ChatGPT 5.5：变革人工智能对话

SpaceX正在告诉投资者，他们打算自己研发GPU。

SpaceX is telling its investors it wants to start building its own GPUs.

这事儿可不小，因为这清楚地说明了算力市场现在有多紧张。

This is a massive deal because I think this is really just showing us how tight the compute market has gotten.

你想想，一家火箭公司居然开始琢磨自己造芯片了。

You have, like, a rocket company that is looking at building its own silicone.

还有，微软和Anthropic联手，把Claude直接集成到他们的安全编码系统里；同时，OpenAI也在悄悄给联邦机构介绍他们的新模型GPT-5.5的网络安全版本。

We also have to talk about Microsoft teaming up with Anthropic to plug Claude directly into their secure coding stack, and OpenAI is quietly briefing federal agencies on the new GPT 5.5 cyber model.

现在整个AI和网络安全融合的趋势越来越明显，而且在好多地方都在同时发生。

There is a whole kind of AI meets cyber story that we are seeing, and this is playing out in a lot of different places.

另外，谷歌这周在云峰会上发布了一个非常重磅的智能代理平台。

Also, Google dropped a very serious agent platform on their cloud event this week.

他们在模型花园里已经有两百多个模型，还推出了专门针对OpenAI和Anthropic的新工具。

They have a 200 plus model in their model garden and new tooling aimed exactly at OpenAI and Anthropic.

我觉得这个特别有意思，因为谷歌之前一直被说在智能代理这场赛跑里掉队了，所以这明显是他们的一记反击。

I think this one is very interesting because Google has been criticized for falling behind on the agent race, and so I think this is basically their very clear response to that.

我们还有一条来自Meta的消息。

We also have news from Meta.

他们要裁掉大约一万八千人，占员工总数的百分之十左右。

They're doing layoffs of about 18,000 people, which is about 10% of their workforce.

扎克伯格已经很直白地表示，这波裁员就是因为AI。

Zuckerberg has pretty openly pointed out that AI is the reason for this.

就在同一天，微软也宣布了有史以来第一个自愿离职计划。

Microsoft, on the exact same day, also announced its first ever voluntary buyout program.

这个计划覆盖大约八千七百名员工。

This is targeted at about 8,700 employees.

我觉得这事儿可不只是Meta一家的问题。

So I don't think this is just a meta story.

这反映的是AI正在彻底重塑全球巨头公司的薪资结构。

I think this is a story about how AI how AI is starting to really reshape payroll at the biggest companies on the planet.

接下来，咱们深入聊聊OpenAI昨天发布的GPT 5.5。

Then we're gonna do a little bit of a deep dive on OpenAI shipping GPT 5.5 yesterday.

这距离GPT 5.4发布才刚过去六周。

This is just six weeks after GPT 5.4 came out.

性能指标简直吓人。

The benchmarks are wild.

定价策略特别有意思。

The pricing story is very interesting.

我觉得这会为很多智能体AI树立新标杆，今天节目里我来好好拆解一下。

I think this is gonna set a new bar for a lot of agentic AI, so I'm gonna unpack all of that on the show today.

在进入正题前，先跟大家聊聊AI Box。

Really quick, before we get into the stories, I wanted to tell you about AI Box.

这工具我现在几乎天天用，真的是不用搞一堆订阅就能直接用上AI的最省事方法。

This is the tool that I personally use pretty much every day now, and it's honestly one of the easiest ways to actually get value out of AI without having to figure out a dozen different subscriptions.

AI Box让你在一个平台就能访问80多个不同的AI模型。

AI Box gives you access to over 80 different models on of AI all in one place.

ChatGPT、Claude、Gemini、图像生成、视频工具，全都有。

So you get ChatGPT, Clawd, Gemini, image models, video tools.

所有功能都集中在一个地方。

All of it is in the same place.

对我个人来说，最实用的是，我只要说清楚想要什么，就能自己搭自动化流程。

And the part that is really useful for me personally is that I can build my own automations by just describing what I want.

我不是程序员。

So I'm not a coder.

根本不用去搞那些稀奇古怪的流程图。

I don't have to, you know, use some weird flowchart.

你只要打字说，比如‘帮我生成一个通讯稿工具，能自动出图、写内容，再去网上找资料，然后给我打包好’，它真就能给你办了。

You basically can type in something like, you know, create a newsletter generator that, you know, spits out different images and different content, and it goes and researches the web to gather information, and it can create that for you.

每个月只要8.99美元，就能用上所有这些工具，比单独订个ChatGPT还便宜。

It's all $8.99 a month to get access to all of these tools, and it's less than a subscription to something like ChatGPT.

如果你感兴趣，赶紧去看看。

So if you're interested, go check it out.

描述里有个链接，点进去就能到aibox.ai。

There is a link in the description to aibox.ai.

我想先聊的 biggest 新闻是 SpaceX。

So the biggest story that I wanted to kick off with is SpaceX.

这周有报道称，SpaceX 告诉投资者，他们计划进行一笔所谓的巨额资本支出。

There's reports that came out this week that SpaceX is telling their investors, they're planning what they're calling a, quote, substantial capital expenditure.

其中他们提到的一个方向，是自己制造 GPU。

And one of the things that they've floated is manufacturing their own GPUs.

他们没说明天就干，但光是把这事写进PPT，就说明问题不简单。

Now they're not saying that they're going to start this tomorrow, but I do think the fact that they're even putting this on a pitch deck tells you something really interesting.

现在我们正卡在一个巨大的算力瓶颈上。

Right now, we have this massive compute bottleneck.

这事儿真不是闹着玩的。

It's very real.

现在每个认真的AI公司都在抢NVIDIA的芯片，等货时间长得离谱，价格也贵得吓人。

Every serious AI company right now is fighting for NVIDIA chips, and the, you know, the wait times and the prices are super, super brutal.

SpaceX 自己的AI计算需求可不小。

SpaceX has a ton of in house AI workloads.

他们有星链系统。

They have Starlink.

他们有星舰的航电系统。

They have Starship Avionics.

他们有自动驾驶项目。

They have Autonomy.

他们还有旗下的xAI公司，整个Grok项目都用着同样的基础设施。

They have the entire Grok project over at x AI that shares their infrastructure.

所以他们的算力需求大得惊人。

So they have a ton of demand.

说实话，我觉得这恰恰说明，现在全球最大的科技公司，已经不再完全信任GPU的供应链了，光当客户不够了。

And honestly, I think this is what is tell what this is telling us right now is that we've basically reached the point where the biggest tech companies in the world no longer trust the GPU supply chain enough to just be customers.

Meta早就开始自研芯片了。

Meta is already doing custom silicon.

谷歌也有自己的TPU。

Jill Google has TPUs.

亚马逊有Tranium芯片。

Amazon has Tranium.

苹果正在自研推理芯片，我觉得SpaceX也加入这个行列了，因为他们确实看到了需求。

Apple is building their own inference chips, and I think SpaceX is basically joining that list because they they see the need.

我觉得这完全说得通。

This makes a lot of sense to me.

他们能不能做成是另一个问题。

Whether they pull it off or not is definitely a different question.

芯片制造太难了。

Chip fabs are very hard.

马斯克手头的事已经多得忙不过来，但方向很明显。

Elon's plate is already very full, but I think the the direction's pretty clear.

AI的算力竞争，现在已经不输模型竞争了。

The compute side of AI is becoming just as competitive as the model side.

微软刚宣布要整合先进模型，顺便提一下，他们正在测试一个叫Anthropic Claude Mythos的预览版。

Microsoft has just announced that they are integrating advanced models, including, by the way, what they're calling Anthropix Claude Mythos preview.

所以这个Claude的Mythos是个超级强大的网络安全AI模型，能找出各种代码里的漏洞，他们担心这玩意儿会被黑客拿来用。

So Mythos from Claude is kind of this huge cybersecurity AI model that can find bugs in anything, and it's got, like, all of these different issues, and they're concerned that it could be used by hackers.

所有这些风险。

All of that.

对吧？

Right?

但微软其实已经在整合这个模型了，不过目前只是预览版。

Well, Microsoft is actually integrating that, but it's just a preview.

它直接嵌入在他们的安全编码框架里。

It's directly inside of their secure coding framework.

目的是让Claude在微软的开发工具里负责威胁检测、漏洞扫描和事件响应。

And the idea is that Claude gets used for threat detection, vulnerability scanning, and incident response inside of Microsoft's dev tooling.

与此同时，OpenAI也在向美国联邦机构、各州政府，还有五眼联盟的情报伙伴们介绍他们的一款叫GPT-5.5 Cyber的模型。

And at the same time, OpenAI has been briefing US federal agencies, state governments, and also the Five Eyes intelligence partners on a version of their model called GPT 5.5 cyber.

我不是故意唱衰，但真觉得这事儿有点儿跟风——就是看Anthropic搞出个Mythos hype，大家立马跟着蹭热度，说什么‘天啊这模型太猛了’，还得专门去给政府做汇报。之前政府还跟Anthropic闹得挺僵，现在倒好，突然开始主动找他们谈了，就因为这模型厉害到能一键干掉全球所有代码库。

Now I don't wanna be skeptical, but it just feels so much like opening eyes trying to tag on to what Anthropic did with the the kind of the mythos hype, where there was like, oh my gosh, the model's so crazy, and they had to go, like, debrief the government, and, like, the government had all this beef with Anthropic, but now they're, like, talking to Anthropic because they have this super hardcore model that could, like, you know, wreck all the code bases in the world.

所以说，这波操作给Anthropic带来了巨大的公关红利。

So, anyways, huge PR for Anthropic.

他们的股价涨了。

Their stock rose.

他们的信誉也一下子飙升了，大家对他们的看法完全不一样了，感觉OpenAI就是想蹭这波热度。

Their credibility, everyone's view of them definitely got a massive boost, and it feels like OpenAI is trying to just, like, tack onto that.

瞧瞧，我们也有个超厉害的网络安全模型。

Like, oh, look.

而且我们这个模型也超牛的。

We got a really good cyber model too.

总之吧。

And so anyways.

管他呢。

Whatever.

我不是说他们没那么强，但这一连串操作，怎么看都像在搞营销套路。

I'm not saying it's probably not just as capable, but it it does feel like a lot of this is marketing kind of ploy vibes.

不过话说回来，OpenAI在华盛顿办了一场活动，邀请了大概50位网络安全防御领域的从业者，看得出来他们在这件事上砸了不少精力。

But, anyways, OpenAI held a Washington event for about 50 cyber defense practitioners, so they're obviously putting a lot of energy into this.

行吧。

Okay.

谷歌这边这周刚在拉斯维加斯办完他们的大型云服务峰会。

Google just had their big cloud event in Las Vegas this week.

他们正式推出了Gemini企业代理平台。

They rolled out the Gemini Enterprise Agent platform.

我觉得谷歌这步棋，本质上就是想在企业级AI智能体赛道上和OpenAI、Anthropic掰掰手腕。

I think this is basically their attempt to take a swing at OpenAI and Anthropic in the enterprise for the AI agent race.

首先，他们马上就要推出模型花园（Model Garden）功能，能接入200多种不同的模型。

I think first, access to over 200 models through Model Garden is something that they're gonna be rolling out.

这样一来，企业用户就不用被死死绑在Gemini这一棵树上了。

So enterprises aren't gonna be locked into just Gemini.

其次，他们还发布了Gemini 3.1 Pro、支持图像处理的Gemini 3.1 Flash，还有全新的音乐生成模型Lyria 3。

The second thing is that they launched Gemini 3.1 Pro, Gemini 3.1 Flash Image, a brand new music model called Lyria three.

他们推出了 Gemini 4 的开源版本，还有 Gemma 4，同时也搭建了一整套管理平台。

They launched Gemini four on the opens, or Gemma four on the open source side, and they also built a whole management layer.

比如一个共享收件箱，AI代理可以在里面更新进度、交接任务，还能跟人类协作。

So a shared inbox where AI agents post progress updates, hand off tasks, coordinate with humans.

这挺有意思的，因为他们真正在为AI代理搭建一套能实际干活、干得好的基础设施。

This is interesting because they're building a lot of this infrastructure for the agents to actually go and accomplish tasks and do it well.

我觉得谷歌在底层技术上早就领先了，但在AI代理这一块，他们一直没把事情真正落地，没让企业觉得真有用。

In my opinion, Google has had the research in in this infrastructure forever, but what they've looked kind of not so not so good at on the agent side is actually pulling it off in a meaningful way for enterprises.

所以之前呢，OpenAI大发布，Anthropic的Cloud Copilot也抢走了一大堆企业客户，尤其是五百强公司。

And so because of that, we have this you know, OpenAI had their huge release, and then we've had Anthropics' Cloud Cowork in the past that has swept up a lot of enterprise customers and and Fortune 500 buyers.

看来这片战场马上就要更热闹了。

It looks like we're gonna have more competition in that space now.

接下来，Meta刚给全体员工发了封备忘录，宣布要裁掉大约10%的员工。

Next up, Meta has just sent a memo to all of their employees announcing that they're cutting about 10% of their workforce.

员工。

Workforce.

这涉及到大约八千人。

This represents about 8,000 people.

他们的裁员将在5月20日全面生效。

Their cuts are gonna all roll out on May 20.

他们还取消了原本要招聘的六千个岗位。

They're also scrapping a bunch of plans to hire, for 6,000 open roles.

所以算上那些还没填上的空缺，总的人数缩减其实接近一千四百个职位。

So total, the reduction in headcount is closer to about 1,400 positions if you're looking at kind of the unfilled ones.

Meta给出的裁员理由是AI。

And the reason what they're that they're giving for this over at Meta is AI.

扎克伯格年初就告诉过大家，2026年将是AI彻底改变我们工作方式的一年。

Zuckerberg basically told everyone this at the start of the year when he said 2026 would be, quote, the year AI starts to dramatically change the way we work.

当然，这事儿也可能是个好事。

Now, you know, that could be a really positive thing.

但很多人觉得这事儿挺让人不安的。

A lot of people view this as sort of an ominous thing.

但不管怎样，我们还是到了这一步。

But here we are regardless.

Meta 对2026年的资本支出指引是1150亿到1350亿美元。

Meta's 2026 CapEx guidance is a 115 to a 135,000,000,000.

这比去年的720亿大幅上涨了。

This is up massively from 72,000,000,000 last year.

他们正在大幅增加AI基础设施的投入，同时削减人力成本。

So they're spending way more on AI infrastructure, and they're gonna be cutting back on people.

花更多钱在AI模型上，却少雇人来操作它。

So spending more for the AI models and less on the actual people running it.

就在同一天，微软也宣布了类似举措——他们不叫裁员，而是搞了个所谓的‘自愿退休回购计划’，明显是在避讳‘裁员’这个词。

I think on the exact same day, we also have Microsoft that has announced that they also are doing not really layoffs, but they're trying to avoid, I think, the word layoffs, but it is the it's what they're calling the first, quote, voluntary retirement buyout program.

据CNBC和彭博社报道，这可是微软51年历史上头一回这么做。

This is the first time that the company has ever done this in their fifty one year history according to CNBC and Bloomberg.

这个计划适用于年龄加上工龄总和达到70岁及以上的员工。

This offer applies to any employees whose age plus the years that they've been there total seventy or more.

比如说，一个52岁、在微软干了18年的员工，就符合条件。

So for example, if there's, like, a 52 year old who's been working at Microsoft for eighteen years, they would qualify for this.

公司现在大概有12万名员工。

And the company has about a 120,000 employees.

按这个比例，自愿离职计划可能覆盖7%左右，也就是将近8750人。

With this, the buyout could reach potentially 7% of that, which is about 8,750 people.

这还没算上去年夏天那9000人的裁员。

That's on top of the 9,000 layoffs that they did last summer.

这么一来，微软明显是在大力削减人员规模。

So with all of this, Microsoft is obviously trying to cut down their headcount.

他们现在想用一种没那么生硬的方式，不像以前那样大规模裁员，但方向其实没变。

They're trying to do it right now in a bit of a less abrasive way than kind of a mass layoffs, but the direction is is basically the same way.

再算上Snapchat刚裁了1000人，占他们总人数的16%，一周前Exxon也干了同样的事，而且他们的CEO说，Aon现在超过65%的代码都是AI生成的。

And if you also add into that Snapchat just cut a thousand jobs, which was 16% of their workforce, and a week ago the ex for the exact same reason, and their CEO said Aon now generates over 65% of their code.

所以综合来看，听起来可能有点吓人，好像不是什么好事。

So with all of that, I know it sounds like, I don't know, it could be like scary, could be bad news.

我觉得有意思的是，我们现在正目睹科技巨头们首次大规模因AI引发的白领岗位重组。

What I what I do think that is interesting here is that we're kind of watching the first real wave of AI driven white collar restructuring in a lot of the biggest tech companies.

这可不是什么‘AI迟早会抢走你的工作’的空话了。

You know, this isn't kind of like the AI is gonna take your job someday.

今天，这些变化其实已经在真实发生了。

I think today we're seeing a lot of these shifts happen.

但关键的一点是，那些真正懂AI、会用AI提升工作效率的人，反而被留了下来。

But I do think an important story here is that people that are deeply embedded and skilled with AI and using it to leverage their roles better are the ones that are actually being held onto.

这一点特别重要。

And I think this is an important part.

有了AI，公司用更少的人就能干更多的活。

Companies are able to do more with less people, due to AI.

所以，如果你在公司里能充分利用AI，把自己变成最高效的那个，说实话，你真没必要担心——不光是因为你不太可能被裁。

So if you are someone inside of your organization that is, you know, taking full advantage and becoming the most productive as possible, I don't think you have anything to fear truly, not just because maybe you wouldn't get laid off.

就算真被裁了，只要你精通这些AI工具，凭你的本事，找下一份工作也轻松得很。

Because even if you did, it will be easy for you to find something else if you are really an expert using your tools these AI tools for your expertise.

所以啊，这其实就是现实，没办法。

So that is that's kind of I mean, this is just the world.

我想表达的就是，实话实说，现状就是这样。

So I'm I'm trying to say you know, I'm trying to tell it like it is.

这世界本来就是这样的。

This is kind of the world we live in.

但真正能充分利用这些AI工具的人，才是未来能脱颖而出的。

But I think that the people that are really taking advantage of these AI tools are the ones that are gonna be successful going forward.

说白了，如果你在听这个播客，那你大概率就属于那个最上层的群体——认真对待这一切的人。

And if you're you know, news flash, if you're listening to this podcast, you're probably in that top bubble, that top tier that's taking all of this very seriously.

明白了。

Okay.

咱们今天聊个大新闻：GPT-5昨天刚发布，这事意义重大，原因有很多。

Let's get into the big story today, which is that GPT five just got released yesterday, and this is a big moment for a bunch of different reasons.

首先，GPT-5.44才刚推出六周而已。

First, I think speed GPT 5.44 dropped just six weeks ago.

才六周过去，就出了个全新的模型。

So six weeks later, we have an absolutely, new model.

他们推出新版本的速度真快，这说明OpenAI现在压力有多大。

The cadence is really fast on how they're pushing this out, and I think this tells you a lot about the pressure that OpenAI is under right now.

他们现在同时被Anthropic和谷歌逼得够呛。

They have pressure both from Anthropic and Google.

我觉得，最让他们头疼的就数这两家了。

Those are the two people I think that are pushing them the hardest.

Anthropic的年收入已经冲到三千亿美元了，谷歌也刚推出了企业级智能代理平台。

Anthropic just hit a $30,000,000,000, you know, annualized revenue run rate, and Google just shipped their enterprise agent platform.

OpenAI现在可再也没法喘口气了。

OpenAI doesn't really get to take any quarters off anymore.

他们必须不停推新功能，才能勉强跟上Anthropic的步伐。

Right now, they have to be shipping aggressively if they're going to keep up with Anthropic.

还有一件事，我觉得特别能说明问题，挺有意思的。

And one, like, I guess, story that I think is really important on why this is so critical for them this is kind of interesting.

其实我没怎么看到有人报道这件事。

I actually don't see a lot of people reporting on this.

不过有个二级市场，员工可以在那儿卖掉自己手里的公司股份。

But there is the secondaries market where employees are able to sell their shares in an organization on a secondary market.

这中间有不少规矩和复杂的地方，但大致就是这么个机制。

And there's a bunch of regulation and tricky stuff with that, but that's kind of the basis of what happens there.

在二级市场上，OpenAI的估值挺有意思的，现在大家买股份的价格，有时候能反映出未来股价的走向。

On the secondaries, OpenAI's valuation, what people are buying the shares for right now it's kind of interesting because it's a indicator of kind of where the stock will go sometimes.

不过说回来，现在OpenAI的股份卖价是8500亿美元。

But anyways, what OpenAI's shares are selling for right now is $850,000,000,000.

如果你想买OpenAI的股份，这就是当前的估值。

That's the valuation if you want to go buy OpenAI shares.

Anthropic的股份呢？

Anthropic shares?

一万亿美元。

1,000,000,000,000.

所以Anthropic的估值现在更高了。

So Anthropic is being priced and valued at higher.

有可能，这部分估值不是基于他们现在的状况，而是大家看好的未来前景。

Possible, you know, possible part of that is maybe not even where they are where they are today, but where people can see them going in the future.

这一切都给即将上市的OpenAI和Anthropic制造了巨大压力——他们其实根本不想上市。

All of that, I think, is giving OpenAI, who's about to IPO, and Anthropic is probably about to IPO, they're giving them kind of a really big moment where they need to make a massive push because they don't want, you know, they don't want an IPO.

Anthropic这家公司融的钱少得多，用户也少得多，可突然间估值却更高，股价涨得还更猛。

And Anthropic, this company that's raised way less money that's, you know, has way less users, all of a sudden is valued higher and their stock rips more.

OpenAI真不想被甩在后面。

OpenAI really wants to avoid that.

所以他们现在拼命推GPT 5.5，就是为了应对这个局面。

So they're that is what they're trying to fight here with GPT 5.5.

这个模型本身，OpenAI的定位是：它是个‘真正’的全自主代理模型，也就是说，它能独立完成一连串复杂的电脑任务，几乎不需要人工干预。

So the model itself, basically, OpenAI's positioning on this is that it is quote, unquote, fully a fully agentic model, meaning that it's designed to complete all these multistep computer tasks with minimal human direction.

他们特别提到了五大能力：分析数据、编写和调试代码、直接操作软件、在线研究，以及自主创建文档和表格。

They specifically high highlighted five different categories, analyzing data, writing and debugging code, operating software directly, researching online, creating documents and spreadsheets autonomously.

所以他们现在是在给我们所有人推销怎么在工作中用这个，这挺有意思的，因为他们的九亿周活跃用户其实都是普通老百姓，但他们却大力押注企业市场——毕竟Anthropic在这块儿已经杀疯了。

So this is like they're they're giving us all a pitch on how to use this inside of work, which is what's interesting to me because their audience of 900,000,000 weekly active users is I mean, it's just kind of everyday people, but they're really putting it put putting a big push here into enterprise because Anthropic has just crushed it in that area so much.

至于基准测试，它在SWE基准上拿到了88.7%。

So as far as the benchmarks go, it's got eight point or it's got 88.7% on the SWE bench.

在MMLU上是92.4%。

It's got 92.4% on the MMLU.

在终端基准上也有82%。

It's got 82% on the terminal bench.

表现真的挺亮眼的。

It's doing really good.

他们还声称，相比GPT-5.4，它的幻觉率下降了60%，这个数字还挺靠谱的，毕竟以前我们觉得它是最强的，现在他们说：瞧，我们进步了。

They also come claim that it has a 60% drop in hallucinations versus GPT 5.4, which, it kind of holds up, as what we used to be the greatest, and now they're saying, look.

它在这方面的表现越来越好了。

It's getting better at this.

说到定价，这才是最让人感兴趣的地方。

The pricing, think, is also where this gets a lot gets very interesting.

标准版GPT 5.5的输入令牌是每百万5美元，输出令牌是每百万30美元。

The standard GPT 5.5 is $5 per million input tokens and $30 per million output tokens.

这从纸面上看是GPT 5.4的两倍。

That doubles GPT 5.4 on paper.

但OpenAI说这个模型用令牌更高效，所以实际每个任务的成本基本持平，甚至更划算。

But OpenAI is claiming that the model uses tokens more efficiently, so the real world cost per task should be roughly flat or better.

对吧？

Right?

这挺有意思的。

So this is interesting.

他们说的是，你看。

They're saying, look.

输出的令牌、文字，价格确实是翻了一倍。

It's, you know, for the output, for the tokens, for the words that it gives you, it's twice as expensive.

但因为它用令牌特别省，从技术上讲，能用更少的钱干更多活，所以总成本其实差不多。

But because it's so efficient with its tokens, technically, it can get a lot more done for a lot less, so it's about the same cost.

这真的太有意思了。

That's really, really interesting.

GPT 5.5 Pro 输入收费30美元，输出要180美元，贵得离谱。

GPT 5.5 Pro is $30 in and a $180 out, which is very expensive.

比Claude Opus还贵。

It's more than Claude Opus.

很明显，OpenAI的意思就是：你要高端服务，就得付高端价钱。

So OpenAI is, you know, obviously saying if you want the premium tier, you have to pray pay the premium prices.

他们还指望企业客户来买单呢。

They're also betting on enterprise customers doing this.

他们现在已经在ChatGPT里给Pro Plus、企业用户这些高级用户都上线了。

Rollout is they they have pro plus and business and enterprise users inside of ChatGPT that are all getting it.

我试用了一阵子，还用它生成了几份报告，说实话，效果真不错。

I've been playing around with it and generating some reports with it, and it's pretty good if I'm being honest.

不过我得说，我发现每次回复最底下都会弹出一句：嘿。

But I will say I also noticed that right at the bottom of my, you know, response, it's like, hey.

你想不想得到一个更棒的回答？

Do you want an even better response to this?

干脆升级到GPT 5.5 Pro试试。

Try upgrading to GPT 5.5 Pro.

我一点击，结果发现是个每月一百刀的订阅，明显是想让我掏钱。

And I click that, and it's like a $100 a month subscription that it's trying to get me to.

我觉得跟Anthropic一样，他们也明白，对重度用户就得推高阶付费，才能换来更好的效果。

So I think just like Anthropic, they're realizing that they need to for power users, they need to try to push them towards paying more for getting better results.

不过现在这事儿有点复杂，因为Anthropic那边是按token计费的。

Now it's kinda tricky because with Anthropic, it's kind of like token based.

对吧？

Right?

你看啊。

It's like, look.

你想要更多token？

You you want more tokens?

就是你一个小时的配额用完啦。

Like, you ran out of tokens for your hour allotment.

升级一下就能多用点。

Upgrade to get more.

我每个月花两百块买Anthropic的服务时，就包含这个。

And that's something that I pay for when I pay the $200 a month, Anthropic thing.

但OpenAI这边，说白了就是想让你得到更好的回复，搞得我一脸懵。

But for the OpenAI, when it's just like get better responses, it feels like I'm like, what?

所以你们是把最好的回复都锁起来收费了？

So you're just like paywalling the best responses?

我不太喜欢这种做法，相比之下，我更愿意听这样的说法：

I don't like that as much as like, look.

我们给你的是顶级体验，但就是特别耗token。

We're giving you the best, but it just burns your tokens really fast.

所以你想多要点token？这种说法我才真能接受。

So if you want more tokens, that that's a pitch I actually like better.

所以我很好奇这事儿最后会发展成什么样。

So I'll I'm curious to see where that goes.

API马上就要上线了，但还没出来，因为我得加更多安全措施，这事儿真让我有点沮丧。

The API is, quote, unquote, coming very soon, but it's not out yet because I have to put more safeguards on it, which I'm bummed about.

我一向特别期待新API推出，毕竟我也会把它们集成到我的AI工具箱里。

I always love the APIs coming out, obviously, because I'm also adding those to my platform AI box.

现在方方面面都在飞速变化——基础设施、劳动力市场、能力曲线，全都在快速推进。

So, anyways, I think there is so much going on right now as far as infrastructure, the labor market, the capability curve, all of this are moving very quickly.

这周最显眼的大概是GPT-5.5，但所有我们之前聊过的那些东西其实都在协同发力，行业接下来怎么变，真挺值得看的。

GPT 5.5 is, I think, kind of the most visible piece we're seeing this week, but all of the stuff is working together that we've talked about on the show, and, it's gonna be interesting to see some of the shifts in the industry.

明白了。

Okay.

今天节目就到这里。

That's it for the show today.

如果你喜欢这个播客，麻烦在Apple或Spotify上给个评分和评论，这对我真的特别重要。

If you've been enjoying the podcast, it would be a huge favor to me if you could leave a rating and review over on Apple or Spotify, wherever you get your podcasts.

这能帮算法更好地推荐。

It helps the algorithm out.

如果你以前没做过，但已经听了好几期，天啊。

If you haven't done it before, but you've listened to more than one episode, oh my gosh.

如果你能这么做，我真会特别开心，特别感激。

I would be so thrilled and grateful if you could do that.

另外，别忘了去逛逛 aibox.ai。

And, also, make sure to go check out aibox.ai.

描述里有链接。

There is a link in the description.

谢谢你的收听，咱们下期见。

I appreciate you listening, and I'll catch you in the next