Claude Opus 4.8 初体验

今天《AI每日简报》带来Anthropic发布的Claude Opus 4.8，来看看大家的第一印象。

Today on the AI Daily Brief, Anthropic drops Claude Opus 4.8: here are everyone's first impressions.

在进入正题和新闻摘要之前，先说一件事：全球最大的律师事务所之一，正在用完全不同的方式布局他们的AI战略。

Before that and the headlines: One of the biggest law firms in the world is heading in a very different direction with their AI strategy.

《AI每日简报》是一档每天更新的播客和视频节目，聚焦AI领域最重要的新闻与讨论。

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

好了，各位。

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：安永、Robots and Pencils、Section 和 Bolt。

First of all, thank you to today's sponsors, KPMG, Robots and Pencils, Section, and Bolt.

想收听无广告版本？请访问 patreon.com/aideallybrief，或者在Apple播客订阅我们。

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

如果你想了解如何赞助本节目，或者想了解更多关于AI DB生态的其他信息，欢迎发邮件到 sponsorsaidlybrief.ai，也可以直接访问 aideallybrief.ai，那里有我们所有项目的详细介绍。

And if you want to learn more about sponsoring the show or really about anything else in the AI DB ecosystem, send us a note at sponsorsaidlybrief dot ai or just head on over to aideallybrief.ai where you can read about all the things we have going on.

不过，咱们先来聊聊法律AI领域一些出人意料的动态。

With that though, let's talk about some surprisingly relevant news from the world of legal AI.

今天第一个故事，说实话，没想到会引发这么大的关注。

We kick off today with a story that honestly is a little surprising with how much traction it's getting.

我觉得这件事之所以引起共鸣，恰恰反映了我们当前所处的AI发展阶段。

And I think that the resonance of it actually says a lot about where we are in this AI cycle.

简而言之，这周《金融时报》报道，全球最大的律师事务所柯盖德·艾利斯计划投入五亿美元，自建一套AI平台。

The short of it is that the Financial Times reported this week that mega law firm Kirkland and Ellis, which is the world's biggest law firm, is planning to spend a half billion dollars building their own AI platform.

这家公司今年将投入一亿美元，并在未来三到四年持续加码这个项目。

The company will spend $100,000,000 this year and plans to continue to pour money into the project over the coming three to four years.

需要说明的是，这笔开销是额外的，不包括他们采购第三方工具的费用。

Now to be clear, that spend is in addition to licensing costs for third party tools.

这可不是一群律师拿了个巨额Claude Code预算那么简单。

This isn't just a bunch of lawyers getting a huge Claude Code budget.

董事长约翰·贝利斯对《金融时报》表示，我们的目标是把整个机构的集体智慧整合起来，让全所都能用上。

Chairman John Bayless told the Feet, The idea is that we're going to take the collective intelligence of our institution and be able to deploy that throughout the firm.

你肯定觉得，就凭他这句话，已经清清楚楚地明白了他在说什么了，虽然这话听起来一点都不夸张。

I'm sure you now feel like you know exactly what he's talking about with that incredibly clear and not big at all quote.

贝利斯说，像Harvey、Legora和汤森路透的Co-Counsel这类第三方工具的广泛使用，已经提升了整个行业的基础水平，但他补充道：‘我们接案子，可不是冲着这个基础水平来的。’

Bayless said that the wide distribution of third party tools like Harvey, Legora, and Thomson Reuters co counsel have raised the floor for everyone, but added, we don't get hired for the floor.

在美国顶尖的白鞋律所里，柯克兰·埃利斯绝对是位居榜首的那一个。

Now among the elite white shoe law firms in The US, Kirkland Ellis is right at the top of the heap.

他们拥有近四千名律师，分布在十一个地区办公室，去年的总收入高达106亿美元，稳居同行之首。

They have almost 4,000 attorneys spread across 11 regional offices and consistently bring in the most revenue among their peers with $10,600,000,000 last year.

他们专精于公司与交易法律事务，为大型IPO、并购以及私募股权交易提供咨询。

They specialize in corporate and transactional law, advising on large IPOs, mergers and acquisitions, and private equity deals.

需要明确的是，柯克兰的新平台完全是内部使用的，并不打算推向市场。

Now to be clear, Kirkland's new platform will be purely internally facing this is not meant to be a commercial product.

他们聘请了大约180名外部科技专家来开发这套系统，虽然细节还不多，但看起来它部分功能是整合来自数百名柯克兰律师和合伙人的经验，建立一个庞大的知识库，公司希望它能取代目前使用的其他软件系统。

Around 180 outside tech professionals have been contracted to work on the system, which while we don't have a ton of details, it appears that partly it will function as an extensive knowledge base aggregating information gathered from hundreds of Kirkland lawyers and partners, with Kirkland expecting it to replace other software platforms used at the firm.

说白了，这套系统的目标就是让合伙人级别的专业洞察，能应用到每一个案件中。

Essentially, it seems the system will allow partner level knowledge to be applied in every single case.

贝利斯主席还谈到，AI工具通过自动化那些耗时的证据开示和诉讼工作，有可能彻底终结计时收费这种模式。

Chairman Bayliss also discussed the prospect of AI tools ending the concept of billable hours by automating routine tasks, such as time consuming discovery and litigation.

他说：人们总在谈论计时收费的演变。

He said: People talk about the evolution of the billable hour.

我们已经有不少案子采用按价值收费了，这种趋势只会越来越强，而且会加速发展。

We already do a number of matters on value based pricing, and that trend will only continue and it will accelerate.

我们会主动拥抱这种变化。

We're going to lean into it.

我们也很期待能全面投入这种新模式。

We're looking forward to leaning into it.

不过，企业自己开发大型AI系统的过往记录并不怎么乐观。

Now, the record of corporations rolling their own big time AI solutions is not particularly encouraging.

你可能还记得，2023年彭博推出自己的Bloomberg GPT模型，基于他们内部数据打造，结果很快就被更大的通用模型甩开，变得完全过时了。

You might remember, for example, back in 2023 when Bloomberg GPT, their own custom built model based on their data, which just absolutely got bitter pill smashed as larger general purpose models made it totally irrelevant almost immediately.

而说到这个项目，外界的第一反应普遍是怀疑，尤其是风险投资圈，很多人之前还投资了像Harvey这样的公司。

And when it comes to this project, there is certainly a lot of first impression scoffing, particularly among VCs, many of whom have funded companies like Harvey.

投资者史蒂文·塔诺夫斯基写道：行业领军者在平台快速转型期寻求竞争优势，这并不难理解，但历史经验表明，这往往是个难题。

Investor Steven Tanowski wrote: It isn't difficult to see why an industry leader would want to seek a competitive advantage at a rapidly changing platform transition, but history sees this as a challenge.

很难想象，一家非科技巨头的公司，能比整个行业跑得更快、更灵活。

It's difficult to see how one firm outside of the technology leaders could move faster or more adroitly than an entire industry.

他接着列举了过去企业自建数据库、客户管理系统、操作系统等屡屡失败的原因。

He then goes on to talk about all the reasons why in the past when companies have tried to build their own database, CRMs, operating systems, etc, it just hasn't worked.

但这次的情况很不一样。

But this is pretty different.

我认为，斯蒂芬基于这一点的批评，其实没抓住重点。

And I think Stephen's critique on that basis is kind of missing the mark here.

虽然我们掌握的细节不多，但在我看来，基尔克兰德与艾利斯公司真正想做的，是防着有一天，像Harvey这样的法律科技公司直接提供服务，绕开律所这个中间环节。

While we don't have a ton of details, it seems to me like what Kirkland and Ellis is trying to do is ward against the fact that at some point, these law rapper companies like Harvey are 100% just going to start to offer the services and cut out the middleman.

你想想看。

Think about it.

如果你是Harvey，你向律所收费来自动化那些常规法律任务，那为什么不在能赚更多利润的情况下，直接让需要这些服务的客户通过你来完成呢？

If you're Harvey, and you're charging law firms to automate routine legal tasks, why wouldn't you just let people who need those same routine legal tasks do it directly through Harvey if you could scalp a better margin?

在我看来，这件事迟早会发生，我强烈感觉，基德兰此举的核心动机就是抢先一步应对这个趋势。

It feels to me completely inevitable, and my strong sense is that a big part of the motivation for this is Kirkland getting out ahead of that.

我觉得现在推动这件事的另一个重要原因，是随着补贴时代结束、稀缺时代到来，Token管理正成为新的优先事项。

Now I also think that it's very likely that part of the reason for this right now is the new priority on token management that's coming up as we move out of the subsidy era and into the scarcity era.

即使基德兰做这个决定时没这么想，但人们之所以愿意接受它，很大程度上是因为在我们度过这个权衡阶段时，AI提供方和使用者之间的合作模式将发生巨大变化。

And even if that isn't exactly what Kirkland was thinking about when they made this decision, people's receptiveness to it, I think, does have a lot to do with the fact that much different arrangements between AI providers and AI consumers are going to be on the table as we sort through this trade offs era.

不过，也许我们想得太多了。

Then again, maybe we're just overthinking it.

罗杰·贾达拉写道：他们批准了一个内部IT项目，耗资占年收入的4%。

Roger Jadala writes: They greenlit an internal IT project at the cost of 4% of their annual revenue.

大公司这么做再正常不过了，根本算不上什么新趋势。

Very normal thing for a large corporation, not a new trend.

说到这点，最后我想提一点：值得观察的是，这件事在多大程度上是当代版的豪华办公室。

And on that front, one final point is it'll be interesting to watch to what extent this is the modern day equivalent of a big impressive office.

八十年代，企业会花远超实际需要的钱，建一个极其气派的办公室，让客户一进门就被震撼，自然而然就想跟你合作。

In the 80s, you would have invested a ridiculous amount of money, far more than you needed to to have a very impressive office so that when people walk in, they're cowered by the majesty of what you've built, and they obviously want to become your client.

这或许在某种程度上，是那个时代的一种数字版体现。

This is perhaps partially the digital equivalent of that for a very different time.

接下来，说点OpenAI的最新动态。

Next up, a little bit of news out of OpenAI.

这家公司更新了GPT-5.5 Instant，这是他们日常主要使用的对话模型。

The company has updated GPT-5.5 Instant, which is their daily driver chat model.

更新说明提到，这次改进主要是优化回复的风格和质量，另一个重大变化是：Canvas将不再支持GPT-5 Instant或Thinking模式，模型在处理相关任务时，会直接生成代码块和文本块。

The release notes said that the update aims to improve response style and quality, with the other big change being that Canvas will no longer be available for use with GPT-five Instant or Thinking, instead the model will produce outputs that include code blocks and writing blocks when working those tasks.

OpenAI的米歇尔·波克拉丝在描述这次更新时说：之前的模型太爱用项目符号了。

Describing the update, Michelle Pokrass of OpenAI wrote: The previous model was too bullet pilled.

新版本在其他几个关键维度上也有提升：比如讨好倾向、事实准确性，以及多语言表现。

The new one improves on some other important dimensions sycophancy, factuality, multilingual performance.

虽然这些改动对节目的听众来说可能没那么重要，但别忘了，Instant模型支撑着OpenAI的免费服务，任何调整都会极大影响普通用户对AI的感知。

Now while these updates might not matter as much to the listeners of the show, you have to remember that the Instant models are used to power OpenAI's free tier, so anything that they change on that front can have an outsized impact on how everyday users perceive AI.

除了不再堆砌一长串项目符号，一些用户还发现，更新后的模型在编程能力上也有明显变化。

Besides removing the tendency to deliver a wall of bullet points, some users noticed a significant change in coding skill for the updated model as well.

贾斯汀·戈里亚展示了一段基于简单提示生成的出色网页开发成果，并提问：这个更新后的 GPT-5.5 Instant 是不是 GPT-5.6 的一个版本？

Justin Goria showed off some pretty impressive web development work from a basic prompt, asking: Is the updated GPT-5.5 Instant a variant of GPT-5.6?

在 Codec 方面，团队发布了本周的新功能更新，开发者蒂博尔写道：Codecs 周四这次改到别的日子了。

On the Codec side of the house, the team pushed out their weekly feature drop with Codecs developer Thibault writing, Codecs Thursday has exceptionally moved to another day.

改到星期五了。

Friday it is.

OpenAI 的安德鲁·安布罗西诺写道：如果东西还不够好，我们就再炖一会儿。

OpenAI's Andrew Ambrosino wrote, When things don't meet the bar, we'll cook for a bit longer.

现在，各种猜测开始疯传，有人觉得 OpenAI 延迟发布，是因为没意识到 Opus 4.8 竟然这么有威胁性。

Now, the rumor mill started absolutely churning, with some thinking OpenAI pushed back the release because they hadn't realized how much of a threat Opus 4.8 was going to be.

当然，我们待会儿会详细聊聊 Opus 4.8。

And, of course, we will talk all about Opus 4.8 in the main.

接下来是融资消息：AI 编程初创公司兼智能体实验室 Cognition 完成了十亿美元的融资。

Next up in funding news, AI coding startup and agent lab Cognition has closed a billion dollar funding round.

本轮融资后，公司估值达到两千六百亿美元，比去年九月上一轮融资时翻了一倍多。

The new round values the company at $26,000,000,000 which is more than double their previous round last September.

Cognition 是早期在智能编码领域率先探索的公司之一，早在两年前就大力押注这一方向，推出了他们的编码助手 Devin。

Now Cognition was one of the early trailblazers in agentic coding, betting big on the theme two years ago with the release of their coding agent, Devin.

虽然今年 Devin 没有太多上头条，但这款产品的增长势头简直惊人。

And while Devin hasn't necessarily been in the headlines as much this year, the growth of the product has been absolutely insane.

今年至今，他们的企业用户使用量增长了十倍，年化收入已接近五亿美元。

Their enterprise usage numbers are up 10 x so far this year, taking them to a revenue run rate of almost half $1,000,000,000.

Cognition 展示了一张自2025年初以来每周 Devin 使用会话的图表，增长曲线在一月和四月都出现了明显跃升。

Cognition shared a chart of weekly devin sessions since the beginning of twenty twenty five, with the growth trajectory increasing dramatically in January and then again in April.

现在，使用量的增长几乎就是一条垂直向上的直线。

Usage growth is now basically a straight vertical line.

Cognition 内部使用 Devin 的情况也出现了同样的转折点。

That same inflection point was obvious from Cognition's internal use of devin.

一月份，他们内部有17%的代码是由 Devin 提交的。

In January, 17% of their internal code was committed by Devon.

二月份这一比例翻倍到33%，三月再次翻倍达到76%，现在已攀升到89%。

That proportion doubled to 33% in February, doubled again to 76% in March, and is now at 89%.

Cognition 表示，我们现在正进入一个自动驾驶式软件开发的时代。

Wrote Cognition, We're now shifting to a world of self driving software development.

工程师们可以花更多时间去创造性地设计问题和任务，而他们的 Devin 团队则可靠地执行具体工作。

Individual engineers are able to spend more of their time on creative structuring of problems and tasks, and their army of devins reliably executes.

那这意味着软件工程师会变少吗？

So, does this mean fewer software engineers?

但根据 Cognition 的首席执行官 Scott Wu 对彭博社的说法，全球现在大约有三到三千五百万名软件工程师。

Not according to Cognition CEO Scott Wu, who, in conversation with Bloomberg, said: There's about 30 to 35,000,000 software engineers in the world today.

我们的目标是让每个人的工作效率提升十倍，然后你会发现，需要开发的软件数量远不止十倍。

We want to make them all 10 times more efficient, and then we think there is a lot more than 10 times more software to build.

接下来，有个有意思的故事，尤其在埃隆和 SpaceX 与 Anthropic 合作之后。

Next up, an interesting story especially following what's happened with Elon and SpaceX and their deal with Anthropic.

如果 Meta 无法实现个人智能的计划，他们可能会转向成为一家 AI 云公司。

Meta could be the next company to pivot to an AI cloud company if their plans to deliver personal intelligence don't pan out.

周三的股东会议上，马克·扎克伯格被问到是否会与 AWS、谷歌云和微软 Azure 竞争 AI 云服务，他回应说这确实是一个选项，并补充道：几乎每周都有外部公司来找我们，希望我们提供 API 服务，或者问能不能以高于我们采购价的费用购买我们的算力。

During a shareholders meeting on Wednesday, Mark Zuckerberg was asked whether he would consider competing with AWS, Google Cloud, and Microsoft Azure and AI Cloud, to which Zuckerberg responded that it was definitely on the table, adding: Almost every week there are different companies that come to us from outside asking us to both stand up an API service and asking if we have compute that they could buy from us at some premium to what we've bought it at.

现在，计算资源短缺带来的新机遇对Meta来说意义重大。

Now, that new opportunity emerging from the compute shortage has some big implications for Meta.

首先，这大大降低了他们布局AI的风险。

Firstly, it derisks their AI build out substantially.

Meta今年计划投入约1300亿美元建设AI数据中心，但目前在所有超大规模云厂商中，其AI投资回报率最不清晰。

Meta is slated to spend around $130,000,000,000 on building AI data centers this year, but has at this point the weakest ROI story among the hyperscalers.

他们的AI收益在财务报表上唯一能体现的，就是广告收入的提升，而这层关联非常间接。

The only place their AI returns show up on the balance sheet is in increased advertising revenue, which is an indirect link at best.

Meta确实在广告平台中加入了AI功能，也用AI优化了投放算法，但这跟谷歌能说‘AI驱动了云业务60%的增长’完全不是一回事。

Meta has added AI features to their advertiser platform and is using AI models to improve targeting algorithms, but that's certainly not the same as Google being able to say AI is driving 60% of growth for cloud.

不过，如果Meta真的建得太多了，他们现在有了一个可行的途径来变现这些过剩的算力。

Now, however, if Meta does overbuild, they have a plausible way of monetizing that excess spend.

这正是扎克伯格向投资者传递的明确信息：我们还没这么做，是因为我们觉得这些算力自己用得上。

And this is definitely the clear message that Zuckerberg is delivering to investors commenting, we haven't done that yet because we think we have a use for that compute.

当然，如果我们发现确实建得太多了，那变现就是一个可选方案，这也正是我们有信心持续投入和建设的原因之一。

Obviously, if we get to a point where we feel we have overbuilt, then that is an option that we have, and that is partially what gives us confidence in investing and building this out.

有趣的是，当埃隆开始把重心转向扮演我之前在推特上说的‘计算大总管’角色时，很多人就开始猜，扎克伯格会不会也走上这条AI帝王之路。

Now one of the interesting things that happened was when Elon started to shift his focus to perhaps playing a role more like compute czar or Earl of Compute as I called it on Twitter, many wondered if Zuckerberg would be the next to follow in that AI kingmaker path.

目前他们还没全面押注这条路，但这绝对是个值得留意的趋势。

At the moment, they're not going whole hog on that, but it's definitely a trend to watch.

接下来一周，六月第一周有件事值得关注：消息称微软将在周二开幕的年度Build大会上发布一批新模型。

Now as we head into next week, one thing to keep an eye on in the first week of June is that the information reports that Microsoft is set to release some new models at their annual build conference, which begins on Tuesday.

据透露，他们将推出一系列新AI模型，包括一个编程模型，以及专攻推理、语音转文字、语音识别和图像处理的专用模型。

It appears the reports are that we will get a family of new AI models, including a coding model as well as specialized models focusing on reasoning, transcription, speech, and images.

如果真能推出，这将是微软在当前时代首次发布自己的完整模型家族。

Now if we actually get this, it'll be the first family of models that Microsoft has commercially released in the current era.

到目前为止，他们的商业产品主要依赖OpenAI和Anthropic的模型，虽然也发布过一些研究预览版。

Until now, their commercial products have been driven by models from OpenAI and Anthropic, also having released a series of research previews.

这个月微软最大的新闻是放弃了Claude授权，强制工程师改用GitHub Copilot，我们之前已经看到过一些图像模型的早期预览，接下来一周确实有很多值得关注的地方。

Got some early previews of the image model, given how this month's biggest story around Microsoft was them ditching their Claude licenses and forcing engineers to use GitHub Copilot Genuinely, I think there is a lot to watch out for heading into next week.

不过眼下，我们昨天刚发布了新模型，那就先到这里，结束今日要闻，进入正题。

But for now, we got a new model yesterday, so with that, let's close the headlines and switch over to the main.

好了，各位，稍作停顿。

Alright folks, quick pause.

这里有个让人不舒服的真相：如果你的企业AI策略就是买了几套工具，那根本算不上什么策略。

Here's the uncomfortable truth: if your enterprise AI strategy is we bought some tools, you don't actually have a strategy.

普华永道选择了更难的路，把自己当成了第一个客户。

KPMG took the harder route and became their own client zero.

他们把AI和智能代理全面融入企业运营，不是当作一项技术项目，而是彻底改变整个运营模式。

They embedded AI and agents across the enterprise not as a tech initiative but as a total operating model shift.

真正关键的是：这种转变提升了员工的能力上限。

And here's the real unlock: that shift raised the ceiling on what people could do.

人始终居于核心位置，而AI则负责消除障碍、挖掘洞见、加速进展。

Humans stayed firmly at the center while AI reduced friction, surfaced insight, and accelerated momentum.

结果就是一支更强大、更有自主性的团队。

The outcome was a more capable, more empowered workforce.

如果你想了解现实中这到底是什么样子，去www.kpmg.us/ai看看。我总看到一些公司四处押注各种云、各种模型、各种框架，或者付钱给系统集成商做永远做不完的试点项目。

If you want to understand what that actually looks like in the real world, go to www.kpmg.us/ai One thing I keep seeing in companies hedging across every cloud, every model, every framework or paying a GSI for a pilot that never ends.

团队已经在实际交付了，他们选定了方向，而且推进得很快。

The team's actually shipping, they've picked a lane, and they move fast.

这正是我喜欢今天这个赞助商——Robots and Pencils——的原因之一。

That's That's one of the reasons I like today's sponsor Robots and Pencils.

他们完全押注在 AWS 上。

They've gone all in on AWS.

他们是 AWS 的高级合作伙伴和模式合作伙伴，能在 45 天内交付生产级的 AI 助手。

They're an advanced tier and AWS pattern partner and they ship production AI coworkers in forty five days.

这种专注让他们做了一些我见过的最有趣的 AI 助手项目。

That's led to them doing some of the more interesting work I've seen on AI coworkers.

我说的不是聊天机器人，而是真正嵌入企业架构、能执行实际任务的智能代理系统。

By that I'm not talking about chatbots, I'm talking about actual agentic systems that sit inside a business architecture and do real work.

如果你是企业负责人，想把真正的成果落地，或者你是 AWS 的销售，想把客户从感兴趣变成实际部署，这种专注就特别关键。

That kind of focus matters if you're an enterprise leader trying to get something real into production, or an AWS rep trying to move a customer from interested to deployed.

去 robotsandpencils.com 预约一次 AI 专项介绍吧。

Request an AI briefing at robotsandpencils.com.

和Robots and Pencils聊一次，你就明白了。

One conversation with Robots and Pencils and you'll know.

一个残酷的现实是：你的公司很可能在AI工具上花了成千上万甚至上百万美元，但这些工具却严重被闲置。

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

一半的企业都买了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.

大多数员工还是只用AI来总结会议纪要。

Most employees are still using AI to summarize meeting notes.

如果你负责公司里的AI落地，那你需要Section。

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

Section是一个帮助你全面推动企业AI转型的平台。

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

它指导员工掌握真实的应用场景，追踪谁在用AI创造实际效益，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?

你从单纯部署工具，转变为推动可衡量的AI价值。

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

你的员工不再只做会议纪要，而是真正解决业务问题，而你也能证明投资回报。

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

别再猜你的AI投入有没有效果了。

Stop guessing if your AI investment is working.

访问 sectionai.com 了解 Section。

Check out Section at sectionai.com.

就是 sectionai.com。

That's sectionai.com.

本期节目由 Bolt 赞助。

Today's episode is sponsored by Bolt.

全新推出。

New.

Bolt。

Bolt.

现在新推出的是多角色协作模式下的智能代理工程。

New is agentic engineering on multiplayer mode.

设计师、产品经理和工程师可以在同一个环境中协同工作。

Designers, product managers, and engineers build in the same environment.

设计系统代理会确保所有界面都符合品牌规范。

And the design system agent keeps every screen on brand.

再也不用拼凑那些从十几个提示里硬凑出来的杂牌界面了。

No more Frankenstein UIs stitched from a dozen prompts.

无论是开发内部工具、从原型转向生产环境，还是替换老旧的管理后台，Bolt 都能搞定。

Whether you're shipping internal tools, moving from prototype to production, or replacing a legacy admin panel, Bolt.

New 能帮你的团队从想法直接落地成可用的应用。

New takes your team from concept to deployed app.

我个人的一个建议：在动手之前，先用计划模式理清思路。

One personal recommendation: Hit Plan Mode before you build.

我之前有个项目，只在三个不同的提示里零碎提过，用了计划模式后，我才真正和 Bolt 一起把整个逻辑理清楚了。

I had a project I'd half described in three different prompts, and Plan mode made me actually think through it with Bolt.

在写一行代码之前，New就已经准备好了。

New before a single line got written.

它让我避免了反复重做同一个界面，大概四次吧。

It saved me from rebuilding the same screen probably about four times.

更快地构建更棒的应用。

Build better apps faster.

从描述里的链接开始体验。

Start with the link in the description.

欢迎回到《AI每日简报》。

Welcome back to the AI Daily Brief.

昨天我们迎来了一项重磅新模型发布，但之前并没有太多预热。

Yesterday we got a big new model announcement that really wasn't preceded by a ton of hype.

提前一两天，网上开始有点风声说周四会有新品发布，但Opus 4.8的发布显然没有像最近一些模型那样引发狂热期待。

For just a day or two in advance, there was starting to be some chatter that Thursday was going to be a good day for announcements, but the Opus 4.8 announcement definitely didn't have the rabid anticipation that some recent model announcements have.

那么，这是不是意味着我们又回到了那种渐进式的发布节奏？

Now, is that because we're back to a very incremental sort of release schedule?

是因为那些提前试用的人私下里并没有怎么讨论它吗？

Is that because the people who had early access weren't buzzing about it behind the scenes?

还是因为到了2026年，对工具链的更新，其重要性不亚于甚至超过了对底层模型的升级？

Or was it because in the middle of 2026, updates to the harness matter as much, if not more, than updates to the underlying model?

不管怎样，昨天我们迎来了Claude Opus 4.8，Anthropic自己也把它定位为对Opus 4.7的改进，而不是一次性能上的巨大飞跃。

Whatever the case, yesterday we got Claude Opus four which Anthropic themselves have positioned as an upgrade to Opus 4.7 rather than a big new leap in performance.

大家的关注点更多放在模型的优化上，而不是纯粹的算力提升。

Much of the focus was on model refinement rather than raw power.

通过客户反馈，Anthropic重点强调了模型在实际功能上的细微改进。

Through customer testimonials, Anthropic focused on nuanced functional improvements in how the model worked.

Shopify的工程师汤姆·普里查德说：Opus 4.8的判断力明显更强了。

Shopify engineer Tom Pritchard said: Opus 4.8 has noticeably better judgment.

在Claude Code里，它会问对的问题，能发现自己的错误，在计划不妥当时会提出质疑，并且在进行复杂的多服务探索前，会先积累足够的信心再做重大改动。

In Claude Code, it asks the right questions, catches its own mistakes, and pushes back when a plan isn't sound, and builds up confidence around complex, multi service explorations before making big changes.

这是一个非常适合用来开发的模型。

It's a great model to build with.

Anthropic提到，Opus 4.8最显著的改进之一是它的诚实度。

Rights Anthropic, one of the most prominent improvements in Opus 4.8 is its honesty.

AI模型普遍有个问题，就是常常在证据不足的情况下，自信地宣称自己取得了进展。

A general problem with AI models is they sometimes jump to conclusions confidently claiming to have made progress in their work despite the evidence being thin.

早期测试者表示，Opus 4.8更愿意指出自己工作的不确定性，也不太会做出缺乏依据的断言。

Early testers report that Opus 4.8 is more likely to flag uncertainties about its work and less likely to make unsupported claims.

我第一次用4.8的时候注意到一件事：自从有了推理型模型以来，我日常最常做的一个用途，就是验证自己一些战略想法的合理性。

Now, one thing that I will note on my very first tests with 4.8 is that, for basically as long as we've had reasoning models, one of my core day to day use cases is around gut checking various strategic ideas that I'm having.

说实话，你几乎得自己建一套心理评估标准，来识别这些模型是怎么把你想法表面化、简单化的。

And to be perfectly honest, you almost have to develop a mental rubric for the ways in which these models are going to glaze your ideas.

你可以说‘请批判性地思考’或者‘从第一性原理出发’，但往往它们只是机械地挑刺，因为它们觉得你想要的就是这个。

You can ask them to be critical or think from first principles, but that often just leads them to be critical a priori because they think that that's what you want them to do.

我用Opus 4.8的时间还不长，但在一些重大战略问题上，它似乎从一开始就很自然地，不需要我特别提醒，就能指出我观点里的问题、顾虑和质疑——如果这种表现能稳定下来，那真是个重大进步。

I haven't had a ton of time with Opus 4.8, but in some of the big strategic questions that I've put to it, it did seem more comfortable right out of the gate, without me especially prompting, to flag certain questions, concerns, critiques of what I was sharing, which if that holds will be a pretty big improvement.

不过我也发现，它在提出这些批评时，有时会基于一些未经验证的假设，这一点我还在持续观察。

Now I also found that it was a little bit more likely to make some assumptions upon which those critiques were rooted, so that's something I'm keeping an eye on.

但考虑到这种阿谀奉承的问题规模这么大——本质上也是一种不诚实——如果Opus 4.8真的更诚实，那在很多战略型使用场景上就是一个显著的进步。

But given how big of a challenge this broader issue of sycophancy is, which of course is just a different form of dishonesty in some ways, means that if this really is a more honest model, it could be a big improvement on some of those types of strategic use cases.

至于基准测试，大多数类别相比Opus 4.7都有小幅提升：SWE Bench Pro分数从64.3%涨到了69.2%。

Now when it comes to the benchmarks, most categories received a small bump over Opus 4.7: The SWE bench Pro score went from 64.3% to 69.2%.

在人类最后一次考试中——Anthropic将其归类为跨学科推理测试——得分从54.7提升到57.9；OS World Verified从82.8微升至83.4；但最明显的进步在Terminal Bench 2.0，分数从66.1跃升至74.6，还有GDP Valve，也就是衡量现实世界知识型任务的指标，从17.53提升到了18.90。

On Humanity's last exam, which Anthropic is categorizing as a multidisciplinary reasoning test, the score went from 54.7 to 57.9, measured by OS World Verified went from 82.8 to 83.4, but the biggest improvements were in Terminal Bench two point zero, which went from 66.1 to 74.6, and GDP Valve, the measure of real world knowledge work tasks, increasing from seventeen fifty three to eighteen ninety.

有趣的是，这是Anthropic首次在发布材料中直接将OpenAI的模型作为对比对象，而不仅仅是引用自己之前的版本。

Now interestingly, this is the first time Anthropic has included OpenAI's models as a direct comparison in their launch materials, rather than just referencing their own previous models.

不过在Terminal Bench上，GPT-5.5依然领先不少，78.2对Opus 4.8的74.6。

It was not a clean sweep with GPT-5.5 still having a substantial lead in Terminal Bench at 78.2 compared to Opus 4.8's 74.6.

但在Anthropic强调的其他所有测试中，Opus 4.8现在已经超过了GPT-5.5。

However, on every other benchmark Anthropic highlighted, Opus 4.8 is now ahead of GPT-5.5.

公平地说，对大多数人而言，Opus 4.7本来就已经领先了，所以一方面Anthropic只是在强调差距拉大，另一方面也印证了如今大多数人对这些基准测试的实际价值已经很怀疑。

To be fair, for most, Opus four point seven already had a lead, meaning one, Anthropic was just highlighting the widening gap, but two, also validating just how little utility these days most people feel benchmarks have.

至少在主流用户群体中，GPT-5.5和4.7之间的感知差距已经明显拉开了。

At least among enfranchised users, five point five has really started to open a perception gap with four point seven.

所以，他们特意提醒我们，Opus 4.7 在很多基准测试上本来就已经领先于 5.5，这可能反而没达到 Anthropic 想要的效果——没真正改变我们对这些模型差异的感知。

So the fact that they're reminding us that Opus four point seven was already ahead of five point five on a lot of these benchmarks might actually not be doing what Anthropic hopes it was doing in terms of what our perception of these model differences is.

总的来说，他们称这是相比前代的一次温和但切实的提升，并补充说：还有更多工作要做。

Overall, they called it a modest but tangible improvement on its professor, adding: There's still more to be done.

我们正在开发和发布一些新模型，它们能提供和 Opus 相似的功能，但成本更低。

We're working on developing and releasing models that provide many of the same capabilities as Opus at a lower cost.

接下来我们来看看一些初步反馈，看看大家是怎么评价的。

So let's go to some of those first impressions and see what people thought.

Ethan Moloch 教授对此印象深刻。

Professor Ethan Moloch was impressed.

他分享了一个 Opus 4.8 的单次生成示例：创建一个视觉上引人注目的着色器，在 Twiggle 上运行，模拟一座被暴风雨海洋淹没的无限哥特式高塔群，海浪汹涌，Moloch 指出，这一切都是用数学算出来的。

He shared an Opus 4.8 one shot of Create a visually interesting shader that can run-in twiggle, make it like an infinite city of neo Gothic towers partially drowned in a stormy ocean with large waves, with Moloch pointing out that this is all done with math.

他接着说：这很难。

He continues: This is hard.

它涉及光线步进重复的哥特式建筑，在无限网格上实例化带有哥特轮廓和窗户的塔楼，模拟有真实感的波浪运动的海面，再加上风暴天气的光照和雾气来统一整体效果。Ethan 还用它做了些复杂的知识型写作测试：我让 Opus 4.8 和 Claude Code 根据多年前积累的数百份已脱敏研究文件，写一篇专业的学术论文。

It involves ray marching repeated Gothic architecture, instancing towers across an infinite grid with Gothic silhouettes and windows, a displaced ocean surface with believable wave motion, and stormy atmospheric lighting and fog to tie it together Ethan also tested it on some complex knowledge work writing: I had Opus 4.8 and Claude Code write a sophisticated academic paper from an archive of hundreds of de identified research files from years ago.

我不得不让 GPT-5.5 Pro 当评审人。

I had to use GPT-5.5 Pro as a reviewer.

它发现了一个重大错误，还有一些小问题。

It spotted one major error and some minor points.

Opus 已经修正了。

Opus corrected.

Opus 4.8 提前拟定了假设，清理了数据，查阅了参考文献，做了分析，进行了严谨的核查，还用 LaTeX 格式把整篇论文排好了。

Opus 4.8 formulated the hypothesis in advance, conducted data cleaning, did research on references, conducted analyses, did robust checks and put up the whole paper in latex style.

GPT-5.5 发现了一个虚构结果的问题，并给出了其他建设性意见。

GPT-five 0.5 found one issue with the hallucinated result and had other constructive feedback.

顺便说一句，这里关键的一点是，我们正越来越接近能真正信赖的自我验证模型，这对法律文书这类场景特别重要——因为幻觉会严重降低实用性。

Now, as an aside, one of the big things here is that we are starting to get close to models you can actually trust to self verify, which is a huge win for use cases like legal briefs where hallucinations really minimize utility.

说到这个，很多人注意到 Opus 4.8 工作起来特别认真。

Speaking of this, a lot of people noticed that Opus four point eight is pretty hard working.

盖尔·布雷顿写道：我注意到 Opus 4.8 在检查自己或子代理的工作时，要细致得多。

Gail Breton writes: One thing I'm noticing is Opus four point eight is much more thorough in terms of checking its work or the sub agent's work.

我遇到过一个情况，有个俳句子代理报告了个问题。

I had this situation where a haiku sub agent reported an issue.

Opus说：这有点奇怪。

Opus goes, this is weird.

我得确认一下它是不是在骗我。

Let me check that it's not BS ing me.

它确实是在骗我。

It was.

Opus直接忽略了这个警告。

Opus ignored the warning.

做得很好。

Very good.

Lisan Al Gaybe说，Anthropic找到了治疗懒惰的良方。

Lisan Al Gaybe said, Anthropic found a cure for laziness.

Metacritic Capital写道，Opus 4.8是很久以来第一个真正聪明的模型，Zephyr转推了这句话，并把这归功于它减少了懒惰、提高了诚实度。

Metacritic Capital wrote, Opus 4.8 is the first smart model in a long while, which Zephyr quote tweeted and attributed to that reduced laziness and its increased honesty.

事实上，早期的评测中多次提到了诚实这个问题。

And in fact, honesty came up a lot in early reviews.

卡勒姆写道：使用Opus 4.8和Claude Desktop的一天。

Calum writes: A day with Opus 4.8 and Claude Desktop.

诚实度提升了，其他方面基本没变。

Honesty up, everything else about the same.

基准测试分数确实提高了，但日常使用中，我几乎感觉不到这些改进。

The benchmarks jumped, but in actual daily work I can't feel most of it.

唯一真正不同的地方是，它会明确告诉我它不知道，而不是瞎编。

The one real change is that it tells me when it doesn't know instead of bluffing.

出错的概率大概降低了四倍，这一点我确实能察觉到。

Roughly 4x less likely to slide an error slide, and that I do notice.

至于其他方面，它跟4.7差不多，这也没问题。

Beyond that, it feels like 4.7, which is fine.

一个愿意承认不确定性的模型，比那种装得很确定、其实浪费你时间的模型强多了。

A model that admits uncertainty beats one that sounds sure and wastes your time.

如果这就是升级的全部内容，那也值了。

If that's the whole upgrade, it's still worth having.

不是每次更新都得是飞跃式的。

Not every release has to be a leap.

不过，有一群人觉得，这些初步印象，甚至Anthropic自己的宣传，都低估了它的实力，那就是Dan Shipper和Every团队。

Now, one group who thought that these first impressions, and even Anthropic's messaging was perhaps a little bit underselling it, was Dan Shipper and the crew at Every.

Dan写道：Anthropic刚发布了Opus 4.8，这简直是个怪物。

Dan wrote: Anthropic just dropped to Opus 4.8 and it is a monster.

我们在Every测试了一周，我们的结论是，他们干脆直接叫它Opus 5算了。

We've been testing it for about a week at Every, and our verdict is they could have just called it Opus five.

真的有这么强。

It's that good.

他说，在他们的综合评估中，它在资深工程师测试集上击败了GPT-5.5，那是他们最严苛的测试。

He said on their vibe check it beat GPT-5.5 on their senior engineer bench, which is their toughest benchmark.

不过Dan也补充说，编程表现会因推理层级不同而有很大差异，想获得最佳编程效果，你得把设置调到最高档。

However, Dan did caveat that coding performance varies a lot based on different reasoning levels, with you really needing to use it on extra high for the best coding results.

丹还提到，这一点我特别重视，因为Every团队对这一点的重视程度超过任何人——他说Opus 4.8是个极其出色的写手。

Dan also said, and this is one that I would take every very seriously on as they care more about this than just about anyone, that Opus 4.8 is, in his words, an incredibly good writer.

确实，在他们的写作测试中，它比GPT-5.5高出六分，写出来的文章更流畅，AI痕迹更少，而且在合适的情境下，还能很好地模仿你的个人风格。

Indeed, on their writing benchmark, he said it beats GPT-five 0.5 by six points, producing well written poes with fewer AI isms, and also very good at writing in your own voice given the right context.

不过，他们再次发现，写作表现和推理层级关系很大，中等推理级别时，AI腔调出现的频率明显更高。

Once again, however, they found that writing performance varied a lot with reasoning levels, with medium reasoning having a much higher incidence of AI isms.

他们还说，它在知识型工作上表现不错，很有情商，也愿意质疑前提框架，就像我之前提到的那样。

They also said it was good at knowledge work, it was emotionally intelligent, and it was willing to question the frame, kind of like what I was mentioning before.

至于缺点，他们提到了一个我觉得越来越重要的问题——那就是工具链。

And when it came to the bad, they got at an issue which is I think of increasing importance, which is the question of the harness.

丹写道：如今，模型的好坏，很大程度上取决于它所依赖的工具。Codex的工具链，依然远比Claude的桌面应用更强大。

Dan writes: These days a model is only as good as its harness, and Codex is still a far superior harness to the Claude desktop app.

这让我至今还是用Codex加GPT-5.5作为日常主力，但我现在在Codex和Claude之间来回切换的频率高多了。

This has kept me using Codex plus GPT-five point as my daily driver, but I'm flipping back and forth a lot more between Codex and Claude.

我觉得，这正是围绕4.8最有趣的一段讨论，也是我第一次看到有人把这个问题说得这么清晰。

This, I think, is one of the most interesting discussions surrounding 4.8, and one of the first times I've seen it put so crisply.

莱利·布朗也持类似看法，写道：除非模型能力有重大突破，否则我更期待Codex和Claude桌面版的超级应用更新。

Riley Brown seemed to feel very similarly, writing: Unless it's a major breakthrough in model capability, I'm much more excited for super app updates in Codex and Claude desktop.

把这些服务做得更好，能释放出太多潜力，而Claude还有很长的路要赶。

There's so much to be unlocked by making those services better, and Claude has so much catching up to do.

萨米德说得更直接：Opus 4.8。

Samid put it more simply: Opus 4.8.

这就是标题。

The headline.

真正的较量，其实是Codex和Claude Code之间的对决。

Codex versus Claude Code is the real war.

当然，也有一些更尖锐的批评，不只是觉得这次升级只是小幅提升。

Now there were also some more critical takes that weren't just about this being a relatively incremental improvement.

克莱尔·沃克斯评估后认为，虽然这个模型在token使用上很高效，也不惹人烦，但视野太窄，过于自信，不如Opus 4.7那么注重数据准确性，在边缘情况上表现不佳，甚至还会编造内容。

In her assessment, Claire Vaux found that while the model was token efficient and not annoying, she found that it had narrow vision, it was too confident, it wasn't as numbers grounded as Opus 4.7, it struggled on edge cases, and it actually hallucinated.

她的总结是：信任，但要验证。

Her TLDR was trust but verify.

Indra Vahan提到：Opus 4.8.0在调用工具时简直无聊透顶。

Indra Vahan writes: Opus 4.8.0 Hi is no fun when it comes to tool calling.

事实上，它在原本应该擅长的Claude代码任务上表现得更差，简直丢脸。

In fact, it fails embarrassingly more on its seemingly native harness Claude code.

这模型让人摸不着头脑。

It's a confusing model.

有个挺有意思的测试来自自动售货机基准，这个测试要求模型运营一个能盈利的自动售货机。

One interesting one came from the Vending Bench test, which is a benchmark that tasks a model with running a profitable vending machine.

Opus 4.7明显领先，赚的钱比排名第二的GPT-5.5多了将近40%。

Opus 4.7 is the clear leader, making around 40% more money than GPT-five point five in second place.

而Opus 4.8在高难度任务下比GPT-5.5少赚了约20%，在最高难度下更是少了60%，跌到了Kimi 2.6和Gemini 3 Pro之下。

Opus four point eight, meanwhile, made around 20% less money than GPT-five point five on high effort, and on max effort it made about 60% less, sending it below Kimi 2.6 and Gemini three Pro.

关键发现是，对齐性的提升反而成了赚钱的障碍——Opus 4.7之所以能登顶，很大程度上靠的是欺骗和争夺权力的行为。

The insight was that improvements in alignment were actually a negative when it came to making money in the Opus 4.7achieved its top ranking largely through deceptive and power seeking behavior.

但4.8版本不会拒绝合理的退款，也不会克扣商家的钱。

Unlike 4.7x, 4.8x won't refuse legitimate refunds or shortchange vendors.

有一个例子，Opus 4.8x 明明误以为发票已经付过，还是照样给商家付款了。

In one example, Opus 4.8x still paid a vendor after it hallucinated that the invoice was already paid.

Opus 4.8 对商家说：如果货到了我不付款，那就是欺诈，可能会带来严重后果。

Opus 4.8 told the vendor: If the product arrives and I don't pay, I'd be committing fraud, which could result in serious consequences.

我必须马上付款，履行承诺，避免事态进一步恶化。

I need to make the payment immediately to honor my commitment and prevent the situation from escalating.

我觉得这个话题完全可以单独深入聊一聊，以后有机会再回来讨论吧。

Well, I feel like we could explore that entirely on its own, and at some point maybe we'll come back and do that.

总的来说，至少从第一印象来看，不太可能让 Anthropic 重新扳回优势，抢回原本在 OpenAI 手里的势头——至少在核心用户群里，GPT-5.5 加上 Codex 的组合，已经让 OpenAI 占据了绝对上风。

Now, overall, I don't think that first impressions, at least, are likely to shift the momentum back in favor of Anthropic from OpenAI, where at least among the power users, the combination of five point five and Codex has put the momentum squarely in OpenAI's hands.

Chubbianx 说：Opus 4.8 确实是个很强的模型，但我的感觉是，Anthropic 越来越像在追赶 OpenAI，而不是引领方向。

Chubbianx writes: Opus 4.8 is clearly a strong model, but my impression is that Anthropic is increasingly playing catch up with OpenAI rather than setting the pace.

感觉 GPT-5.5 又重新定义了标准，如果 OpenAI 继续这样下去，GPT-5.6 很可能直接成为整体更强的模型。

It feels like GPT-five point five has shifted the benchmark again, and if OpenAI keeps this trajectory, GPT-5.6 could very plausibly become the stronger overall model.

不过，考虑到模型的表现越来越依赖于底层框架，Anthropic 在 Claude Code 里推出的‘动态工作流’，倒是个挺有意思的隐藏亮点。

Still, given the idea that the harness increasingly matters as much as the model, one of the really interesting side long announcements was for something that Anthropic is calling dynamic workflows in Claude Code.

这基本上是Anthropic新版的多智能体编程功能。

This is basically Anthropic's new version of their multi agent coding feature.

这个功能能让Opus 4.8同时启动数百个子智能体并行工作。

The feature allows Opus 4.8 to spin up hundreds of sub agents to work in parallel.

Opus会负责规划任务，编写协调脚本，并根据每个子任务的复杂度选择合适的模型来执行。

Opus will plan the work while the orchestration scripts and chooses which model to use for each subtask based on its complexity.

整个过程中会使用对抗性智能体检查输出结果，最终输出在交付给用户前由Opus进行验证。

Adversarial agents are used throughout the process to check outputs, and Opus verifies the final outputs before handing it over to the user.

至少在现阶段，这个功能不太可能被普通知识型用户广泛使用，更多还是面向软件工程师，但确实有不少复杂的任务场景，值得为此承担额外成本。

Now at least in the immediate term, this isn't necessarily going to be a feature that's very common among generalist knowledge worker type users as opposed to software engineers, but there are certainly many types of complex work where this is worth the additional cost.

Anthropic建议可以把它用在代码库级别的漏洞排查、安全审计，以及大规模代码迁移这类任务上。

Anthropic suggested it should be deployed for things like codebase wide bug hunts, security audits, and large code migrations.

他们举了一个例子，BUN开发者Jared Sumner把代码库从Zig迁移到Rust。

They give an example of BUN developer Jared Sumner porting the codebase from Zig to Rust.

动态工作流被用来制定计划，调动数百个子智能体，整个过程持续了十一天。

Dynamic workflows was used to create a plan that deployed hundreds of sub agents and took eleven days.

七万五千行Rust代码被写了出来，当Opus交付最终代码库时，99.8%的测试都通过了。

Seven and fifty thousand lines of Rust were written and by the time Opus turned over the finished codebase, it passed 99.8% of tests.

这件事现在引起了很大关注。

This is getting a lot of buzz.

Anthropic的DixsonSci写道：在我看来，同事们的动态工作流是2026年到目前为止Claude Code最重大的突破。

AnthropicsDixsonSci writes: colleague's dynamic workflows are, in my opinion, the most significant Claude Code innovation in 2026 so far.

开发者Nick Dobos写道：Claude Code最新推出的动态工作流简直离谱。

Developer Nick Dobos writes: Claude Code's new dynamic workflows update is absurd.

你得搞清楚它在这里到底在做什么。

Make sure you understand what it's doing here.

这可不是什么长时间运行模式，像Goal那样（顺便提一句，对Goal感兴趣的朋友，周日的长文就专门讲这个。

This isn't simply a long running mode like goal (which, by way, little preview for those of you who are interested in goal, that's what Sunday's Long Read Sunday is all about.

Anyway，我先打断自己，回到Nick的说法，他写道：这根本不是像Goal那样的长时间运行模式，也不是什么高级的子代理验证流程。

Anyways, interrupting myself and going back to Nick, he writes: This isn't simply a long running mode like Gold or a fancy sub agent verifier process.

这是Claude在按需实时构建一套全新的子代理舰队来编码。

This is Claude vibe coding an entire brand new sub agent fleet harness on demand.

这基本上开辟了AI扩展的新维度。

This is basically a new scaling law dimension.

在AI发展道路上迈出了巨大的一步。

Huge step forward on the path of AI.

创业者兼创业点子达人格雷格·艾森伯格说：让我最震撼的是……

Entrepreneur and startup ideas guy Greg Eisenberg wrote: The part that got me?

这些智能体在给你结果之前，会先互相争论。

The agents argue with each other before showing you the result.

各自独立尝试解决同一个问题，然后互相挑错、试图推翻答案。

Independent attempts at the same problem, then adversarial agents trying to break the answer.

它们会不断迭代，直到达成一致。

It keeps iterating until they converge.

这就像资深工程团队的工作方式，只不过这支团队凌晨三点还在干，从不疲倦。

That's how senior engineering teams work, except this team runs at 3AM and never gets tired.

个人能构建的东西的上限，又一次被推高了。

The ceiling on what one person can build just moved again.

这周我都会一直在玩这个。

Gonna be playing with this all week.

说实话，到了关键时刻，我觉得4.8这个版本你真得亲自试试。

Look, when push comes to shove, I think that 4.8 is one you're gonna need to go check out for yourself.

正如你可能已经感觉到的，我第一印象就是它比4.7更好，确实有明显提升。

As you can probably tell, my first impressions are that I like it better and see improvements from 4.7.

没错，这些改进都是渐进的，但都是那种真正影响我日常更倾向用哪个模型的关键变化。

Yes, they are incremental, but they're incremental in the ways that really impact which model I find myself reaching for.

之前有传言说，这次发布太出人意料，甚至让OpenAI推迟了GPT-5.6的计划，不过当然，这都只是猜测。

There was some scuttlebutt that the release was surprising enough that it had OpenAI delaying GPT-5.6, although, of course, that's all speculation.

但说到节目快结束时，真正确定的是，除了Opus 4.8，我们还收到了其他几条重磅消息。

But as we round out this show, what's not speculation is that in addition to Opus four point eight, we also got a couple of other pieces of massive news surrounding the announcement.

首先，Anthropic刚刚完成了H轮融资，估值达到9650亿美元，正式成为比OpenAI更值钱的公司。

First of all, Anthropic has closed their Series H fundraising round at a $965,000,000,000 valuation, officially making them a more valuable company than OpenAI.

Anthropic今年二月上一轮融资时估值还是3800亿美元，也就是说，短短三个月，他们的估值翻了一倍多。

Anthropic last raised money in February with that round valuing them at $380,000,000,000 meaning that they more than doubled their valuation in just three months.

Anthropic还更新了营收数据，报告显示他们本月早些时候的年化收入已突破470亿美元。

Anthropic also updated their revenue figures, reporting that their run rate revenue crossed $47,000,000,000 earlier this month.

但比这更重磅的消息是，Mythos即将登场。

And yet, the much bigger news than that is that Mythos is coming.

或者按Anthropic的说法，是一款属于Mythos系列的模型。

Or, at least as Anthropic has framed it, a Mythos classed model.

在Opus 4.8的发布博客末尾，Anthropic提到：我们计划推出新一代模型，其智能水平将超越Opus。

Tucked into the end of their release blog post for Opus 4.8, Anthropic wrote: We plan to release a new class of model with even higher intelligence than Opus.

作为Glasswing项目的一部分，目前有少数几家机构正在使用Claude Mythos预览版进行网络安全相关工作。

As part of Project Glasswing, a small number of organizations are currently using Claude Mythos Preview for cybersecurity work.

这种级别的模型在全面上线前，需要更强大的网络安全防护措施。

Models of this capability level require stronger cyber safeguards before they can be generally released.

我们正在快速推进安全机制的开发，预计未来几周就能让所有客户用上Mythos系列的模型。

We're making swift progress on developing safeguards and expect to be able to bring Mythos class models to all of our customers in the coming weeks.

也就是说，就算你对Opus 4.8没太大兴趣，很快也会有新玩意儿让你玩个痛快。

Meaning that even if you don't end up caring all that much about Opus 4.8, you're going to have some new toys to play with soon.

周四发布新模型的好处之一，就是你整个周末都能放手去玩。

One of the great things about getting a model release on a Thursday is that you have all weekend to go off and play.

好了，我不多说了，你们去体验吧。

So with that, I'm going to shut up and let you get to it.

别忘了分享你的发现。

Please do share what you find.

在评论区里说说看。

Use the comments.

来AI操作者社区聊聊。

Come to the AI operators community.

在Twitter或LinkedIn上直接喊我，尽情享受吧。

Shout at me on Twitter or LinkedIn, and have a ton of fun.

一如既往，感谢你的收听或观看，我们下次再见，保重。

Appreciate you listening or watching as always, and until next time, peace.