Considering all the problems they've been having with over-charging Claude Code users over the past few weeks it's the very least they could do. Max subscribers are hitting their 5 hour usage limits in 30-40 minutes with a single instance doing light work, while Anthropic have no support or contact mechanism for users that they respond to.
This hasn't been my experience either. I personally find the max plan is very generous for day-to-day usage. And I don't even use compact manually.
However, when I tried out the SuperPower skill and had multiple agents working on several projects at the same time, it did hit the 5-hour usage limit. But SuperPower hasn't been very useful for me and wastes a lot of tokens. When you want to trade longer running time for high token consumption, you only get a marginal increase in performance.
So people, if you are finding yourself using up tokens too quickly, you probably want to check your skills or MCPs etc.
As a regular user, I hit these walls so often. I am experimenting with local model and open code. I am hoping to see some good results with qwen3 coder
It's known that Anthropic's $20 Pro subscription is a gateway plan to their $100 Max subscription, since you'll easily burn your token rate on a single prompt or two. Meanwhile, I've had ample usage testing out Codex on the basic $20 ChatGPT Plus plan without a problem.
As for Anthropic's $100 Max subscription, it's almost always better to start new sessions for tasks since a long conversation will burn your 5-hour usage limit with just a few prompts (assuming they read many files). It's also best to start planning first with Claude, providing line numbers and exact file paths prior, and drilling down the requirements before you start any implementation.
> It's known that Anthropic's $20 Pro subscription is a gateway plan to their $100 Max subscription, since you'll easily burn your token rate on a single prompt or two.
I genuinely have no idea what people mean when I read this kind of thing. Are you abusing the word "prompt" to mean "conversation"? Or are you providing a huge prompt that is meant to spawn 10 subagents and write multiple new full-stack features in one go?
For most users, the $20 Pro subscription, when used with Opus, does not hit the 5-hour limit on "a single prompt or two", i.e. 1-2 user messages.
Today I literally gave Claude a single prompt, asking it to make a plan to implement a relatively simple feature that spanned a couple
different codebases. It churned for a long time, I asked a couple very simple
follow up questions, and then I was out of tokens. I do not consider myself to be any kind of power user at all.
The only time I've ever seen this happen is when you give it a massive codebase, without any meaningful CLAUDE.md to help make sense of it and no explicitly @ mentioning of files/folders to guide, and then ask it for something with huge cross-cutting.
> spanned a couple different codebases
There you go.
If you're looking to prevent this issue I really recommend you set up a number of AGENTS.md files, at least top-level and potentially nested ones for huge, sprawling subfolders. As well as @ mentioning the most relevant 2-3 things, even if it's folder level rather than file.
Not just for Claude, it greatly increases speed and reduces context rot for any model if they have to search less and more quickly understand where things live and how they work together.
I have a tool that scans all code files in a repo and prints the symbols (AST based), it makes orienting around easy, it can be scoped to a file or folder.
I am on $100 max subscription, and I rarely hit the limit, I used to but not anymore, but then again, I stopped building two products at the same time and concentrate to finish up the first/"easiest" one.
> you'll easily burn your token rate on a single prompt or two
My experience has been that I can usually work for a few hours before hitting a rate limit on the $20 subscription. My work time does not frequently overlap with core business hours in PDT, however. I wonder whether there is an aspect of this that is based on real-time dynamic usage.
When I used it before Christmas (free trial), it very visibly paused for a bit every so often, telling me that it was compressing/summarising its too-full context window.
I forget the exact phrasing, but it was impossible to miss unless you'd put everything in the equivalent of a Ralph loop and gone AFK or put the terminal in the background for extended periods.
However I run like 3x concurrent sessions that do multiple compacts throughout, for like 8hrs/day, and I go through a 20x subscription in about 1/2 week. So I'm extremely skeptical of these negative claims.
Edit: However I stay on top of my prompting efficiency, maybe doing some incredibly wasteful task is... wasteful?
You hit a vague, never-quite-explained 5h window limit that has nothing to do with what you're doing, but with what every user is doing together. It's totally not downtime, you're just "using it too much" and they're telling you to fuck off until the overall usage slows down.
The order of priority is: everyone using the API (you don't want to calculate the price) → everyone on a $200/month plan → everyone on a $20/month plan → every free user.
Let's be perfectly clear: if user actions had anything to do with hitting these limits, the limits would be prominently displayed within the tool itself, you'd be able to watch it change in real time, and you'd be able to pinpoint your usage per each conversation and per each message within that conversation.
The fact that you cannot do that is not because they can't be bothered to add such a feature, but because they want to be able to tweak those numbers on the backend while still having plausible deniability and being able to blame it on the user.
Instead, the little "usage stats" they give you is grouped by the hour and only split between input and output tokens, telling you nothing.
For the same reason they use "tokens" instead of kilobytes: so that you don't do the conversion yourself and realise that for example spending a million "tokens" on claude-opus-4.6 costs you anywhere from $10 (input tokens) to $37.5 (output tokens). Now, 1 million tokens sounds pretty big and "unreachable" until you realise that's about 4 megabytes of text. It's less than three floppy disks of data going back and forth.
Now let's assume you want to send a CD worth of data to Opus 4.6. 700 megabytes * $10 (price per million input tokens) / 4 (rounding down one megabyte to roughly 250k "tokens") = $1750. For Opus 4.6 to return a CD amount of data back to you: $37.50 * 700 / 4 = ~$6.5k.
A terabyte worth of data with a 50:50 input/output ratio would cost you $5.7 million. A terabyte worth of data with a 50:50 input/output ratio on gpt-5.2-pro would cost you $25.2 million. (Note: OpenAI's API pricing still hasn't been updated to reflect 5.3 prices.)
So we get layers upon layers upon layers upon layers upon layers of obfuscation to hide those numbers from you when you simply subscribe for a fixed monthly fee!
you can just watch the limit on the claude usage settings view.
itd be nice to know how much the session context window applies wrt token caching, but disabling all those skills and stopping sending a screenshot every couple messages gets that 5hour limit and weekly limit a bunch better
It hasn't always done this, it's a relatively recent problem in the last 1-4 weeks (roughly). (note: I'm on the $160AUD/mo plan, so I think that's $100USD).
“Light work” is a pretty bold statement my dude. I run max for 8+ hour coding sessions with 3-4 windows where I’m babysitting and watching the thing and I never even get session warnings. The only time I bump up against limits is on token hungry tasks like reverse engineering 3M+ LOC codebases or 5-6 agents generating unit tests in parallel. Something tells me that what you call “light work” is not remotely the same as what I consider “light work”
That would be heartening, if I wasn’t consuming tokens 10x as fast as expected, and they just had attribution bugs.
Do you have references to this being documented as the actual issue, or is this just speculation?
I want to support Anthropic, but with the Codex desktop app *so much better* than Anthropic’s combined with the old “5 back and forths with Opus and your quota is gone”, it’s hard to see going back
Yeah I think it's either a billing bug, or some sort of inbuilt background sub-agent loop gone wild inside Claude Code, if you have a look at recent issues on the Github relating to 'limits', 'usage', 'tokens' you'll see a lot of discussion about it: https://github.com/anthropics/claude-code/issues?q=sort%3Aup...
exactly my experience. i am on pro subscription and when coding with claude in console i can only do about 30 min of work before i have to wait for 4 hours. i actually code more by chatting with claude via poe.com than using the subscription that i have
pro is the $20, right? It runs out quickly, especially using opus. But what do you expect for that kind of money? For serious work at least Max $100 is needed.
I spend not even 20 monthly when using gemini cli.
Claude's subscription model is forcing you to space out equal sessions unnaturally during day - but you don't work like that. You work N hours of day and then go to have personal life/sleep - meaning you sometimes code more and sometimes you don't. Forcing you to do only little work and then lock you out for extra usage will make you pay much more when actually using the model much less - because the times you are not coding would even out or even benefit you in credits
Tl;dr - it's claude's greedy credit system that sucks, i am not going to pay $100 for overcoming these weird limits, i would much rather use gemini cli
Microsoft's everything is running into problems. Seriously though, it always seems like a dumpster fire over there but the last year is next level with all their Windows 11 problems, Windows update problems, Office365 outages, Copilot issues, Azure outages etc... not to mention the features no one seems to want.
Sounds like they've been using skills incorrectly if they're finding their agents don't invoke the skills. I have Claude Code agents calling my skills frequently, almost every session. You need to make sure your skill descriptions are well defined and describe when to use them and that your tasks / goals clearly set out requirements that align with the available skills.
I think if you read it, their agents did invoke the skills and they did find ways to increase the agents' use of skills quite a bit. But the new approach works 100% of the time as opposed to 79% of the time, which is a big deal. Skills might be working OK for you at that 79% level and for your particular codebase/tool set, that doesn't negate anything they've written here.
I have a skill in a project named "determine-feature-directory" with a short description explaining that it is meant to determine the feature directory of a current branch. The initial prompt I provide will tell it to determine the feature directory and do other work. Claude will even state "I need to determine the feature directory..."
Then, about 5-10% of the time, it will not use the skill. It does use the skill most of the time, but the low failure rate is frustrating because it makes it tough to tell whether or not a prompt change actually improved anything. Of course I could be doing something wrong, but it does work most of the time. I miss deterministic bugs.
Recently, I stopped Claude after it skipped using a skill and just said "Aren't you forgetting something?". It then remembered to use the skill. I found that amusing.
I have a couple skills invoked with specific commands ('enter planning mode' and 'enter execution mode') and they have never failed to activate. Maybe make the activation a very rigid phrase and not implied to be a specific phrase.
Re: Flynn effect reversal - The article attributes this entirely to classroom technology. Researchers have proposed multiple explanations including changes in education, nutrition, cultural factors, and media exposure. The researchers studying the US decline noted "there is no shortage of theories in the scientific community, including poor nutrition, worsening health, media exposures and changes to education" and emphasised that "not every domain is going down; one of them is going up" - spatial reasoning scores actually increased.
Re: Denmark's "Solution" - The article states Denmark has "taken tablets, laptops, and computers out of the classroom" since the 2025/2026 school year with "undeniable" results.... however Denmark’s phone ban is set to take effect from 2026, and it applies to “smartphones and other private electronic devices with access to the internet" Critically, "this does not include computers and tablets that are used as part of the instruction.", and - the policy was only agreed upon in September 2025 and isn’t fully implemented yet - meaning claims of "undeniable results" are premature.
The article is basically a summary of Dr. Cooney Horvath testimony in the US congress and it is linked in the article.
He talks about correlation vs causation in this part:
https://youtu.be/Fd-_VDYit3U?t=134
There is plenty of research starting from 60's , introduce technology in the classroom and the learning goes down.
reply