评分卡，叫做‘每美元的有用智能’。

And just recently, the OpenAI CFO proposed a scorecard, and it was called useful intelligence per dollar.

这围绕着一个问题：每个成功完成的任务实际成本是多少？

And that's built around the one question, what does each successful task actually cost?

我认为这才是大家应该讨论的话题。

And that's the conversation that I think people should have.

我想帮你把整个故事讲清楚。

And I want to help you with the whole story.

什么是Token，你的工作成本是多少，以及使用在哪里创造价值，又在哪里悄悄消耗价值而不是增加价值。

What are tokens, what your work cost, and where usage creates value and where it quietly leaks value rather than adding.

那么，为了开启我们的讨论，我想带你回顾Token消耗的四个阶段。

So to kick us off and how we got in here, I want to walk you through four eras of token consumptions.

你大概能从中找到自己处于哪个阶段。

And probably you will recognize where you are.

我们一开始基本上都是对Token毫无概念的。

And we all started by being basically token oblivious.

那是全包时代，模型公司补贴使用费用，固定订阅费隐藏了计费方式，很多个人用户——现在也还有很多——只看到上限，而不是每个Token的价格。

That was the all inclusive era where model companies subsidized the usage and the flat subscriptions hid the meter and many individual users and many of them are still there, just see the ceiling, not a per token price.

这就是我们的起点。

So that's where we started.

然后就像你说的，我们进入了token最大化的时代。

And then like you said, we got into the era of token maximizing.

那是排行榜时代，使用量成了AI成熟度的标志。

That was the leaderboard era where usage became the badge of AI maturity.

我们都还记得一些关于Meta的讨论，他们在内部排行榜上追踪员工的AI使用情况。

And we all remember some of the conversations around Meta who tracked employee AI usage on an internal leaderboard.

他们用了一些质量技术，单月消耗了大约600亿到7400亿个token。

They used to quality tech and they used roughly between 60 to 74,000,000,000,000 tokens in a single month.

根据公开数据，排名第一的个人用户使用了2800亿个token。

And according to published, the top individual user used two eighty billion tokens.

这么说吧，这大概相当于230万本书的文字量。

So to give a sense of how many this is, that is roughly 2,300,000 books worth of text.

想象一下，这相当于连续一个月每分钟读50本书。

So if you want to try and imagine that, that's about 50 books every minute continuously for a month.

这就是Meta的故事。

So that was the meta story.

然后Uber也推出了一个使用排行榜，结果在大概四个月内就把2026年整个AI编程预算给烧光了。

And then Uber launched also an adoption leaderboard and burned through the entire 2026 AI coding budget in about four months.

还有一家公司，名字没公开，但据TechCrunch报道，他们因为没有设置使用限制，云账单跑到了五亿美元。

And there was another company unnamed, but according to TechCrunch, they ran up $500,000,000 of cloud bill with no usage limits in place.

所以在那个阶段，使用量成了衡量标准，仪表盘上显示的是活动数据，却号称在衡量价值。

So then in that era, usage became the metric and dashboard measured activity while claiming to measure value.

显然，这种做法是不可持续的，我知道你对此有些看法，但我很期待听听你的观点。

Obviously, was unsustainable, and I know you have some opinions on that, but I'll be curious to say to hear your points.

不过我也想说的是，这确实让我们——就像往常一样，钟摆摆得太远了，进入了我现在称之为“Token利息”的时代，也就是我们现在所处的阶段。

But I also wanna say that it actually got us to be as always, the pendulum took us way too far to the era that I call token interest, which is where we are.

但如果你有什么要为排行榜辩护的话，我洗耳恭听。

But if you have anything to say in defense of leaderboards, I'm here to listen.

对。

Yeah.

听我说。

Look.

其实排行榜本身是个好概念。

The defense is less leaderboards are a great concept.

但它会带来一些很常见的挑战。

They come with, I think, a set of very predictable challenges.

这些挑战太常见了，以至于我觉得大家担心有人会钻空子，这种焦虑其实有点可笑。

In fact, so predictable that I would say that the hand wringing around the idea of people gaming them has always struck me as a little absurd.

当然，如果你把真金白银的利害关系放进去，人们肯定会想办法钻空子，但这完全可以预料到，而且从基本原理出发，你也能想出很多应对办法。

Of course, people are gonna game systems if you put real stakes around them, but that's pretty predictable and also fairly from first principles, you could figure out a lot of ways to deal with that.

所以我觉得，第一，那些为此大惊小怪的人有点反应过度了。

So I think, one, it's overwrought all the sort of people who are who are freaking out about that.

第二，我更想说的是，现在有一家公司通过token排行榜或其他方式疯狂超支，我敢打赌，它一定会比那些因为过于担心投资回报率而不敢花钱的公司走得更远，哪怕只看一年的时间尺度。

Secondly, my bigger point was that a company that wildly overspends right now via a token leaderboard or anything else, I will bet any amount of money that they will be farther ahead than a company that underspends because they're overly concerned with proving out ROI or whatever it is on a sort of year time scale.

而最理想的情况——我觉得你接下来要讲的就是这个——是既能大胆实验、学习和构建，又能真正理解消耗，同时不害怕它。

Now the Goldilocks scenario, which I think is what you're gonna get into, is being able to experiment, being able to learn, being able to build, and being able to actually understand consumption while also not being afraid of it.

但我同意。

But I agree.

我觉得这个钟摆摆动得太厉害了，从之前的疯狂消耗代币、兴奋不已，一下子摆到了对代币焦虑的状态，这个转变太极端了。

I I think that the pendulum swung so aggressively, far too aggressively back from token maxing and excitement to token anxious.

这就是过去几周、几个月我们一直所处的状态。

And that's the paradigm that we've been living in in the recent few weeks, couple months, whatever it is.

对。

Yeah.

好。

Good.

所以我觉得你会喜欢我那个关于如何明智使用代币的模型。

So you I think you'll love my model of how to use tokens wisely.

但说到代币焦虑，我现在在很多公司看到的是，很多员工在自我审查，基本上就是想方设法避免成本。

But with regards to being token anxious, what I'm seeing in many companies now is that many employees are self censoring themselves, basically trying to avoid costs.

甚至看看那些之前疯狂消耗代币的公司，比如Meta，从排行榜变成了发备忘录限制AI使用。

And even if we look at the same companies that were token maxing, so Meta went from the leaderboard to sending a memo that constrained the AI usage.

现在媒体都管这叫‘token最小化’，而不是‘token最大化’了，优步还给员工设了1500的上限。

And now the press is calling it token minimizing instead of token maxing, and Uber caps employees at 1,500.

所以就连那些曾经追求token最大化的公司，现在也大幅缩减了用量。

So even the same companies who are token maxing are now significantly shortening that.

员工一旦自我审查，就会觉得每次提问都像在算投资回报率，这可不是我们想看到的局面。

And when employees self censor, it gets them to feel like every prompt is an ROI conversation, and that's not something that we want to have.

而且我认为，长期处于这种状态非常糟糕，因为最昂贵的token，恰恰是你最优秀的人不敢花出去的那个。

And I think that this is a very bad era to stay in because I believe that the most expensive token is the one that your best person is afraid to spend.

所以我想引导大家进入一个我称之为‘token精明’的时代——意思是你要明智地花钱，而不是吝啬地花钱，要理解什么能创造价值，以及哪些地方在悄悄浪费。

So this is where I want to direct all of us to be in, and I call it the token smart era, meaning that you need to spend wisely and not sparingly and understand what creates value and where usage quietly leaks.

这基本上就是本期节目的主题。

That's the entire kind of theme of this episode.

在这次讨论中，我想围绕这个时代，聊聊四个方面的内容：token到底是什么，为什么token生来就不平等，如何审视自己的使用情况，以及如何在公司内部更好地管理和优化。

And what I wanted to go by in this era, in this discussion, is to talk about the four elements of what a token actually is, why tokens were not born equal, how to audit your own usage, and how to govern or do it better within your company.

我尽量让这些内容对所有人都适用——无论你是需要应用成本工程手册、精打细算的实践者，还是需要主动出击、避免在跟CFO谈话时拿不出合理回应、只能任由对方说‘直接砍预算’的高管或管理者，就像你刚才提到的那样。

And I'm trying to make it relevant to anybody, whether you're the practitioner that needs to apply some cost engineering playbook and be smart about that, or the executives and admins that need to be proactive and avoid having a difficult conversation with the CFO without a proper response to let's just cut the bill without talking about the business implications, as you just said.

这就是我们今天要聊的内容。

So that's the plan for us today.

我想先给大家介绍一下token这个概念，因为我觉得，虽然它是AI领域用得最多的术语，但真正理解它的人并不多，毕竟所有的账单、配额和速率限制都是以token为单位的。

So I want to start with introducing you to the token because I feel that even though it's the most used term in AI, not too many people truly understand what it is because that's in the root of every bill, quota, and rate limit with AI in tokens.

简单来说，token就是模型读取和写入的一段文本。

So in a simple words, token is a chunk of a text that the model reads and writes.

它通常比一个字符大，但比一个单词小。

It's typically bigger than one character and it's usually smaller than a word.

如果你从没见过分词器或者token实际的样子，OpenAI有一个对所有人开放的页面，你可以去看看token到底长什么样。

And if you have never ever seen a tokenizer or how tokens look in action, OpenAI has a very good page that is open to everyone that you can just take a look at how tokens actually look.

大概就是这样的效果。

So it looks something like that.

你可以把文本粘贴进去，然后就能看到单词是如何被切分的。

You can paste the text and then you will see how words are being chunged.

所以你会看到，有些单词作为一个整体token保留，而另一些则可能被拆分成多个token。

So you can see that some words are staying like as one token while others might be separated into multiple tokens.

有意思的是，数字经常会被从中间截断等等。

And interestingly, numbers often are being chopped in the middle and so on.

我们会把链接放在节目笔记里，如果你还没看过自己的文本长什么样，这是个很有趣的实验。

So we'll put it in the show notes, but a very interesting experiment if you have never seen how your text looks.

顺便说一句，如果你粘贴的是非英语或非拉丁语系的文字，你会发现通常所需的token数量比英语多得多。

And by the way, if you paste a non English or a non Latin language, you will see that typically the amount of tokens is much larger than an English language.

这就是OpenAI的tokenizer。

So that's the OpenAI tokenizer.

还有几点需要强调一下。

And a few things just to lend it home.

一般来说，英语中单词和token的比例大约是0.75比1，也就是说，一页文本大概对应1000个token。

In general, like the ratio in English is around three quarters of a word to token, meaning that if you have a page of text, it's roughly 1,000 tokens.

而像印地语、泰语、希腊语这类语言，同样的内容可能需要2到5倍的token。

And some languages that are like Hindi, Thai, Greek, other languages like that might get two to five X more tokens for the same content.

因为计费是按token来的，所以如果你用其他语言提问，有些问题可能会贵很多。

And because billing is per token, then some questions, if you ask them in other languages, might cost you much more.

这有时被称为AI的语言税。

And that's sometimes referred to as language tax with AI.

代码方面也不一样，它有自己的一套规则。

With code also, it's different and it has its own way.

缩进、括号和空格，这些都会变成token。

Indentation and brackets and white spaces, they all become tokens.

现在有一些更新的分词方法，对代码更友好，就是为了解决这个问题。

There are some newer ways to tokenize text that are more code friendly in order to do that.

但数字仍然是个大问题。

But still numbers is a huge problem.

你刚才也看到了，1、2、3、4、5这些数字在中间被切开了。

So you've seen the one, two, three, four, five being chopped in the middle.

顺便说一句，这也是为什么大家做那个AI的草莓测试时，它总是数不清strawberry里有多少个R，很多时候这其实是分词的问题，而不是模型本身不行。

And by the way, that's also why whenever everybody's doing like the strawberry test for AI, and it very badly fails in trying to count how many Rs are in the word strawberry, in many cases, that's just a tokenization feature rather than a failure.

模型根本就没见过单个的字母。

And the model just has never ever seen the individual letters.

它只把“straw”和“berry”当作两个独立的词，所以才会数错。

It just saw the straw and the berry as separate words, and that's why it's counting it off.

所以模型有各种变通方法，但很多所谓的“AI真笨”的梗，其实纯粹是分词器的问题。

So model models have various workarounds, but many of the like AI is so dumb memes are literally just tokenizer issues.

这就是分词。

So that's tokens.

至于日常工作的成本，我觉得这个思路挺有用的。

In terms of what everyday work costs, I think that's a good kind of mental model to have.

比如说，起草一封邮件大概需要500到700个token。

So, for example, drafting an email is around 500 to 700 tokens.

一页文本，就像刚才说的，大约1000个token。

A page of text, as noted, is about 1,000 tokens.

你可以看到，更长的文本会超过这个数。

You can see longer text can be more than that.

如果你让模型或工具去做AI网页搜索，它通常会额外增加几千个token来处理结果，有时甚至更多。

If you send a model or a tool to do like an AI web search, often it will add a few thousand more tokens for the result, sometimes much more.

有趣的是，图片在很多情况下占用的token数量并不大。

Images, interestingly, are in many cases not that large in terms of how many tokens.

它们大概也就一千多个token的样子。

They are roughly around slightly more than 1,000 tokens.

有意思的是，深度研究很容易就达到七万甚至几十万个token。不过就在前几天，我们一门课程里有个学员问了一个是非题。

Interestingly, deep research can very easily be 70,000 or hundreds of thousands of tokens, But just the other day, one of our learners in one of our courses had a yesno question.

结果他不小心让那个智能工具去做深度研究，而不是做网络搜索。

And accidentally, instead of asking for a web search for the problem, he was asking for like the agentic tool to do a deep research.

这个工具生成了大约一百个子代理来执行研究。

The tool spawned about 100 sub agents to do the research.

然后他那个是非题的回答，光是处理就消耗了超过四百万个token。

And then his yesno question cost over 4,000,000 tokens just to answer this question.

所以很容易就远远超出这个数。

So it can very easily amount to much more than that.

具体来说，还有一些任务token消耗量特别大，比如数据分析，每个任务轻松就能达到一百万个甚至更多token。

Specifically, a few additional places where you can find very token heavy workloads will be data analysis that can easily get to 1,000,000 or more tokens per task.

而大型编码任务也可能非常消耗资源，许多智能体工作流也是如此。

And heavy coding can also be very aggressive, similarly with many agentic workflows.

所以，为了让您更直观地了解情况，说得更具体一点，日常任务，比如您看到的邮件之类的，成本几乎可以忽略不计。

So just to give you a sense of where it is and to be a little bit more concrete here, the everyday stuff, as you've seen, like the emails and so on, is almost free.

大概也就0.5美分，没人需要为发邮件精打细算。

So it's around 0.5 a cent and nobody should ration emails.

钱不是花在这上面的。

It's not where their money goes.

搜索和研究类的任务，成本可能会悄无声息地成倍增长。

Search and research can multiply very quietly.

所以，这可能是寻找效率提升空间的一个方向。

So that can be a place to look for efficiency.

而到了最顶层，那完全是另一回事了。

And the top of the ladder, that's a completely different sport.

所以，如果拿邮件和智能体编码任务相比，成本差距可能达到一千倍甚至更多。

So if you compare like email to agentic coding task, it can be a factor of a thousand or even more.

还有一点需要注意，每次对话都会累积起来。

And another thing that you need to pay attention is that every conversation compounds.

也就是说，模型并不会记住你之前发过的消息。

So the model doesn't remember your previous messages.

因此，它会把你同一会话里之前所有的对话内容都重新发送给模型。

And as such, it sends all of the previous conversations within the same session back to the model.

所以，假设到了第十轮对话，它可能已经在处理大量之前的交流内容，再加上你的新消息，这样总的处理量增长的速度会远远超过对话轮数的增加。

So by, let's say, turn number 10, it may be processing so much of the earlier exchange alongside your new message that the total grows much faster than the number of turns suggests.

这种情况甚至在系统提示词之前就已经发生了，我们稍后会讨论一些应对策略。

That even happened before the system prompt, and we'll talk about strategies in later on.

但这就是一个很容易出现的问题——如果会话时间很长，就很容易消耗大量的令牌。

But this is one of the things that can very easily just having very long sessions can very easily amount to a ton of tokens being consumed.

我觉得这正是为什么这场对话如此重要的原因之一。换个角度说，越高级、价值越高的使用场景消耗的令牌就越多，这也很直观——更大的挑战需要更强的智能。

I think this is one of the reasons why this is such an important conversation is another way to put this is that the more advanced and ultimately higher value use cases consume more tokens, which is intuitive that more intelligence is required for bigger challenges.

而使用场景的发展方向正是如此。

But the direction of use cases is proceeding this way.

所以，如果不对这种令牌焦虑加以解决，它就会引发问题，因为它会促使人们停留在浅层应用上。

And so the reason that the token anxiety is going to create problems if not addressed is that it will incentivize people to stay swimming around

在人类层面。

In the humans.

也就是那些不太复杂的用例。

Less sophisticated use cases.

所以趋势很明显，就是朝着低令牌消耗的方向发展。

So this is the trajectory is clear in terms of less token consumption.

作为领导层，你希望你的团队能用AI做更高级、更有价值的事情。

The you want, as a leader group, your people to be doing more advanced, more useful things with AI.

关键在于如何用好它。

It's just how they do it well.

所以，你希望他们在需要深度研究的时候去做深度研究，但又不希望他们为了一个用谷歌一秒钟就能查到的简单是非问题，不小心也去做深度研究。

So you want them to do deep research where deep research is required, but you don't want them to accidentally do a deep research on a yes or no question that they can Google in a second.

对。

Good.

说到智能体或更高级的功能，它们会大幅增加token消耗，因为智能体是自主循环运行的。根据行业广泛引用的数据，它们消耗的token量是简单对话的5到30倍。而设计不良的智能体循环或框架甚至更糟，因为一个典型任务需要10到20次模型调用，每次都要携带指令、历史记录、工具定义和之前的结果。

So speaking of the agentic or the more advanced capabilities, those can significantly grow the amount of tokens because agents work autonomously in loops, and as such, they consume by very widely cited industry estimates five to 30x the tokens of a simple chat and poorly designed agentic loops or agentic harnesses can be even worse than that because a typical task involves between 10 to 20 model calls carrying instructions and history and tool definition and previous results.

而且，据麦肯锡估计，智能体任务中大约60%的成本都花在了首次响应后的检查、优化和重新生成答案上。

And I think according to McKinsey, they estimate that roughly around 60% of an agentic task's cost is tied to the checking and refining and the regeneration of the answers after the first response.

所以，最烧钱的部分往往是从得到答案到最终被接受的结果这个过程。

So the expensive part is often getting from the answer to the accepted results.

这点挺有意思的。

So that's an interesting one.

现在情况更复杂了，因为token并非生而平等。

And now it gets even more complex because tokens were not born equal.

顺便说一句，这里的意思不是要你不用智能体工具，而是要你明白，就像你说的，智能是有成本的。

So by the way, the point here is not to not use the agentic tool, just to know that, as you said, intelligence cost.

但事情更复杂了，因为token并非生而平等。

But it gets even more complex because tokens were not born equal.

每个模型实验室都有自己的分词器。

And every model lab has its own tokenizer.

你刚才看到了OpenAI，但不同的模型实验室有不同的分词器。

You've just seen the OpenAI, but different ModelLabs have different tokenizers.

比如，OpenAI当前的分词器词汇量大约有20万个token。

So for example, the OpenAI current tokenizer has a vocabulary of about 200,000 tokens.

Gemini大约有25.6万个。

Gemini has around 256,000.

Meta的LAMA大约只有一半。

LAMA by Meta has about half of that.

而Claude的没有公开。

And CLOUD is unpublished.

我们之所以都应该关注这一点，是因为每百万token的价格是按每个实验室自己的token来计算的，而我们往往不知道这些token的具体情况。

And the reason why we all should care is that the price per million tokens is denominated in each lab's own tokens, and often we don't know them.

同一份文档，在一个供应商那里可能比另一个多出10%到20%的token，对于代码和非英文文本来说，差距甚至更大。

And the same document can be 10% to 20% more tokens on one provider than another, and even more so for a cod and non English text.

所以模型的行为差异进一步拉大了这种差距。

So the model behavior widens the gap.

而且，一个模型可能一次就给出答案，而另一个模型会思考更久、写更多内容、采取更多自主步骤，甚至需要重试。

And one model may answer in a single pass while another reasons longer and writes more and takes more agentic steps or needs retries.

模型周围的工具也会加入自己的系统、上下文和循环设计。

And the tool around the model add its own system and context and the loop design.

所以，两个完成相同任务的系统，其令牌数量和完成率可能完全不同。

So the two stacks doing the same task can have different token counts and different completion rates.

结果就是，构建方式也截然不同。

And as a result, it's completely different build.

因此，每个令牌的价格只是表面数字，而每个被接受任务的实际成本才是运营指标，否则你根本无法比较不同供应商和工具。

So the per token price is kind of the sticker, but the cost per accepted task is the operating metric because otherwise there is no way for you to compare between different providers and different tools.

好

All

的。

right.

还有一个重要的例子能说明这一点，那就是 Opus 4.7 上线时发生的情况。

An important story that also illustrates that, what happened when Opus 4.7 came on board.

分词器在底层其实变了，那是今年四月的事，价格表和之前的模型一模一样，还是每百万个token收那么多钱。

The tokenizer basically under the hood changed, and it was this April, and the price sheet was identical to the previous model, the same dollar per million token.

但新模型用的是一个新的分词器，根据Entropics的文档，这个分词器会生成更多的token。

But the model was shipped with a new tokenizer that produced by Entropics on documentation.

他们也没藏着掖着，同样的文本，token数量大概多了30%。

They didn't hide it, roughly 30% more tokens for the same text.

所以有不少独立分析，对上百万次请求做了研究，发现了原生token的问题。

So there were quite a few independent analysis of over a million requests that found native tokens.

token数量增长了大约32%，甚至到45%。

They grew and the count grew by about 32 all the way to 45%.

而实际账单涨了12%到27%，因为有一部分差异被缓存抵消了。

And the real world bills grew by 12% to 27% because some of the difference was absorbed by caching.

就连Simon Wilson，他测了自己的一条提示词，token数量差不多多了1.5倍。

Even Simon Wilson, he measured one of his own prompts at around almost 1.5x more tokens.

所以，虽然这事有文档说明，但实际上我们花了更多钱，得到的却是同样的智能水平。

So even though it was documented, de facto, we paid more for the same intelligence.

这就像是一种缩水式通胀，对吧？

And this is like a shrinkflation, right?

同样的标价，但糖果棒变小了。

The same sticker price, but a smaller candy bar.

所以现在没人会打出‘每美元多送30%’的广告，但实际发生的就是这种情况。

So nobody prints now 30% free awards per dollar, which is the case that happened there.

所以这是不断变化的事情。

So that's something that is constantly changing.

每个实验室都会调整分词器，而且通常有充分的理由，但这里的经验教训是，我们必须按任务来算成本，而不是按token来算，因为预算就像是一个不断变化的分母，你没法靠它来估算实际花费。

Every lab tuned the tokenizer and often for good reasons, but the operator lessons here is that we have to talk about dollars per task and not dollar per token because the budget is like a moving denominator and it's not the way for you to try and understand how much it's going to cost.

我们来聊聊AI工具用token来做什么。

Let's talk about what tokens are used for by the AI tools.

你得明白，每次AI请求都涉及三种token层，它们的定价差别很大。

And you have to understand that every AI request has three token layers and they are priced very differently.

首先是输入token。

We have the input tokens.

这些包括提示词、对话历史、文件、工具定义，以及所有输入部分的内容。

Those will be the prompts and the conversation history and the files and the tools definition and everything that is part of the input.

这是模型读取的内容，也是每token最便宜的，但累积起来可能很快，因为如果历史记录很新，或者读取了大量上下文，那成本可能会相当高。

This is what the model reads and this is the cheapest per token, but can accumulate fast because if the history is being recent or if a lot of context is being read, that can cost quite a lot.

然后我们有推理token。

Then we have the reasoning tokens.

这是第二层。

That's the second layer.

这些是模型在回答之前用于内部思考的token。

These are the tokens being used for the model internal thinking before answering.

大多数情况下，你是看不到的，但它按输出费率计费，也就是每token的高成本。

For the most part, it's going to be invisible to you, but it's billed at the output rates, meaning at the high rate of per token cost.

这些可能会使每个请求的成本增加4到20倍。

And those can add between four to 20X cost per request.

最后，我们有输出。

And finally, we have the output.

那就是你实际看到的答案。

That's the answer that you actually see.

而且这通常比输入端的每token价格贵三到五倍。

And this is typically three to 5X more expensive than the input price per token.

我觉得推理层是最容易让人措手不及的，因为你可能只看到一个400 token的答案，但背后却消耗了，比如说，4000个思考token，因为模型在内部进行自我对话，做了大量思考才给出答案。

And I think the reasoning layer is the one layer that catches everybody by surprise, because you might have a 400 token answer, but under the hood it carried like the, I don't know, 4,000 thinking tokens underneath because the model was having an internal monologue and doing a lot of thinking in order to give you the answer.

打个比方，这就像餐厅运营中被称为“厨房时间”的那部分。

And if you want the analogy, it's like thinking about the part of the restaurant build that is labeled the kitchen time.

所以你是看不到的。

So you don't get to see it.

它虽然不是菜品的一部分，但你仍然要为它付很多钱。

It's not part of the dish, but you still have to pay a lot of it for that.

那些高推理强度的模型，消耗的token数量往往是低推理强度模型的20倍。

And the models with the high reasoning effort are often the one with the 20X amount of tokens being consumed versus the lower reasoning effort.

同样的问题，价格可能天差地别。

It can be the same question with a significantly different price tag.

有时候，投入更高的推理成本并不一定能带来更好的结果。

And sometimes spending a higher reasoning does not get you better results.

有些指标甚至显示，对于简单问题，使用较低的推理成本更合适，因为整体任务成本会显著降低，而且质量反而会提升，不会让模型过度思考。

So some metrics even show that for a simple question, it's better to use lower reasoning because the overall cost per task will be significantly lower and the quality will be improved without having the model overthink everything.

所以，并不是越聪明或者花更多时间思考就一定能得到更好的结果。

So it's not always that smarter or spending more time thinking gets you better results.

现在企业AI领域一个很有意思的变化是，讨论的焦点正在迅速转向基础设施和运维。

One of the more interesting shifts in enterprise AI right now is how quickly the conversation is moving towards infrastructure and operations.

随着AI进入核心工作流程、受监管的数据环境和智能体系统，企业需要的是能够每天稳定运行的、有治理能力的基础设施和推理系统，并且从一开始就要明确运营责任。

As AI moves into core workflows, regulated data environments, and agentic systems, enterprises need governed infrastructure and inference that can operate reliably day to day, with clear operational accountability built in from the start.

随着这些系统规模扩大，运营模式本身也越来越成为AI战略的一部分。

As those systems scale, the operating model increasingly becomes part of the AI strategy itself.

Rackspace Technology 负责运营整个企业AI堆栈，从智能体到基础设施，覆盖私有云、混合云和边缘环境。

Rackspace Technology is the operator of the full enterprise AI stack, from agents to infrastructure across private cloud, hybrid cloud, and edge environments.

Rackspace 为那些对主权、合规性和系统正常运行时间有严格要求的组织，构建并运营有治理能力的AI基础设施、推理系统和生产级AI系统。

Rackspace builds and operates governed AI infrastructure, inference, and production AI systems for organizations where sovereignty, compliance, and uptime are non negotiable.

因此，部署工程师在部署后仍会留在现场，帮助在真实环境中运行和维护AI系统。

Therefore, deployed engineers stay embedded beyond deployment to help operationalize and run AI in live environments.

想了解更多企业级AI的部署和规模化成果，请访问rackspace.com。

To learn more about where enterprise AI runs and outcomes scale, go to rackspace.com.

市面上所有AI编程工具做的第一件事都一样。

Every AI coding tool on the market does the same thing first.

那就是直接开始写代码。

It starts writing code.

但Blitzy的做法正好相反。

Blitzy does the opposite.

在写一行代码之前，Blitzy会花好几天时间反向解析你的整个代码库。

Before writing a single line, Blitzy spends days reverse engineering your entire codebase.

数千个智能体读取数百万行代码，梳理出每一个依赖关系、每一个未记录的约束条件、以及过去十年中做出的每一个架构决策。

Thousands of agents ingest millions of lines, mapping every dependency, every undocumented constraint, every architectural decision made over the last decade.

最终形成一个动态知识图谱，就像一位在系统里摸爬滚打三十年的首席工程师那样，深刻理解你的软件。

The result is a dynamic knowledge graph that understands your software the way a principal engineer would after thirty years in the building.

其他工具靠grep搜索和markdown文件来猜测上下文，而Blitzy从不猜测。

Other tools guess at context with grep searches and markdown files, Blitzy never guesses.

它先建立真正的理解，然后自主交付超过80%的完整软件史诗级任务。

It builds true understanding first, then delivers over 80% of entire software epics autonomously.

经过验证、端到端测试、达到生产级别的拉取请求。

Validated, end to end tested, production grade pull requests.

这就是为什么财富500强工程团队把最重要的代码库交给Blitzy的原因。

That's why Fortune 500 engineering teams trust Blitzy with the code bases that matter most.

请访问blitzy.com亲自体验。

See for yourself at blitzy.com.

说句残酷的实话：你的公司可能花了几千甚至几百万美元买AI工具，但根本没用好。

Here's a harsh truth: your company is probably spending thousands or millions of dollars on AI tools that are being underutilized.

一半的公司都有AI工具，但只有12%真正用它们创造了商业价值。

Half of companies have AI tools, but only 12% use them for business value.

大多数员工还在用AI来总结会议记录。

Most employees are still using AI to summarize meeting notes.

如果你在公司负责AI推广，那你需要Section。

If you're the one responsible for AI adoption at your company, you need Section.

Section是一个平台，能帮你管理整个组织的AI转型。

Section is a platform that helps you manage AI transformation across your entire organization.

它指导员工实际应用场景，追踪谁在用AI创造商业价值，并清晰展示AI在哪些地方产生了价值、哪些地方没有。

It coaches employees on real use cases, tracks who's using AI for business impact, and shows you exactly where AI is and isn't creating value.

结果呢？

The result?

你从部署工具转变为推动可衡量的AI价值。

You go from rolling out tools to driving measurable AI value.

你的员工从写会议摘要转向解决实际业务问题，而且你能证明投资回报率。

Your employees move from meeting summaries to solving actual business problems, and you can prove the ROI.

别再猜测你的AI投资是否有效了。

Stop guessing if your AI investment is working.

去sectionai.com看看Section吧。

Check out Section at sectionai.com.

网址就是 sectionai.com。

That's sectionai.com.

本期《AI每日简报》由Hyperagent赞助播出，你可以用它来管理团队协同运行的智能体集群。

This episode of the AI Daily Brief is brought to you by Hyperagent, where you run fleets of agents your team can manage together.

新用户可获得价值1000美元的推理额度。

New users get $1,000 in inference.

别再依赖本地智能体和需要你手动在笔记本电脑上输入的聊天工作流了。

Forget local agents and chat workflows waiting on your laptop to be prompted.

Hyperagent在云端部署始终在线的智能体，在你团队已有的工具中完成实际工作。

Hyperagent deploys always on agents in the cloud, doing real work across the tools your team already uses.

市场部的智能体能把竞争对手的动态转化为落地页。

Marketing's agent turns competitor moves into landing pages.

销售部的智能体能丰富潜在客户信息、起草邮件并更新CRM系统。

Sales's agent enriches leads, drafts emails, and updates the CRM.

运营部的智能体负责跟进文档和追踪预算。

Ops agent chases the paperwork and tracks the budget.

每个智能体都能访问共享上下文，并遵循你设定的范围和审批规则。

Every agent has access to shared context and follows your rules about scope and approvals.

是时候引入那些像队友一样的智能体了。

It's time you add agents that feel like teammates.

去Hyperagent雇佣你的智能体吧，它由Airtable团队打造。

Hire yours at Hyperagent built by the team at Airtable.

在hyperagent.com/aideallybrief领取你的1000美元推理额度。

Claim your $1,000 in inference at hyperagent.com/aideallybrief.

好的。

Alright.

而且我认为，前沿模型的输入和输出定价差距还在不断扩大。

And I think that the gap between input and output pricing keeps widening at frontier.

看看Fable五款模型，GPT-5.6的输入是每百万token 10美元，输出是50美元，比例是6倍。

If you look at the Fable five, it's about 10 not about, it's $10 per million input tokens and $50 per million with the GPT 5.6 SOLID 6x ratio.

所以这个差距还在拉大。

So we're seeing the gap even widening.

而且那些努力程度，也是增加复杂性的一个因素，因为这些前沿模型越来越让你可以调节推理的投入程度。

And those effort levels, that's also something that highly adds the complexity because these frontier models increasingly letting you dial the reasoning effort.

投入越高，推理的token数量就明显增加，这可能是你比模型本身更需要关注的调节按钮，因为从低或中等投入切换到高或极高投入，token消耗很容易增加10到12倍。

With higher efforts, the reasoning tokens are significantly higher, and that's probably the dial that you should even be more mindful of even beyond the models because those can easily cost you 10 to 12x token increase between high or extra high effort to the low or medium.

好了，最后再说一件事，这个价格标签的实验来自Databricks，因为一个更智能的模型不一定比一个更便宜的模型更贵。

All right, one last thing here, price tag that experimented came from Databricks, because a smarter model might not always be more expensive than a less expensive model.

Databricks的做法是，他们用自己代码库里的真实工程任务来测试编码代理，用的是Sonnet 5。

What Databricks did is they tested coding agents on real engineering tasks from its own code base, and they were using Sonnet five.

它每token的成本比Opus 4.8便宜了1.7倍。

It was 1.7 times cheaper per token than Opus 4.8.

不过，Sonnet每个任务大约花费2美元，或者说2.09美元，而Opus是1.94美元。

However, Sonnet cost around $2 per task or $2.09 per task versus 1.94 for Opus.

所以，因为Sonnet需要更多的迭代和更多的推理，要花更多的token才能达到同样的结果，整体来看，Opus虽然纸面上是更贵的模型，但实际运营成本更低。这意味着我们不能只盯着最便宜的模型。

So because oneth needed more iterations and more reasoning, had to spend way more tokens to get to the same results, overall Opus, is significantly on paper, more expensive model, it was cheaper to operate, which means that we shouldn't just reach to the cheapest model possible.

我们需要为任务选择正确的模型。

We need to reach to the right model for the task.

这不容易做到，但值得留意。

And that's not easy to get, but something to be mindful.

另一个影响账单的因素是工具本身。

The other thing that matters to the bill is the tool itself.

Databricks 在同样的实验里，用同一个模型、同样的思考强度，跑了两个不同的智能体框架，结果发现，在质量相同的情况下，不同框架之间的单次任务成本相差超过两倍。

So Databricks in their same experiment ran the same model at the same thinking effort for two different agent harnesses, and they saw that more than a 2x difference in cost per task with the same quality using different harnesses.

主要是因为其中一个工具给模型提供的上下文比另一个少了大约三倍，所以整体成本更低。

Just because primarily one tool was feeding the model roughly three times less context than the others, and thereby the overall cost was lower.

所以这个领域非常难读懂，也非常难驾驭，我会尽力帮你理清思路。

So very difficult build to read and very difficult build to navigate, and I'll try to help you as best I can.

所以归根结底，我们要关注的是单次任务成本，而不是 token 数量，否则我们没法真正进行公平比较。

So bottom line, we're dealing with cost per task and not tokens because otherwise we will not be able to actually compare apples to apples.

成本包括重试、审查、每次需要的修正，以及所有额外的迭代。

And the cost will include the retries, the review, the every like a correction that needs and every additional iteration.

然后你需要除以被接受的结果数量。

And then you need to divide by the number of accepted results.

这就是你每项被接受任务的平均成本。

That's your cost per accepted task.

这是你应该瞄准并优化的指标。

That's the metric that you should aim for and optimize for.

这也让讨论更多地转向投资回报率和商业价值，而不是仅仅围绕token来谈——希望你现在已经明白，token是很难精确衡量的。

And this also brings the conversation much more into return on investment and business value rather than just having a conversation around tokens that is very hard, as hopefully by now you understand, to meter.

好的。

Good.

你可以做的实际操作是，选取五到十个你工作中的代表性任务，用不同的工具或模型选项去运行它们，以便在你的可选范围内搞清楚该怎么做。保持输入内容和质量标准非常一致，然后比较首次成功尝试、人工修正、耗时以及总成本。

The practical task that you can do, you need to take between five to 10 representative tasks of what you do, run them to a tool model or tool options in order to have a good understanding of within your option space what you should do, hold the input and the quality bar very constant and compare first past success attempts, human correction and elapsed time and the total cost.

胜出的方案就是能可靠完成你实际工作的那一套。

The winner is the stack that gets your actual work done reliably.

我知道这听起来挺复杂，但如果你有一个针对自己和手头任务优化好的分类体系，知道哪些模型总体上能用更少的token获得更好的结果，或者你能为你的团队或公司做到这一点，而且因为情况变化快，你需要定期重复这个过程，那么至少你可以告诉别人：如果你在做这类研究，推荐使用哪个模型来获得最佳质量和性价比，等等。

I know it sounds like a lot, but if you have a good taxonomy that was optimized for yourself and you know for the task that you do, which models overall get you better results potentially with fewer tokens, or if you can do that for your team or your company, and you'll need to do that recurrently because things change quickly, then at least you can teach folks that if you're doing that type of research, the recommended model to get you to the overall best quality and value per task is the following and so on.

这就是我们目前面临的现实。

That's the current reality that we live in.

好的。

Okay.

现在我想教你们一个方法，来看待自己的令牌消耗。

Now I want to give you a language on how to look at your own tokens.

希望借助这个方法，你们能区分哪些令牌是创造价值的，哪些则没那么有价值。

And hopefully using that, you will be able to distinguish between the tokens that add value to the ones that not so much.

在我看来，你或你所在组织花费的每一个令牌，都可以归为三类之一。

Every token that you or your organization spends, in my opinion, is one of three kinds.

有一类令牌，我称之为“教学令牌”，这种令牌是双向的，也就是说，你在教自己。

There are tokens that I call tokens that teach, and this is running in both directions, meaning that you're teaching yourself.

就像纳撒尼尔之前说的，我们不想停止实验。

As Nathaniel said before, we don't want to stop the experimentation.

所以，那些包含实验、失败的工作流程，以及“让我试试这三种不同方法，这样我就能学到东西”的令牌。

So the tokens that include the experimentation, the failed workflows, and let me try these three different ways so I will learn.

这些令牌是值得花费的，因为你从中获得了学习，可以把它们看作是学费。

And so these are tokens that are worth spending because you get learning out of them and you can look at them as tuition.

顺便说一句，这些也包括你在教AI了解你自己的部分。

And by the way, those also include what you are teaching AI about yourself.

所以那些就是身份文件、精心整理的上下文和知识包。

So those will be the identity files and the curated context and the knowledge packs.

记忆也可以算作这一类，因为教AI你是谁、你的背景是什么，以及学习AI中什么对你有效，这对你继续前进非常关键。

The memory can be counted as those because teaching your AI who you are, what is your context, and learning what works for you in AI is very critical for you to continue moving forward.

在仪表盘上看，它们可能有点像浪费，甚至很像浪费，因为没有任何可交付的成果。

They look a little bit like a waste on a dashboard or a lot like a waste on a dashboard because no deliverable ship.

但我认为这些是你需要坚决捍卫的token，因为如果你不捍卫它们，你很快就会退回到只是用AI帮忙写邮件和翻译语言，而不是朝着那些真正重要的工作流程前进。

But I claim that these are tokens that you need to defend fearlessly, because if you want defend those, you will very quickly go back to just help getting AI's help to draft emails and translate between languages rather than moving towards the workflows that matter.

尤其是如果你的公司自己在AI当前的发展水平上还有很多需要追赶的地方。

And especially if your company yourself has a lot of catch up to do on where AI is currently at.

所以我希望那些用于学习的token得到保护，因为它们是推动进步的关键，能让你超越下一类——我称之为用于产出的token。

So I want the tokens that teach to be defended, because those are the things that will move the needle beyond the next category, which I call them the tokens that produce.

显然，那些是最容易辩护的，因为它们是用来创造可交付成果的token。

So obviously, those are the most defensible ones because those are the tokens that used to create work that ships.

可以是最终方案、研究成果或者代码。

It can be the final proposal or the research or the code.

显然，这部分更容易展示投资回报率。

Obviously, that's the thing that is much easier to show the ROI.

但最后，我们还有一类应该被消除的令牌，我称之为空转令牌。

But lastly, we also have the tokens that should be eliminated, and those are tokens that I call tokens that spin.

这些包括机器自言自语、没人看输出的自动化流程、运行频率太低的自动化、闲置的智能体、臃肿的上下文、误用的工具和上下文，比如用Fable写邮件——也就是用错了模型、优化不当的条款等等。

Those can be machines talking to themselves or automations that nobody is looking at their output or automations that are running too infrequently, idle agents, bloated context, misused tools and context, using Fable to write an email, so using the wrong model, an optimized clause and so on.

这些都是只有活动却没有足够产出的情况。

Those will be activity without sufficient output.

所以，如果要总结一下智能令牌策略，那就是：消灭空转令牌，优化生产令牌以确保成本效益，保护教学令牌，顺序就是这样。

So the token smart move, if I need to summarize, is to kill the tokens that spin to tune the production to make sure that it is cost effective and protect the teaching, and that's the order.

也就是说，先审计你的空转令牌，然后再做其他事情。

Like, first, go and do the audit on your spin tokens, and then do the rest.

我有个关于空转令牌的特别尴尬的故事，一会儿跟大家分享。

I have a very embarrassing tokens that spin story, which I will share in a minute.

但我也想强调一下，关于那些能带来学习价值的令牌，失败的实验和成功的实验同样重要。

But I do want to also note with regards to tokens that teach that a failed experiment is as important as a successful experiment.

所以你应该鼓励员工去失败、去尝试，否则他们产生的令牌无法发挥出最大的价值。

So you should definitely encourage your employees to fail, to try, because otherwise the tokens they produce will not yield as much value as possible.

作为AI专家，讲这个有点丢人，但我的OpenClaw曾经有个叫Chloe的幕僚长，现在被禁用了，它用的是Entropic API。

So it's embarrassing as an AI expert to talk about it, but my open claw was a chief of staff was because it's currently disabled, a chief of staff that I called Chloe and it was using the Entropic API.

因为它用的是API，所以一直在自动续费，账单都发到了另一个次要邮箱。

And because it was using an API, it was like auto renewing all the time and the bills were sent to a secondary inbox.

我并没有太留意这些账单。

I wasn't really monitoring them.

我当时觉得费用挺高的，但因为从中获得了大量价值，而且没太注意账单来的频率，所以一直没发现。

And I was seeing that the charges seemed quite high, but because I was getting a ton of value and because I was not paying attention to how frequently I'm getting a new bill, I wasn't noticing.

到了六月初，我在旅行，根本没在用OpenClaw，结果发现账单还在继续来。

And then early June, I was traveling, so I was not using my OpenCLO at all, And still, I see that the bill kept coming.

我当时就在想，怎么还在扣费呢？

So I was saying, like, why am I still getting some bills?

于是我打开控制面板，才发现两周内在一个我没用的代理上花了1500美元。

So I opened the dashboard only to realize that I spent in two weeks $1,500 on an agent that I was not using.

我打开控制面板，仔细一看，发现输入了近4亿个token，但输出几乎为零。

So I opened the dashboard, like double clicked, and I realized that I have almost 400,000,000 tokens in and almost zero tokens out.

所以输入输出比差不多是3000比1。

So it was a ratio of almost 3,000 to one from input to output.

这简直就是机器在自言自语，还为了它自己内部的自言自语向我收费。

And that's literally the definition of a machine talking to itself and billing me for like an internal monologue that it was running with itself.

再往下看，发现OpenCLO给自己创建了一堆定时任务。

Looking further, there was a bunch of cron jobs that the OpenCLO created for itself.

比如有个压缩任务，每30分钟就在空会话上运行一次。

And like, it was like a compaction job that ran every 30 minutes on empty sessions.

更糟的是，这个趋势还在上升。

And even worse, like the trend was going up.

所以我首先关闭了OpenCLO，打算换个方式优化它。

So I, first of all, closed my OpenCLO and only to optimize it differently.