They lied... Transcript, AI Summary & Key Points

00:00 Starting September 14th, we're permanently raising standard weekly limits by 25% for Pro Max and seatbased enterprise plans. Huh, that's interesting. I thought the title of this video was that Anthropic was lowering our limits. Oh, there's more here. Until then, the current 50% increase will stay in place until Huh? This screenshots cut off. Let's go look at the official post.

00:23 Oh, they deleted it. Wow. As always, Anthropic is doing bad things and is doing even worse at communicating these bad things. The new version of the post is up here and in my opinion is still an egregious failure of communication. Starting September 14th, we're permanently raising standard weekly limits in cloud code by 25% for Pro Max teams and seatbased enterprise plans.

00:48 Until then, the current 50% increase will be in place. What this means is that they are dropping your usage by a meaningful percentage. It's not 25% less to be clear. It's a 25% increase reduction. So, comes out to 17% when you do the math. But this still sucks and is particularly egregious both in terms of the way Anthropic is communicating as well as the other things they should have done that they failed to do here.

01:09 I have been paying attention closely to Anthropic's rate limits for a while now. In fact, I've gone as far as having five accounts. I have not been free enough to code much this week, which is why most of them don't have their limits down. But I have had plenty a week where I managed to burn all of these accounts to the ground. So, I am deeply intimately familiar with Enthropics rate limits, and as you can probably guess with that, I have a lot of thoughts about these changes.

01:34 There's also a pretty crazy history here that goes around a lot of corners, including but not limited to SpaceX's involvement, which is deeper than you probably think. And I guess I'm just not allowed weekends anymore because they put this post out on a Saturday morning. Anthropic, anthropic, anthropic. You were supposed to drop a new model 2 days ago.

01:52 Instead, you're strangling us for compute. I want to talk about these changes, the history that got us here, and just how bad Anthropics compute issues are. But if I want to keep using Fable, I need some way to pay these absurd bills. So, pardon me for a moment because we need to hear from today's sponsor. It's never been easier to ship code. Whether you're trying to add features, fix bugs, clean up tech debt, or whatever else, making code changes is trivial now.

02:13 The hard part is knowing what to do and how to do it right. The best way to figure these things out is, of course, to figure out what users actually want. Doing that requires a lot of data on how they do things. Hopefully, you're already using a product like Post Hog in order to track all of the things your users are doing because from that data, you can figure out a lot.

02:31 That said, it takes a lot of effort to go through all of this data and figure out what your users actually need to use and stay using your applications. Wouldn't it be nice if something could just automatically take all this data from how people are using your apps and constantly improve them based on the real needs of real customers? Well, I have some good news because Post Hog agrees and they figured it out.

02:51 They've been transforming their all-in-one suite of product tools into a self-driving product iteration loop. What this means is that they can look at the analytics and see what users are doing and automatically start fixing bugs and introducing things that users need to succeed. Kind of wild that we're in a world where you don't even have to write the prompt.

03:08 You just have to click the merge button. But there's a reason why they built this and they've been using it themselves. And honestly, the more I've been playing with it, the more I've been wanting to use it, too. Tools like Cloud Code and Codeex are incredible at building things, but they don't know what your users actually need and what experiences they're actually having.

03:25 That's what makes Post Hog spin on this so powerful. They have the data. They know what your users actually need and what problems they're actually facing. And not only is it free to get started, 97% of Post Hog users pay 0 because their free tier is so absurdly generous. Start building what your users need at soy.link/postto. First and foremost, I want to cover all the funny things people are saying to Anthropic about this, like Matt PCO saying, "Stop hitting yourself."

03:50 While also tweeting a pretty solid post here, had some personal communication with members of the Claude Code team. They seem lovely, well-meaning, intelligent, and gifted. All of this is to say, "I find communication from Claude Devs, the official account here, absolutely baffling. It's fumble after fumble. They used to make their employees do these cringe tweets.

04:09 They made this new Claude Dev account seemingly largely to be the place where all of the bad announcements go as well as the occasional good one. And this is one of those bad announcements. Also, the wording here is egregious, especially when you consider the first time they made the post it was misleading enough that they deleted it. To come back and effectively mislead again by opening so similarly, it's actually absurd.

04:31 I cannot believe that they had the foresight to delete the bad post and then make another bad post. For this to all make sense, we have to go back in time a bit all the way back to May, which at this point feels like it was years ago. We've agreed to a partnership with SpaceX that will substantially increase our compute capacity. This, along with our other recent compute deals, means that we've been able to increase our usage limits for Cloud Code and the Claude API.

04:54 Effective today, we are doubling the Cloud Code 5H hour rate limit for Pro, Max, and Teams plans. We're removing the peak hour limit reduction on Claude Code for Pro and Max plans. and we are substantially raising our API rate limits for Opus models. At this time, a lot of people, myself included, were not pumped because actually wasn't the hourly limits.

05:11 It was the weekly limits. I'm also obligated to tell you guys that Codeex subscriptions don't have 5 hour limits on the $100 and $200 plans. They might bring them back in the future, but they don't have them right now. And it is very nice, even if you can accidentally burn your whole weekly a little easy. So Claude doubled the 5 hour limits, which cool, nice, whatever.

05:30 Not great when you can burn through your weekly in two sessions. So they followed up a week after this in midmay, May 13th at this point to say that cloud code weekly limits have been increased by 50% from now until July 13th. Huh. Last I checked, it's been a long time since July 13th. Well, in that window, some fun things happened. I don't know if you remember, but there was a certain model called Fable or something, and I think it got banned by like the US.

05:52 Yeah. When that all happened, a lot of people weren't happy with the value they were getting from their anthropic plan because they no longer had access to Fable in it. I know I had registered one of my accounts just for Fable right when it dropped and losing that sucked. So they end up extending a 30% increase in limits all the way to August 19th because of the Fable removal but then also because of Fable being added back.

06:15 But when they added Fable back they had one interesting change that has been I'll be real kind of killer for me in particular. It's been very painful. When they brought back Fable 5 they didn't give you Fable 5 for your whole weekly limit. So, let's say theoretically you can do $1,000 of inference on your $200 plan. It's actually closer to eight grand, but we'll say a thousand.

06:32 If you can do $1,000 of inference, once you hit $500 of Fable 5, you're not allowed to use Fable anymore for that window. You have to move over to other models like Opus and Sonnet. This sucked because it was not how it worked before and it means that I'm effectively losing half my usage, being forced to use Fable. It's actually like a meme now that at any given time I have practically unlimited opus usage because I have multiple Claude code accounts that have used their whole Fable limit and still have 50% of their

06:58 week left. Just so you know how much I'm struggling here. I have accidentally referred to my Claude account as my Fable account multiple times. I literally just use it for Fable. Opus and Sonnet piss me off. They're just not good right now. I thought Opus would be good. I was wrong. I will keep eating that one for a while. I am sorry for saying it was good when it's not.

07:15 I just use Fable and it's worth the money for me for Fable. it is that good a model. So, the 50% limit killed me and I was upset, but I was hoping that as compute got better, this would be more usable. Thank you for our patience over the past several weeks. We're continuing to invest in new capacity and we'll keep everyone updated as we do. I don't feel like finding the tweet for it, but they did ultimately end up extending one last time to this new September date.

07:39 And now we have today's comms, which is the world's worst way of saying your limits are going down by pretending that they're actually going up. The harsh reality is that most Cloud Code users aren't locked in people on Twitter. They are relying on things like email communications and clarity in the thing they're buying when they buy it. So, this being the only way users are getting informed that their Cloud Code accounts are going to have their limits reduced is not great.

08:03 I would guess the majority of heavy users right now registered their account or started paying for their account or at least bumped their subscription tier at some point since this 50% increase happened. That means that the whole time they've been using Cloud Code, they were used to a limit that will no longer exist starting on September 14th. That is stupid.

08:22 That is a really bad experience and I would argue goes as far as misinforming customers, which means if you cancel because you aren't getting what you're used to anymore, I totally understand. As a person who's admittedly a little too nerdy about comms for a developer, I took my own spin at this as well because I wanted to see if I could show people how you would format something like this correctly with more empathy towards your users.

08:42 I want to call out that the last chunk at the end here isn't something they decided to do. I was giving an example of a goodwill gesture they could make. I still hope they do, but anything I say about Fable in this should not be considered real. It isn't. I wish it was. Back in May, we did a one-month promotion where we increased the weekly limits of Cloud Code by 50%.

09:02 We extended this promotion multiple times since, but sadly, it must come to an end on September 14th. We know that this will impact the work of many users. We've run the numbers and found that we were able to confidently make a permanent increase of 25% from the original base. This nets out to a 17% decrease from the limits as they are today. We understand this is a big hit for a lot of users.

09:25 We hate making changes like this. We're doing everything in our power to secure more compute to improve availability of cloud. We do have some good news. This is the part I said isn't real, so don't get too excited. When these changes go live on September 14th, we will be removing the separate weekly limit for Fable. So, you will now be able to use your full limit with your preferred Clawed model.

09:44 You get the idea. You lead with the bad news and be very realistic and empathetic about it, talking less about what the change is and more about how you will feel the change. This is the thing Enthropic just doesn't do. Funny enough, they are capable of doing this. For example, right here where the chief comput officer at Anthropic took the opportunity to on OpenAI indirectly when the OpenAI news about the breakup between them and Cursor happened.

10:09 Curser has been a trusted partner of Anthropic since Sonnet 3.5. We'll continue to increase compute to support cloud models in Cursor and are excited for what comes next with them at SpaceX. Looks like they're willing to talk about these things when they're dunking on competition and also sweetening their relationship with SpaceX because they need the compute so bad.

10:24 But they're not willing to say things like this when it is affecting their users because I'm going to be super harsh and critical here. Anthropic doesn't give a about their customers. They just don't. Their customers are a means to an end. The end being inventing a god that is actually friends with them. And they just seem to treat customers like numbers.

10:43 So the announcement they made had nothing to do with smoothing over the harsh experience that their users are about to have and everything to do with the internal debate they had over how much allocation they could or couldn't give to the users of the subscriptions and then some random person was given the task of turning that into a post. I'm guessing Claude and I'm guessing that Claude has specific instructions to make everything seem good even when it isn't.

11:05 Which is why when they talk to customers they can't really acknowledge pain points. They instead talk about it as though it is a good thing. This is Anthropic's specialty. This is particularly funny when at the same time OpenAI and SpaceX are having their weird breakup with cursor where OpenAI models are being banned. I already did a video on this. It's probably up by now.

11:25 Take a look. This one is a much better example of how to communicate these things. To maximize the time the developers can retain access to our models through cursor, we're giving the maximum notice provided by our contract. This decision was incredibly tough as we care deeply about our models being broadly available for developers. We're making this choice because we cannot be confident yada yada yada.

11:43 You get the idea. They are focusing much more on the developers who are affected and how tough this decision is. It's just is written with anything resembling empathy, which is not the case at all. When anthropics cruise up, they either hide it or they pretend this was some greater problem beyond their control or they blame compute or other bad decisions they made while pretending it's not their fault.

12:03 Anthropic does not admit fault when they are wrong. and barber just refuses to take the L even when they cause it. And this is such an egregious example of it that I felt obligated to come here and crash out of it. I think this all just comes down to compute availability, especially during the workday because Anthropic makes so much money using their compute for their enterprise customers.

12:23 And since that compute is being paid per usage the way that you would over the API and with much less subsidization than you get from using the model with a subscription, the subscriptions are now a loss leader that is restricting compute from being available for the API sales they want to make. Which means that during the business day when those companies are the most interested in using those API calls, if there isn't enough compute available, those requests fail and then they don't make the money.

12:48 I'm surprised they haven't brought back the limits during peak business hours. I would suspect that will become a more popular pattern because that is the best way to save money when you have subscriptions like this, but we'll see where that all ends up. I don't love where it's landing right now, especially when you combine the reduced weekly limits with the much stricter Fable limits, especially since Fable 5.1 just isn't here yet.

13:09 And let's be real, when Aster drops, it might be more expensive and it might use your codeex limits faster, but it's not going to have these types of arbitrary restrictions that are honestly claude specials at this point. I think this sucks and I really really hope Astro turns out to be a good model. I'm surprised they announced this the way they did in the timing they did.

13:27 I hope that this gets followed with a fun announcement on Monday or Tuesday like Fable 5.1 is here. It's more efficient and you can use your whole limit. But that's not what happened here. This is just a downgrade and they are pretending it isn't when it obviously is. Anthropic, I beg you, please stop lying to your users like this. We're smarter than you think.

13:47 I know you only think about your business customers and your employees and Claude and that the users are the ones that get screwed in the end, but if you keep doing this and OpenAI keeps making models as good as it looks like they are, you're not going to hold this position much longer. You do have two models that bench better than anything OpenAI does, but I still choose to use Soul every day and it's a great model.

14:07 Fable's obviously the one I like most for code. I've been very transparent about that. There's a reason I have all those subs and it's not cuz I hate the model. Fable 5 is unbelievably good. When you look at the gap here in how strict Anthropic is about how you can use your sub, banning people for even having OpenClaw mentioned in their commit history versus OpenAI just letting you sign in with codecs wherever you want, using your token however you choose.

14:28 This generally anti-user stance that Anthropic consistently takes is going to destroy them in the end. Posts like this weren't written for the users. They were written for Anthropic and their business partners, even though they are about the users. This is pathetic. They deserve the they're getting and they will continue to get more indefinitely until there's a fundamental change in how they interact with the community.

14:49 In the words of Matt PCO anthropic, please stop hitting yourselves. This stuff is pathetic and it needs to end. I've been trying to be nicer to y'all for a while because for the most part, you've been chiller, but things like this make it harder and I don't want this relationship to end. Fable's a great model. Let me defend it without having to defend the you're doing next to it.

15:08 Hopefully, this message gets through, although I know how this tends to go. Let me know what y'all think. And until next time, pray for my hand. Seriously, I'm so tired of this cast. You guys have no idea.