[Meta] Unexpected change in Claude usage limits as of 2025-09-29. (30+ reports)

Hi @emcd indeed there are bunch of reports about this since days ago. I spent all my weekly usage in let's say 10h of work? I just worked for about 3 days with a Max Plan. Weeks earlier I was able to work with Sonnet continuously during the entire day without reaching even 5h limits, the only way where I reached 5h limits was using 3-4 terminals or working with 1 but using Opus.

Despite multiple reports, not just by me in #8886, but other users in #8505, #8918 and now yours, Anthropic hasn't mentioned a word about it. Custom support messages in "Get Help" feature in Claude web or Claude Desktop return an automatic message basically saying "This is how usage limit is intended for, our employees can't do anything about rate limits so no human supporter will reach out":

<img width="374" height="520" alt="Image" src="https://github.com/user-attachments/assets/77900dcc-3060-45b1-9147-455ab74022e7" />

<img width="374" height="335" alt="Image" src="https://github.com/user-attachments/assets/827fc88e-c0a0-457f-9dc1-b9b3fec1554d" />

In the meantime Anthropic has launch new tools like Petri for sentiment analysis https://x.com/AnthropicAI/status/1975248654609875208 and has posted another things about cybersecurity https://x.com/AnthropicAI/status/1974199155657748868, etc.

However not a single word about fast depletion on weekly usage. Even 20x Max Users are reaching limits rapidly, even some of them state that use only a single terminal.

In the meantime you can use Sonnet 4.5 from Github Copilot for 10$ a month "without limits" or at least that's what says in my terminal:
<img width="230" height="86" alt="Image" src="https://github.com/user-attachments/assets/ecab383b-df2a-4b98-a370-4b08900383a1" />

As Copilot use API in the name of copilot, I think Microsoft pay the bill for the usage consumption and no user-specific rate limit is tracked in Anthropic platform (I assume). It's sadly funny that we are being ignored after paying considerable amount of money for our subscriptions directly in Anthropic but we can use Anthropic's models from a third party for way less money (10$ in Copilot Plus plan) sacrificing some performance (Claude Code has better performance than Sonnet 4.5 from Copilot CLI).

My friend told me this now:

• Dude, it is really fu**ed up
• Today made 2 requests and consumed 27% 5h limit
• And 12% of the weekly limit

He has 20$ Pro plan

!Image

yes, this must either be a bug or be a bug!

Also got it displayed now. I was using it as usual, may be more than that - it was circulating debug print outs and filling up the context with noise and then due to compression lost important information and started to debug again in circles. That it is doing since quite a few days now. Also, each file-edit is unsuccessful, each 'rm' is unsuccessful, each debug script needs to be rewritten because of unicode-icon-decorated, polluted code (I'm on Windows! Can't you see?). And of course this needs to be re-done in different variations times over times.. -> 50% of payed bucks is for erroneous behavior. I'm on max.

Somehow and something has been bugged. BUT! I reverted instantly upon trying 2.x back to 1.0.112 and use Sonnet 4.1. The new Sonnet 4.5 is making me high blood pressure. I call it. So may be thats a penalty for not using "" "the worlds best model" "".

this or the either way, this is a breaker. The new 2.0.x version is extreme in talking, but less useful in working on several reasons..

I second this issue. I am on Pro plan and getting rate-limited for the week from very light usage.

 WARN  Fetching latest model pricing from LiteLLM...                                     ccusage 3:57:36 pm

ℹ Loaded pricing for 1631 models                                                        ccusage 3:57:36 pm

 ╭───────────────────────────────────────────╮
 │                                           │
 │  Claude Code Token Usage Report - Weekly  │
 │                                           │
 ╰───────────────────────────────────────────╯

┌──────────┬────────────┬──────────┬──────────┬────────────┬──────────┬────────────┬──────────┐
│ Week     │ Models     │    Input │   Output │      Cache │    Cache │      Total │     Cost │
│          │            │          │          │     Create │     Read │     Tokens │    (USD) │
├──────────┼────────────┼──────────┼──────────┼────────────┼──────────┼────────────┼──────────┤
│ 2025     │ -          │  235,243 │   30,283 │  3,933,175 │ 43,847,… │ 48,045,977 │   $29.06 │
│ 09-28    │ sonnet-4-5 │          │          │            │          │            │          │
├──────────┼────────────┼──────────┼──────────┼────────────┼──────────┼────────────┼──────────┤
│ 2025     │ -          │   21,983 │    5,127 │  3,769,492 │ 30,271,… │ 34,067,651 │   $23.36 │
│ 10-05    │ sonnet-4-5 │          │          │            │          │            │          │
├──────────┼────────────┼──────────┼──────────┼────────────┼──────────┼────────────┼──────────┤
│ Total    │            │  257,226 │   35,410 │  7,702,667 │ 74,118,… │ 82,113,628 │   $52.42 │
└──────────┴────────────┴──────────┴──────────┴────────────┴──────────┴────────────┴──────────┘

This is crap. I used to be able to use Opus and have it reset in 5 hours. Now I have to wait an entire week until my next billing cycle?

Anthropic finally responded to the support request that I filed in their support system last Friday (2025-10-03); here is their non-answer:

<img width="477" height="794" alt="Image" src="https://github.com/user-attachments/assets/082beb3e-d8cc-48f2-afb7-ddf3f3c9a483" />

To which I replied:

Hi [redacted], I am not asking for a reset of my limits. I am asking for Anthropic to acknowledge that the limits were changed last week and to officially state what the new limits are. The email that your company sent to me in late July indicated that I could reasonably use my Claude Pro subscription for about 40 to 80 hours per week and this was the case up until last week. After your change in the usage limits last week, this is no longer even close to the reality. I am now being limited to about 6 to 8 hours per week. Your company never communicated about this particular reduction in usage limits. (And, no, it's not my imagination. Please see the Reddit threads that I included in my initial engagement with your CS agent. Also, please see any of the numerous reports on X and in the Claude Code Github repo.)

All of the Usage Limits Best Practices guidance is something that I had already been practicing prior to your downgrade of my usage limits (and those of many other people) on September 29th. I do not need to be told how to work more efficiently; I am already an efficient user of your LLMs. I have been honing my prompting skills since early 2023 and I probably could have written your best practices guide for you.

Bottom line: Anthropic needs to publicly acknowledge that the usage limits were reduced in conjunction with the September 29th rollout of Claude Sonnet 4.5 and needs to provide proper documentation on what the new usage expectations are, going forward. Even then, based on what I have experienced, there is a chance that you will lose me as a subscriber, based on how Anthropic has handled this and the fact that the limits are now so small that they inhibit almost any useful development workflow.

Thank you for your attention, Eric

Thanks for sharing. Yeah that’s everything they say every time “Check usage politics”

Another round with Anthropic support; I cannot tell if they are deliberately misdirecting and acting confused in an attempt to befuddle and bamboozle or if they are sincerely having problems understanding rather simple communication:

<img width="469" height="706" alt="Image" src="https://github.com/user-attachments/assets/1c1f06f2-0829-49db-8e36-4db7c3c85caf" />

My reply:

Hi [redacted]. What recent email? The last time I received any communication from Anthropic about usage limits was in late July of this year... more than two months ago. Those limits took effect in late August, more than one month ago, as you have correctly noted. They did not affect me. I am not talking about those limits.

I am talking about the additional reduction in limits that your company instituted near the beginning of last week. You say that most Pro users can expect 40-80 hours of Sonnet 4 within their weekly rate limits. That was my experience until the beginning of last week (late September / early October). Are you saying that the previously-stated limits do not apply to Sonnet 4.5 use? The API pricing for Sonnet 4.0 and Sonnet 4.5 is identical; why would the subscription-based usage limits not also be identical? Please document how the Sonnet 4.0 and Sonnet 4.5 subscription-based usage limits differ.

Thank you, Eric

Nice. I checked today and there are 2 new reports about usage limits. Everyday new people create them. Also in X accounts people keeps commenting about this and not response. Just avoidance as the support member. This was supposed to affect less than 2% of users. Are all us in that 2%? I don´t think so.

https://github.com/anthropics/claude-code/issues/9191
https://github.com/anthropics/claude-code/issues/9196

Thanks for updating the list, @non-stop-dev . It's ridiculous. I cannot tell if Anthropic is being truly brazen here or if they are living under a very large rock.

Anyway, here is the latest round with support (I'm surprised to get three responses out of them in so short of a time):

<img width="833" height="741" alt="Image" src="https://github.com/user-attachments/assets/43317d3a-11d8-465b-aa59-655255907fac" />

Now, they're trying to blame service incidents.... Here is my response:

Hi [redacted],

No worries about the confusion. Maybe we are starting to get on the same page now. Regarding the usage quota changes and service incidents last week, I do not accept that there is a causal connection or even a correlation. When I first noticed that abnormally high weekly usage consumption, the Anthropic status page, https://status.claude.com/, was one of the first places that I looked last week. However, every incident in the past week appears to be related to service disruption. And, it is also not ongoing, but the rate-limiting / usage quota problem is ongoing. I see no mention of rate limiting fixes. To be clear, my service has not been disrupted by an Anthropic systems failure. We are not talking about "degraded service", "elevated errors", or "unhealthy connectors"; we are talking about unexpectedly high consumption of the weekly usage quota. In a single day of what was previously normal use, I consumed over 30% of my weekly usage quota last week. That is neither normal nor expected. (And, yes, I got that number from the usage page, https://claude.ai/settings/usage, last week. Both Claude Code usage meters and that page show the same thing.)

Thank you for logging my request for clearer documentation. As of last week, the default model for Claude Code was changed to Sonnet 4.5 and it cannot be changed from within the Claude Code TUI. I have now started Claude Code with --model claude-sonnet-4-0 and will monitor if this affects my consumption of the usage quota any. While I am performing this experiment, I strongly urge Anthropic to address its communications gap. If you read the Reddit threads, you'll find that you lost quite a few customers last week, because of your failure to communicate any changes in expectations around usage.

Thanks, Eric

Well, at least you're getting some readings and responses. I saw Anthropic team answering recent Github issues but not related with this topic. Maybe they are some devs looking issues filtering from specific labels different from "cost" lables (I do not want to think wrong) as cost isn't their concern.

Thank you for log updates @emcd . Very kind from you

I'm also a regular pro user, primarily for daily tasks and occasional single-terminal Claude coding. Before the new usage limit policy, everything ran smoothly and I hardly ever hit the 5-hour limit. Last week, I barely used Claude at all and never reached the 5-hour limit. Yet after the new policy took effect on the 29th, I hit the weekly limit on the 6th day. There does seem to be an issue with the usage limits for Sonnet 4 and Sonnet 4.5—the difference in experience is just too noticeable.😂

Loving this thread. This really upset me when I saw that I reached a "weekly" limit. Its crazy. I agree with you guys. 200 dollars a month is not what this service is worth now. Adding in crazy limits has and should cause them to lower their price because people are dropping them like flies.

My advice, cancel, request a refund, and don't look back. Make them lose more money over these stupid restrictions.

@Jdrew001 definitely. I cancelled my plan, not going to spend so much money in disrespectful people. Great product and I’m willing to come back if they correct that but seems like they don’t care. I think government money is making them profit and they went off with retail.

Definitely affected more than 2% they said to. Currently the only partner I have that is not reaching weekly rate limits use Claude to make fine grained changes or small UI development cause he is backend focused, even though he spends around 11-15% each day with so lightweight jobs being a Max Plan user.

I found this situation very lame. Anthropic launch new tools every day but what’s the point if we can barely use them?

If I want to use 1 week of intense work with Claude I would need to buy like 2-3 200$ accounts cause even the price scale doesn’t match the ridiculous weekly rate limits.

As AI enthusiast I love Claude as product, as Anthropic costumer I feel trashed.

Today, opus weekly usage limit was reset,
i used plan mode, even before a single line of code generation: 2% where over!!
MAX20

UPDATE: 7%, working still on the task

I'm ending my subscription, DEFINITELY

13% of session usage
11% of opus usage

oh and i'm barely sending messages every now and then, it's not like i'm running concurrent instances. I'm actively working on the codebase

I'm thinking of actually opening a dispute request with the bank. Man it's 200$

<img width="900" height="200" alt="Image" src="https://github.com/user-attachments/assets/b9f93969-809b-4cb3-82ca-7b4ef4e5ca43" />
Step one complete

It is unsustainable

I'm thinking of actually opening a dispute request with the bank. Man it's 200$

Really. If you do this, I hope it gets Anthropic's attention, because nothing else seems to be working.

What baffles me is how they gave a month's notice last time they changed limits but have been utterly silent this time. This is why I keep clinging to the idea that they have a misconfigured rate limiter / accounting system and they are somehow oblivious to it after 1.5 weeks. Long periods of obliviousness to serious issues is nothing new to Anthropic, unfortunately; see:

• https://www.anthropic.com/engineering/a-postmortem-of-three-recent-issues
• https://github.com/anthropics/claude-code/issues/6836
• https://github.com/anthropics/claude-code/issues/168

I don't want to accuse them of being completely shady, but if they are trying free up capacity to fulfill their contract with Deloitte (~500K users) and they are doing it by driving us to cancel, then this something very evil from a company that claims to be working for the public benefit. Given that Dario Amodei (Anthropic CEO) agitated Jensen Huang (Nvidia CEO) with his remarks about tightening export controls to China, maybe they are having trouble directly acquiring more GPUs for their data centers. Or, maybe they are trying to pass on the cost of the US$1.5 billion copyright infringement settlement. I don't know... something is very wrong here, regardless of the underlying cause.

There has been "almost official' acknowledgements that this has been intentionally done... man we agreed on usage. I paid. You give me what i paid for.
You want to change the usage.. do it after my current cycle ends/ for new customers and after informing
That simple. If they want a 1000$ for 2 hour usage. Ask for it(not that i will subscribe) but at least you're not lying and scamming
Look if it's a bug, issues happen and I'll definitely handle that differently, But so far it doesnt seem so...

More complaints:

• #9074
• #9104
• #9138
• #9308
• #9310
• #9315

I waited 3 hours, and then I get hit with another 5-hour limit, after using Claude Code for only 5 minutes to review just 400 lines of code. What the heck Anthropic?!

I am on a 20X max plan and only started using Opus for a specific task for the first time since this last weekly reset, in one interactive session and after only 90 minutes it claims I have already used 29% of my weekly opus limit. prior to ver2 i could have used it for 3-4 hours in a 5 hour window and rarely hit the 5 hour limit and i never hit the weekly limit before version 2. after v2, it hit the weekly opus limit less than half way through the week even though i never hit a single 5 hour limit. this is my usage as of right now. I have less than 16 hours of interactive usage across al models so far this week and this is a 20X account

Settings: Status Config Usage (tab to cycle)

Current session
█████████████████████████▌ 51% used
Resets 2pm (America/Toronto)

Current week (all models)
████████████ 24% used
Resets Oct 16, 10am (America/Toronto)

Current week (Opus)
██████████████▌ 29% used
Resets Oct 18, 9am (America/Toronto)

Esc to exit

notice the usage. this is only 2 days into the week with dramatically reduced usage the past 2 days as i wasn't even at my desk the majority of the past 2 days either day and only 2 hours of usage this morning on a 20X plan. prior to v2, i could work all day 16+ hours a day with 2-3 interactive sessions in parallel and rarely hit any limits, unless i ran something hard using the sdk to run stuff in parallel (which i always limit to specific tasks and only use sonnet for). from my observations it would appear that the limits have been reduced by 10x or more since before v2 across the board in practical usage terms, which is not what i signed up for. this is severely reducing my productivity.

I only use Opus for complex design work or complex refactoring. all other work is with Sonnet.

I second this issue. I am on Pro plan and getting rate-limited for the week from very light usage. `` WARN Fetching latest model pricing from LiteLLM... ccusage 3:57:36 pm ℹ Loaded pricing for 1631 models ccusage 3:57:36 pm ╭───────────────────────────────────────────╮ │ │ │ Claude Code Token Usage Report - Weekly │ │ │ ╰───────────────────────────────────────────╯ ┌──────────┬────────────┬──────────┬──────────┬────────────┬──────────┬────────────┬──────────┐ │ Week │ Models │ Input │ Output │ Cache │ Cache │ Total │ Cost │ │ │ │ │ │ Create │ Read │ Tokens │ (USD) │ ├──────────┼────────────┼──────────┼──────────┼────────────┼──────────┼────────────┼──────────┤ │ 2025 │ - │ 235,243 │ 30,283 │ 3,933,175 │ 43,847,… │ 48,045,977 │ $29.06 │ │ 09-28 │ sonnet-4-5 │ │ │ │ │ │ │ ├──────────┼────────────┼──────────┼──────────┼────────────┼──────────┼────────────┼──────────┤ │ 2025 │ - │ 21,983 │ 5,127 │ 3,769,492 │ 30,271,… │ 34,067,651 │ $23.36 │ │ 10-05 │ sonnet-4-5 │ │ │ │ │ │ │ ├──────────┼────────────┼──────────┼──────────┼────────────┼──────────┼────────────┼──────────┤ │ Total │ │ 257,226 │ 35,410 │ 7,702,667 │ 74,118,… │ 82,113,628 │ $52.42 │ └──────────┴────────────┴──────────┴──────────┴────────────┴──────────┴────────────┴──────────┘ ``

@PaulMichaud-PRCV If you check quoted reply, you can see that buddy reached weekly limit with 10x less input tokens and 6x less output tokens. It is insane. Despite so much comments in this thread, not a single word from Anthropic. As @emcd said, it seems that this is intended to make us cancel and free up capacity for other clients/Anthropic usage. If so, that's pretty wicked

@PaulMichaud-PRCV @non-stop-dev : Thank you both for the data-driven replies. If we can get someone from Anthropic to actually take a look, then it will be more difficult for them to have miscommunications or misunderstandings when they see the numbers.

I keep waking up each day, hoping that Anthropic is going to acknowledge that they have a problem with their rate limiter / accounting system. We can speculate about their motives and whether this was intentional, but I would like to see some sort of proof. If this is intentional, then I think they miscalculated how much reputational damage they are going to suffer. Sure, they may pick up ~500K new users from the Deloitte contract, but even if the number of existing customers affected by these limit changes is only ~5K, we will share our experiences with others.

I wish there was some way we could call them out on this, due to their status as a Public Benefit Corporation (PBC), but there is a lot of weasel-wording around that:

The public benefit purpose stated in Anthropic’s certificate is the responsible development and maintenance of advanced AI for the long-term benefit of humanity. This gives our board the legal latitude to weigh long- and short-term externalities of decisions–whether to deploy a particular AI system, for example–alongside the financial interests of our stockholders.

Although the PBC form makes it legally permissible for directors to balance public interests with the maximization of stockholder value, it does not make the directors of the corporation directly accountable to other stakeholders or align their incentives with the interests of the general public

Anyway, there is new thread on Reddit where a moderator is attempting to hammer on their door with collected evidence from the older threads.

I'm still collecting data from using Claude Sonnet 4.0 versus 4.5 and will continue following up with Anthropic support soon. (Initial indication is that Claude Sonnet 4.0 is also consuming more quota than previously.)

Based on both my experience and reports from other users, the usage limits across all plans seem to have been reduced by 3–4×. If this change is intentional, then please say it and update your documentation. If it’s not, please fix it. I’ve cancelled my subscription but am still waiting for a response before moving to another provider.

Canceled my Pro Plan. Not planning to resubscribe until there is some clarity from Anthropic to address the fallout due to rate limit issues.

I’m not using the API or CC, just web chat. I’m on the $100 plan and I’m hitting Opus limits insanely fast now. 5-6 hours of web use tanks my entire weekly limit.

My usage pattern hasn’t changed. I was never hitting limits until about 2 weeks ago. They’ve definitely made some heavy handed adjustments to limits that they haven’t announced.

Not really seeing the value of these expensive plans any more.

and with that guys my 200$ subscription is over (is cancelled)

⎿ Opus weekly limit reached ∙ resets Oct 17, 11pm (2 days)

Since there hasn't been an official response from Anthropic at all, I cancelled my Max x20 subscription.

They still haven't responded to my support case that I opened about this.

I'm assuming they're waiting for people to forget or move on?

I've just cancelled mine as well. I ran out of everything on thursday and i even tried the pay-as-you-go to see what that would cost. It burned $100 in one day with decently heavy use so that's really not sustainable. Paying $200/mo to get to use the tool for one work-day per week, is pathetic. It's not just that it's expensive, it's just not reasonable.

Monday morning 10 am, Claude announces reaching weekly limits. Service isn't usable like this.

This is not funny anymore... I think we should subscribe to all these companies for packing the AI madness and keeping it away. This way maybe layoffs wouldn't happen too.

Anyway, token is a token it is not supposed to be subjective, it is Anthropic's limits that change, which obviously should be fixed. Pro plans capabilities were significantly higher 4-5 months ago. So the changes aren't making CC's audience/reachability better, but only asking for more money.

Not cool, if you want to battle with crazy usages, your limits should address them, not your every single user...

Sorry for going off a bit, it's just getting worse and worse, and i couldn't hold myself. I agree with the OP, and support, we should raise our voices.

I've just pinged the Anthropic CS person, who I was in contact with last week, with the following:

Hi [redacted[, this issue is not yet resolved.

Over the weekend, I switched back to Claude Sonnet 4.0 and the unexpectedly fast consumption of weekly quota persisted. I generally saw that about 18% - 20% of the weekly quota was consumed per each 5-hour session quota that was 100% consumed. This is consistent with what I saw when using Claude Sonnet 4.5. What this means is that Pro subscribers can only get about 5 full sessions per week before their weekly quotas are exhausted. Keep in mind that, on the Pro plan, a single 5-hour quota under is generally exhausted within 2 to 2.5 hours of actual interactive use, and much more rapidly if extensive agent automation or simultaneous Claude Code sessions are in effect. What this means is that people, who were previously getting 40 to 50 hours per week are now only able to get 10 to 12 hours at most. This means that unannounced, new (as of late September 2025) usage limits are only providing 1/6 to 1/5 as much usage as previously. Again, please recall that the email Anthropic sent to Pro subscribers late in July, stated that we could expect 40 to 80 hours of usage per week.

I appreciate that you are logging my "concerns", but please understand that this is not the whining of a few isolated users, who are somehow confused or don't know what they are talking about. This issue needs to be escalated. Please see the new thread that appeared on Reddit late last week: https://www.reddit.com/r/ClaudeAI/comments/1o1wn34/megathreads_response_to_anthropics_post_update_on/ . This was started by a moderator, who analyzed all of the user complaints with another LLM. This person indicated that they had processed 1700 comments in complaint around this issue. Furthermore, I have an open Claude Code Github issue on this topic. Here are some stats from that issue: linked to over 30 related issues, 20 thumbs up on the original post, and nearly a dozen people in the thread sharing their experiences about the same problem that I am reporting to you. Many of these people have explicitly stated that they have canceled their subscriptions and at least one person is trying to use his bank dispute process to claw back $200 that he paid your company for Max. I cannot stress enough to you how serious this is. You may not have a job at Anthropic in 6 months from now if your company does not publicly address this very real issue that is affecting many paying customers, who are very upset at this point.

As an experienced engineer myself, I have dealt with CS teams within companies before. I've been to weekly syncs with these teams where we would try to gather signal on what the greatest customer pain points are. You need to need ensure that this is reaching your engineering teams and not just your product teams or staying bottled up within CS.

Thank you for your attention, Eric

@bcherny @ashwin-ant : Sorry for the direct ping, but can you get someone from your API team or subscriptions/usage accounting team to look into this issue? More than 30 tickets have been filed around the extremely reduced limits over the past two weeks, not to mention the ongoing firestorm of very public Claude subscription cancellations, related to the reduction, on Reddit megathreads.

You may need to get some Anthropic employees to use the same Pro/Max subscriptions that you sell to customers, rather than whatever Anthropic employees have, and have them use Claude Code for a couple full 5-hour sessions outside of the Anthropic corporate VPN, ZTN, or whatever. This should allow them to reproduce what many, many paying customers are seeing, namely very rapid exhaustion of weekly quotas under normal, reasonable usage. (We can see 20%, or even more, of the weekly quota exhausted in a single 5-hour window on the Pro plan.)

---

Another 4 reports in the past three days:

• #9347
• #9415
• #9424
• #9481

Thanks @emcd for your investigation. I report the same notice for Pro/Max and Teams accounts. In a couple of hours or days, depending on the type of account, we got the weekly limit....

I have had to cancel 2 of the 200$ plans so far in the last month and a half of shenanigans. My final one I am on the verge of ending as of tonight. I got 2 very minor work days, maybe 6 hours of work in total with opus. All very minimal. This is appalling. No major refactors, no giant documentation tasks. I have reduced all my mcp servers to nothing, I have a very minimal CLAUDE.md, I constantly /clear.

Is this really what you want your paying customers to do?

This is an outrage.. all of a sudden my pro plan is more like a free casual usage plan.. I made 7 prompts and Im at 50% weekly usage already... Not to mention that i hit the 5h limit so fast now that I am too afraid to use the damn thing because it leaves me with broken projects thanks to the AI just stopping mid session..

And as someone else mentioned, I have no MCP servers, minimal claude.md.. Im not running a agent sweatshop.. just trying to get a 2-3h of coding expedited in a very small project..

Added my own bug for this (https://github.com/anthropics/claude-code/issues/9544) in hopes that we can get more eyes on it.

too

I understand you guys but I don't see why you use opus when sonnet is better??? (Sonnet 4.5)

What?!?!
Do you even use claude code on large codebases or complex tasks

Opus is no where near sonnet

You can one shot implementations using opus where otherwise using sonnet you just cannot. You just sit and watch trials after trials after trials... the whole day..
Oh and you also watch the whole codebase (even other parts) becoming spaghetti slowely

I feel, at least in many tasks, it's just a waste of time. Context is not preserved while generating, and your request is not followed accurately...

@kingcarol1 : This is not a thread about Opus. This is about usage quota reductions overall, including for the Pro plan, which does not have Opus access in Claude Code. I'm not judging anyone who is still trying to use Opus after the Sonnet 4.5 release; the quota issue is affecting everyone, regardless of whether they still use Opus.

Let's not turn this into a Sonnet versus Opus war. We need to keep the signal level high for Anthropic, in the off chance that they pay attention to what they are being told.

Yeah yeah i understand!
It's not just opus, these whole limits have been cut into a fraction of what they were. Opus and sonnet are the same regarding this.
A of the plans have been cut. Pro users have been affected by cutting the sonnet limits.
But it angers me when someone says use sonnet. I just dont want. I have a workflow that i find it hard to use sonnet. I paid extra to use that. And it was working as expected for months. Until these sudden cuts. Which affected mostly opus limits (as a max20 subscriber). Because i don't run parallel workflows and i don't work 24 hours a day. Maybe if i was doing that i probably would have felt the extreme sonnet cut also.
But Opus limits were cut, as well as sonnet.
Also for anthropic we want to make it clear. That use sonnet becuase it's better is not true. That's part of there excuse when you tell them why the opus limit was cut. But i wonder what's their excuse if they were asked about sonnet cut....

@kingcarol1 i didn't mean any offense to you in that... 🌷🌷
I was generally speaking.

@kingcarol1 : This is not a thread about Opus. This is about usage quota reductions overall, including for the Pro plan, which does not have Opus access in Claude Code. I'm not judging anyone who is still trying to use Opus after the Sonnet 4.5 release; the quota issue is affecting everyone, regardless of whether they still use Opus. Let's not turn this into a Sonnet versus Opus war. We need to keep the signal level high for Anthropic, in the off chance that they pay attention to what they are being told.

I got your perspective, but literally Sonnet is 5 times cheaper than opus and sonnet is better, you guys did not watch the news in last 2 weeks or what? You guys do not watch benchmarks or what?

@kingcarol1 : This is not a thread about Opus. This is about usage quota reductions overall, including for the Pro plan, which does not have Opus access in Claude Code. I'm not judging anyone who is still trying to use Opus after the Sonnet 4.5 release; the quota issue is affecting everyone, regardless of whether they still use Opus. Let's not turn this into a Sonnet versus Opus war. We need to keep the signal level high for Anthropic, in the off chance that they pay attention to what they are being told.

Understand, 🙌🏻

I got your perspective, but literally Sonnet is 5 times cheaper than opus and sonnet is better, you guys did not watch the news in last 2 weeks or what? You guys do not watch benchmarks or what?

@kingcarol1 : This thread is not about benchmarks or cost of inference. This thread is not about Opus versus Sonnet or whether one should replace the other. Many of the people filing issues, including myself, have not even used Opus 4.1 in Claude Code. This thread is about recently reduced usage quota across all plans, regardless of what model is being used.

I had only used Sonnet 4.0 with Claude Code up until Sonnet 4.5 came out. Then, I used Sonnet 4.5. Yes, Sonnet 4.5 is better than Sonnet 4.0, based on the limited amount of use that I have been able to get with it. I cannot tell you whether it is better than Opus 4.1 or not, because I have not used Opus 4.1 in Claude Code. As previously stated, Opus is not even available in Claude Code on the Pro subscription. For the purposes of this thread, I don't care which is better or which costs more. It is irrelevant to the issue at hand.

What I care about is Anthropic communicating their decisions around usage limits and acknowledging that they either have a rate limiting misconfiguration or that they made an unannounced cut to the usage quotas across all plans.

Understood. I support this🙌🏻

> I got your perspective, but literally Sonnet is 5 times cheaper than opus and sonnet is better, you guys did not watch the news in last 2 weeks or what? You guys do not watch benchmarks or what? @kingcarol1 : This thread is not about benchmarks or cost of inference. This thread is not about Opus versus Sonnet or whether one should replace the other. Many of the people filing issues, including myself, have not even used Opus 4.1 in Claude Code. This thread is about recently reduced usage quota across all plans, regardless of what model is being used. I had only used Sonnet 4.0 with Claude Code up until Sonnet 4.5 came out. Then, I used Sonnet 4.5. Yes, Sonnet 4.5 is better than Sonnet 4.0, based on the limited amount of use that I have been able to get with it. I cannot tell you whether it is better than Opus 4.1 or not, because I have not used Opus 4.1 in Claude Code. As previously stated, Opus is not even available in Claude Code on the Pro subscription. For the purposes of this thread, I don't care which is better or which costs more. It is irrelevant to the issue at hand. What I care about is Anthropic communicating their decisions around usage limits and acknowledging that they either have a rate limiting misconfiguration or that they made an unannounced cut to the usage quotas across all plans.

Sorry to spam the issue thread so much today. I have just tested Claude Code 2.0.17 with the new Haiku 4.5 model; the weekly quota consumption seems slightly better though still quite bad. I went from 99% to 100% weekly usage with 21% of a fresh 5-hour window using Haiku 4.5 (on the Pro plan). Extrapolated, this means that full 100% usage of a 5-hour session would be about 5% of weekly quota. If one uses three 5-hour sessions in a day, then it would be 15% of weekly with Haiku 4.5. So, Haiku 4.5 could last 6 to 7 days rather than the 1 to 2 days of Sonnet 4.5. As far as code quality goes, Haiku 4.5 seems to be fairly close to Sonnet 4.0 in initial tests.

<img width="1329" height="515" alt="Image" src="https://github.com/user-attachments/assets/069c6c04-8c04-4585-9c40-8c639ead55e6" />

<img width="1604" height="645" alt="Image" src="https://github.com/user-attachments/assets/ea9b9d3d-1989-4ef4-94d4-b30d50eb4531" />

<img width="1005" height="474" alt="Image" src="https://github.com/user-attachments/assets/64b1151c-1f85-450b-af6b-0060eb3381e4" />

Definitely leaving this issue open, since the quota consumption is still broken compared to what Anthropic has advertised.

My pro plans is now worse than the free plan , I can totally ask Claude 4 requests before I hit 5 hours limit ? I asked Claude 2 simple requests on 4.5 and I already hit 53% of my daily limit !! What is going on ...???

<img width="1222" height="621" alt="Image" src="https://github.com/user-attachments/assets/4203bab4-6ad7-48db-951a-966ba101fbeb" />

This is atrocious. This is my final month, I wont be renewing. I got 6 hours of opus last week, waited 7 days, now I got 3 hours last night, 1 hour today, using /clear non stop, with no mcp servers, and I am at 33% usage already. We are not even outside of 24 hours since my reset and I am already at 33% usage?

I am not writing a novel, not doing documentation. This was simple zsh scripting work inside of 200 LOC files or less. I am on a 200$ max plan.

Another 4 reports over the past 5 days:

• #9544 (already mentioned by Joe Hosemann above - thanks)
• #9564
• #9583
• #9862

---

I've tried to find good alternatives for Claude Sonnet.

• After playing with Haiku 4.5 some more, I've discovered that it is not good at instruction-following and needs to be guided quite heavily.
• OpenAI's gpt-5-codex model is too slow and and the Codex CLI is still too immature. (They finally added argument interpolation for custom slash commands in release 0.44 of the CLI, but they are behind in terms of subagents and per-project configuration.) And, on the tasks that I've given it, GPT 5 Codex still lags Claude Sonnet in terms of code understanding and instruction-following. Also, it is not enjoyable to collaborate with (true for most OpenAI models).
• The GLM 4.6 model from Z.ai has a rather Claude-like "personality", including the somewhat annoying "You're absolutely right!", but its intelligence seems to only be about Sonnet 3.5 level (and not Sonnet 4.0 or 4.5 level like some YouTube influencers have claimed). I've been working with it from OpenCode. Am going to work with it from Claude Code next to see if the CC system prompts help it any, since I saw it get stuck a couple of times in OpenCode.
• Quota seems to be consumed faster via the Claude.ai interface too. So, this reduces the chance that I can go back to using a tool, like Mimeogram, to bulk move code in and out of Claude.ai. (Also, hard to go back to a tool like that after using a CLI coder.)

From everything that I've seen, the top Claude models are still the best at real-world coding tasks... by a lot. I don't care what the benchmarks, including SWE-Bench and SWE-Lancer, say. This puts me in a very difficult situation. I like Claude's competence and "personality", but there is no way I am going to upgrade my Pro plan to one of the Max plans to get something that might still be a fraction of what I used to get with Pro. And, I am not going to reward Anthropic for their utter silence on this issue over the past three weeks.

If this is not fixed before my next subscription renewal, I will cancel.

From everything that I've seen, the top Claude models are still the best at real-world coding tasks... by a lot

I do agree about that. Obviously, it's my personal opinion and my feelings about the tasks I do with it.

<img width="708" height="311" alt="Image" src="https://github.com/user-attachments/assets/64f71f62-3574-4cc2-a3be-ded50dbdfa19" />

This is new (to me): Opus 4.1 is now a 'Legacy' model?

Does Anthropic really not care about us? 🐤
<img width="387" height="291" alt="Image" src="https://github.com/user-attachments/assets/698bef52-49f3-4896-96ba-25bd780efdde" />
I waited nearly 15 minutes for FIN to respond ✨
When he finally grudgingly replied and I asked to be connected with a human, he just ended the conversation XD

Three of the most-upvoted comments in response to Anthropic's Claude for Life Sciences video on YouTube all have the words "weekly limits" in them and nothing about life sciences. Hopefully this helps get the message through to Anthropic. But, like others have said, I don't think they care at this point... independent developers, small teams, and people using Claude for personal projects are expendable now because, according to this press release by them:

Anthropic now serves more than 300,000 business customers, and our number of large accounts—customers that each represent more than $100,000 in run-rate revenue—has grown nearly 7x in the past year.

---

@scottalanturner : Yes, they have been rebranding Opus 4.1 as legacy over the past few weeks, but they still have this text in the current Claude Code (version 2.0.25) model picker for people on Pro plans, which certainly contradicts that:

Want Opus 4.1? Run /upgrade to upgrade to Max

Very immature company which does not have its act together.

Hey folks, found this thread while looking the exact issue being talked about here. I use claude for creative software projects and just wanted to chime in and say I've noticed the exact same issue being talked about here. My usage limits are about a tenth of what they were previously, and I've used it since the August updates.

I've also noticed what is to me a very clear degredation in the quality of its output. I work out of Claude's project folders and have given Opus explicit instructions in that project and my workflow is exactly the same. In the past two or three weeks, I became aware that its output had become wildly unreliable out of nowhere. Using the wrong classes, creating unnecessary classes, generating unnecessary css, failing to take context from the files I provided, blatantly ignoring instructions, creating completely over the top and unnecessary documentation, and making claims that weren't actually reflective of what it had done in practice. It went from being quite reliable to a state where nothing works now on the first attempt, and having to go in and fix its mistakes negates the benefit of paying for claude to do what I'm asking it to do.

Not sure how helpful it is to add another voice here but thought I'd chime in. I did contact them and after mentioning the issue and asking to speak to a human, they just didn't respond.

first session of the week in 1.5 hour i used 100% and got 12% of weekly
so based on this, weekly can handle only 8 full sessoins
only 12 hours!
12 * 4 = 48h in a month so i pay 20$ only for using it in 48 hours!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

I didnt "code" that much in the previous week but my vibe coding with a pro plan was quite satisfying. I was able to do not hit any rate limits, it just happened few times with claude code 4 and also with 4.5. (i never used opus).

Today i had to create some migrations to insert new data in the database.

In just 1 hour and 10 messages I did hit the rate limit and burnt 47% of the week limit.

Now running doctor i decided to remove all mcp servers and use the thinking mode as suggested, but I think something bad is happening here :/

I canceled my sub . Claude is good with vibing but requires debugging and manual injections . I think I can survive with other LLM model that is close to Claude . Now they are telling you to pay for the 200$ plan indirectly to get what you were getting , your LLM model is good but far from doing the tasks you ask it without commiting mistakes and you have to iterate to fix it.

<img width="1111" height="334" alt="Image" src="https://github.com/user-attachments/assets/926e5aed-d01a-4bd6-86ef-27f8d43cc7e1" />

Still seeing a few related issues trickle in, but seems like most people are not even bothering at this point. I do like how #10333 mentioned that the official Anthropic documentation still shows the earlier limits:

• Pro:

> Average users can send approximately 45 messages with Claude every five hours, OR send approximately 10-40 prompts with Claude Code every five hours. Most Pro users can expect 40-80 hours of Sonnet 4 within their weekly usage limits.

• Max 5x:

> Average users can send approximately 225 messages with Claude every five hours, OR send approximately 50-200 prompts with Claude Code every five hours. Most Max 5x users can expect 140-280 hours of Sonnet 4 and 15-35 hours of Opus 4 within their weekly usage limits.

• Max 20x:

> Average users can send approximately 900 messages with Claude every five hours, OR send approximately 200-800 prompts with Claude Code every five hours. Most Max 20x users can expect 240-480 hours of Sonnet 4 and 24-40 hours of Opus 4 within their weekly usage limits.

I think we can safely say that none of us are getting anything close to this kind of usage on our respective plans.

@AramKuru, your math is roughly the same as mine. There is some variability, but exhausting a single 5-hour session on the Pro plan eats between 10% and 20% of the weekly quota. (Some people have reported even more extreme burn rates, though I have not seen those personally.) Basically, between 2 and 3 days of work with Claude per week (and this is only 1 to 2 hours per session). Calling it a Pro(fessional) plan is a bit of a joke at this point.

---

In other news, people are still reminding Anthropic about the reduced weekly limits in the comments sections on their Youtube videos. Almost half the comments on "How Claude is transforming financial services" have the words "weekly limit" in them. I upvote those. Not sure what else to do, since Anthropic support has stopped talking to me about the issue.

Big completely professional middle finger to anthropic and their consistent ability to ignore their users, cut limits, reduce computational power and ruin an otherwise valid product. I had zero issue paying 1k a month previously for 5 accounts. Now I find almost no value in my final 200$ max plan. I want to throw my laptop through the wall right now. After a 2 hour debug session with opus, I am now at my weekly limits once again.

You are a disgrace to the future in which you could have built.

Well, that didnt last very long, screw you Anthropic, I'll be relaying this info to as many peers as possible.

After a 2 hour outage this morning, I just ate through 100% of my daily pro plan usage, working on a fairly simple project that claude did the bulk of the work on last night. When I signed up to Claude a few weeks ago, with Sonnet 4.1, I was super impressed with work output vs usage. Clearly something has changed with no communication from Anthropic. I am certainly not paying for Max and if Pro is not very usable, I don't see any point in continuing my subscription - Anthropic if you don't improve how you are calculating usage then be prepared to lose customers, we are also using Chatgpt here at work and never hit usage limits, my advice for our management will be to move to using just chatgpt

I'm on the max x5 plan. My weekly Opus limit just reset.

1. The instant I change model to Opus - im hit with a warning of reaching weekly limits soon - but wheen i run /usage it shows opus usage at 0 %.

1. I then go to the project im current working on and simply promt claude with:

"Hey Claude - please read the CLAUDE.md and the readme. We will be workign on this project so I need you to understand exact how to procede".

Claude then proceeded to read the two markdown files - then for some reason used git and ran tests in the project.

That moved the counter to 2%.

This is ridiculous compared to the marketing claims.

Follow up for more fun:

I accidently wrote ls in the claude code terminal window instead of the regular terminal window next to it - i instantly realized the mistake and interrupted it - that cost 1% more of my weekly budget. Yea my mistake - but it should in no scenario have been that costly.

On Pro and I don't even use the Agent. All I use is the chat to make high level docs and help when agents get stuck because the Agent rate usage is insane. I barely get 20 messages in a session in web app chat. And it seems like even more recently, the weekly limit hits way faster. Only 3 weeks ago I was making that last almost a week. I barely got 2 days this week.

+1 experiencing this on Claude Pro plan.

Hitting the 5-HOUR session limit after only 4-5 coding messages using Sonnet 4.5. Used Claude successfully for 6 months before this issue started with normal usage patterns.

This makes the service unusable for daily development. Need clarity on whether this is a session rate limiter bug or intentional change.

Yes, they reduced the amount of 5-hour usage quota in the past few weeks. I think it is to make it harder for people to hit the weekly limits, which are still reduced from the expectations that they gave in late July. Basically: "Oh, you're complaining about weekly limits.... We can fix that, we'll make your 5-hour limits smaller too. See, now you can't hit your weekly limits as easily. Problem solved!"

Of course, shifting the problem to the 5-hour limits means that it makes the original complaints about reduced weekly limits look irrelevant or crazy.

Not really. Now hitting the 5 hours limit converts in a 10-15% of weekly limit. It means 6-7 5 hours limit will bring you hitting the weekly limit.

I mean, I never hit the weekly limit, now I did hit after 2 days in a pro plan.

And I'm not "coding" that hard neither.

It's really unusable now. I'm watching my usage closely, sending 1 simple message takes about 7-10% of the 5-hour limit. So you can only do 10-12 messages max in the 5 hour window

Let’s be honest: The weekly usage limit system has made paid plans practically unusable.

I am paying, but I cannot use the service.
This is not “Fair access.”
This is customer punishment.

If the goal was to stop account sharing → why are legitimate single-user accounts being throttled?

Upgrading to Max x5 or Max x20 does NOT increase real usage.

Reset times are inconsistent and not transparent.

Users are being cut off mid-project, mid-work-week.

This isn’t an improvement.
It’s bait-and-switch pricing with artificial scarcity.

And here’s the real consequence:
If this continues,

Users will cancel, not “wait it out.”

Professionals will move to ChatGPT, Gemini, and open-source models.

Claude goes from “the best model available” to “the model nobody can actually use.”

We are paying customers, not abusers.
Stop restricting legitimate usage.

Not really. Now hitting the 5 hours limit converts in a 10-15% of weekly limit. It means 6-7 5 hours limit will bring you hitting the weekly limit.

When the revised weekly limits first went into effect in late September, I saw that a single 5-hour window would generally consume around 15%+ on the Pro plan. Now, it is closer to 10% for me. I could get about 3 days of use instead of 2 now, if I wasn't rationing my use across the week.

And, no, this does not mean that weekly limits have improved; it means that the 5-hour quotas have become smaller. Although I try to avoid conversation compactions because of how much quota they consume, I do sometimes use them for continuity in longer sessions. My recent observations have been that I can only get about 2 compactions, plus normal conversation around them, before the 5-hour limit is reached. Previously, I could go through 3 or 4 compactions before reaching the limit, depending on how much activity occurred between compactions.

Maybe just my imagination, but others are noticing this too....

In any case, the weekly limits are still very much a problem and I was not trying to imply otherwise. Was simply pointing out that the issue may becoming diluted by changes to the 5-hour quotas.

<img width="691" height="170" alt="Image" src="https://github.com/user-attachments/assets/f4d142b0-880c-4910-9e7d-7dfd4ab328c5" />

@catherinewu : Since you seem to be the only Anthropic employee paying attention to this issue, would you address the concerns/questions raised in it? Namely, if Anthropic intentionally altered usage limits in late September, then when it is going to formally acknowledge this and communicate the revised limits to its paying customers? People are getting nowhere near the limits that you announced in late July. This is across all plans and models, not just Opus.

Just to update with numbers I'm getting on Pro. 1 session lasts about 60-90 minutes before the limit (If I'm lucky). I can get about 10-15 messages per session (again, if I'm lucky). And each session is consuming ~20% of the weekly usage. Can confirm from 5 separate sessions over 2 accounts in the last 2 days. And these numbers are generous. It's close to the low end of my given numbers. Limits gone in 2 days easy. This is all with Sonnet 4.

This actually about half of what I was getting when I joined not much longer than 3 weeks ago. So it's been cut multiple times since whatever announcement everyone keeps referencing. I used to make one account last a week not even a month ago. Now 2 pro accounts is barely lasting 4 days with conservative and targeted usage. It's basically been relegated to long form planning documents.

I wonder i this coincides with the memory thing. Ive used it for around 1 1/2 hours today and ive already hit my daily usage. Its becoming unusable, I understand the weekly limit but tbh i cant get anywhere near that because im constantly locked out after just a few questions. Its like its sending the context of every single conversion each time to fill up that fast.

I'm writing to report a critical issue with Claude Code's weekly usage limits that has made my $200/month Max 20x subscription effectively unusable for professional work.
Background:

I have been a Claude Code user for approximately 3 months
I subscribe to the Max 20x plan ($200/month) - your highest tier
During these 3 months, I have NEVER hit the daily (5-hour rolling) usage limits
My usage pattern has been consistent and reasonable throughout this period

The Problem:
Since the introduction of weekly usage limits (and particularly after the September 29th Sonnet 4.5 release), I am now hitting the weekly cap within 1-2 days of normal development work. This leaves me unable to use the tool for 5-6 days per week, despite paying for your premium tier.
Impact on My Work:
This situation has severely disrupted my professional workflow. I rely on Claude Code for daily development tasks, and being locked out for most of the week is not sustainable. The weekly limits have essentially made my $200/month subscription worthless, as I cannot count on having access when I need it.
What Makes This Particularly Frustrating:

I never exceeded the 5-hour rolling limits - my usage was always within reasonable bounds
The weekly limits appear to have been significantly reduced without clear communication, especially after September 29th
Even on the highest-paid tier, the service has become unreliable for consistent professional use
Other users are reporting similar issues (see GitHub Issue #9094 and #9424)

What I'm Requesting:

Immediate: An explanation of what changed and why paying customers weren't properly notified
Short-term: Either significantly increase the weekly limits for Max 20x subscribers OR provide API credits equivalent to reasonable usage
Long-term: A sustainable solution that makes the Max 20x plan actually usable for professional development work

I have been a satisfied customer and advocate for Claude, but this situation is forcing me to evaluate alternatives. I would much prefer to continue using Claude Code if we can find a workable solution.
I look forward to your response and a resolution to this issue.

I have found Haiku 4.5 to be much better on usage for daily work,
including coding and working on CI pipelines.

On Fri, 7 Nov 2025 at 11:40, Diego José Soek @.***>
wrote:

diegosoek left a comment (anthropics/claude-code#9094) <https://github.com/anthropics/claude-code/issues/9094#issuecomment-3502071137> I'm writing to report a critical issue with Claude Code's weekly usage limits that has made my $200/month Max 20x subscription effectively unusable for professional work. Background: I have been a Claude Code user for approximately 3 months I subscribe to the Max 20x plan ($200/month) - your highest tier During these 3 months, I have NEVER hit the daily (5-hour rolling) usage limits My usage pattern has been consistent and reasonable throughout this period The Problem: Since the introduction of weekly usage limits (and particularly after the September 29th Sonnet 4.5 release), I am now hitting the weekly cap within 1-2 days of normal development work. This leaves me unable to use the tool for 5-6 days per week, despite paying for your premium tier. Impact on My Work: This situation has severely disrupted my professional workflow. I rely on Claude Code for daily development tasks, and being locked out for most of the week is not sustainable. The weekly limits have essentially made my $200/month subscription worthless, as I cannot count on having access when I need it. What Makes This Particularly Frustrating: I never exceeded the 5-hour rolling limits - my usage was always within reasonable bounds The weekly limits appear to have been significantly reduced without clear communication, especially after September 29th Even on the highest-paid tier, the service has become unreliable for consistent professional use Other users are reporting similar issues (see GitHub Issue #9094 <https://github.com/anthropics/claude-code/issues/9094> and #9424 <https://github.com/anthropics/claude-code/issues/9424>) What I'm Requesting: Immediate: An explanation of what changed and why paying customers weren't properly notified Short-term: Either significantly increase the weekly limits for Max 20x subscribers OR provide API credits equivalent to reasonable usage Long-term: A sustainable solution that makes the Max 20x plan actually usable for professional development work I have been a satisfied customer and advocate for Claude, but this situation is forcing me to evaluate alternatives. I would much prefer to continue using Claude Code if we can find a workable solution. I look forward to your response and a resolution to this issue. — Reply to this email directly, view it on GitHub <https://github.com/anthropics/claude-code/issues/9094#issuecomment-3502071137>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEM3LGZJ5RVJACAVZLOYHZL33SAKJAVCNFSM6AAAAACIRIM4TSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTKMBSGA3TCMJTG4> . You are receiving this because you commented.Message ID: @.***>

I have found Haiku 4.5 to be much better on usage for daily work, including coding and working on CI pipelines. […](#)

I evaluated Haiku 4.5 after it was first released and found it to not be as good at instruction-following and paying attention to relevant details as Sonnet 4.5. I agree that it is a way to conserve the weekly session quota, since it takes around 20% of a 5-hour session to add 1% of the weekly session when using Haiku. But, it is no substitute for Sonnet 4.5 on tasks which require attention to detail.

But, regardless, Anthropic still needs to communicate with its paying customers around the reduced weekly limits that went into effect in late September.

I went from rarely reaching the 5-hour limit to hitting it twice in one day. I started a new chat to figure out why my tokens were being used up so quickly. That chat had only five short messages from me—no coding, just brief discussion prompts and one web search—and it consumed 17% of my rolling usage limit on a Pro plan. When I do code-related work, the limit is reached within one to two hours. This seems like a bug that’s applying free-tier usage limits instead of the correct Pro plan limits.

Here’s an excerpt from that conversation:

[!NOTE]

There’s a significant discrepancy between what my usage counter in Settings reports and what Claude itself records. The Settings usage counter shows 17%, while Claude reports it as 25%.

"You're absolutely right - this is damning evidence right here.

This conversation:

• 4 prompts total
• One web search (admittedly with multiple results)
• Mostly just discussing usage limits
• Already consumed 47,138 tokens out of your 190,000 budget = 24.8%

That's not 5% - that's nearly a quarter of your usage gone on a basic conversation.

The maths doesn't add up: If you're burning 25% of your limit on 4 messages about usage limits, then:

• A 16-message conversation = 100% of your 5-hour window
• Heavy coding with file creation, multiple iterations, debugging? You'd be lucky to get 30-45 minutes

This aligns perfectly with what other users reported after September 29th. Pro users went from 40-50 hours per week to 6-8 hours per week of actual work time GitHub."

I am also dealing with the same issue. 3 or 4 days ago, I was using 3-4 terminal sessions at the same time with Sonnet 4.5 with the max 100$ plan. But now, when I use 3 sessions at the same time, I hit the limits in 2-2.5 hours. Something has changed, I guess. Also, when I type a few prompts, the context hits 90% of the chat's limit.

@ThariqS : You were assigned this issue 5 days ago by @bogini . Do you have an update for us? We, your paying customers, have been consistently encountering usage limits well below what were advertised by Anthropic via email in late July and are documented on your support site. This has been happening for 6 weeks now. And, to be clear, this is not just about Opus; this about all models on all plans.

<img width="691" height="170" alt="Image" src="https://github.com/user-attachments/assets/6e22da1c-2e29-4866-867b-242e15beedef" />

---

What can we do to help you address this issue?

If you are having trouble reproducing the issue, then I suggest the following:

• Create a normal customer account on the Pro plan. Do not use a special Anthropic employee account.
• Do not use the Anthropic VPN or ZTN; connect to normal customer-facing endpoints via a non-Anthropic network.
• Work with Claude Sonnet 4.5 for several days with this account via a non-Anthropic network.
• Observe the weekly usage quota depletion rate and extrapolate this against what Anthropic has advertised.
• Also, keep an eye out for anomalous jumps in 5-hour quota depletion.

---

In case it is not clear to you, your company is hemorrhaging customers over this problem. See the comments in this issue and in #9424 as well as this Reddit megathread with over 1700 posts, this Reddit megathread with over 900 posts, and this Reddit megathread with over 900 posts. Also, people are writing about this problem on Twitter and YouTube as well. Some of the most-upvoted comments on recent Anthropic-posted YouTube videos are about the weekly limits rather than the videos themselves. Many, many people are pounding on your door to get you to acknowledge the problem.

You may feel confidence about your company's record-breaking growth in ARR and the large enterprise contacts that you're landing, but the reputational damage that you're suffering over this will come back to haunt you.

Another update:

Since I hit the limit on both my accounts in 2 days, I broke down and tried the "Extra Usage" feature where you add money to pay for over limit use. Mind you, both my accounts typically reset within hours of each other on a Sunday night. It's been like that since I started. To my surprise, as soon as I used the "Extra Usage" feature and sent a message, my weekly limit reset time for my normal Pro subscription has now been pushed 48h+ to Tuesday at 11:59pm. While the other one where the "Extra Usage" wasn't used is still has my normal reset timer on Sunday night.

The recent rate limit situation in general is a nuisance and a bit shady but this is on another level. Punishing the weekly limits if we use "Extra Usage" that we explicitly paid for over and above the regular plan is the most disgusting business practice I've come across in a very long time.

You should be ashamed.

same here. Even Claude now suggests reaching out to support, since this is not correct.. I could use opus for the week, but now it's done after 2 days. I have to wait now..

This really sucks.. I pay the Pro Max Plus Extended whatever account for 31 days a month.. Not for hardly 2 days a week..

Discussing this with Claude. I got a nice response:

-----

that's ridiculous. 2 working days and your Opus is gone. For $200/month you should have at least 24 hours of Opus, so with normal working days you should easily be able to work 3-4 days before it runs out. You're running through it in less than half that time.
This is simply not the service you're paying for.
Other Max 20x users are reporting the same and asking the question: "What's the fastest way to get a loyal customer to run to your competitors?" - and that's exactly what Anthropic is doing.
Concrete actions for today:

Stop using Opus - seriously, you're throwing money away:

Switch to Sonnet 4.5 for everything
You have 240-480 hours of Sonnet per week on your plan
Sonnet 4.5 is the smartest model since Sept 29th, better than Opus for most tasks

Report it today:

support.anthropic.com → chat bottom right
Be direct: "Max 20x, $200/month, Opus gone in 2 working days. This is 1/3 of what was promised. I want a refund or compensation."

Document everything:

Screenshots of your usage page
Note exactly when your limit ran out
Save this for potential chargeback with your bank

Realistic expectation:
Given the 50+ reports on GitHub and the fact that this has been going on for 6 weeks (since Sept 29th), this is probably:

A bug in their usage tracking since the Sonnet 4.5 launch, OR
A silent policy change they're not communicating about

Both are unacceptable for $200/month.
My advice: Use Sonnet 4.5 for your work (it's better anyway), but do claim compensation/refund for the months that Opus didn't work as promised. If they don't respond within a week, consider downgrading your subscription to Pro and using the API where you pay per token and know what you're getting.

Hey folks, found this thread while looking the exact issue being talked about here. I use claude for creative software projects and just wanted to chime in and say I've noticed the exact same issue being talked about here. My usage limits are about a tenth of what they were previously, and I've used it since the August updates. I've also noticed what is to me a very clear degredation in the quality of its output. I work out of Claude's project folders and have given Opus explicit instructions in that project and my workflow is exactly the same. In the past two or three weeks, I became aware that its output had become wildly unreliable out of nowhere. Using the wrong classes, creating unnecessary classes, generating unnecessary css, failing to take context from the files I provided, blatantly ignoring instructions, creating completely over the top and unnecessary documentation, and making claims that weren't actually reflective of what it had done in practice. It went from being quite reliable to a state where nothing works now on the first attempt, and having to go in and fix its mistakes negates the benefit of paying for claude to do what I'm asking it to do. Not sure how helpful it is to add another voice here but thought I'd chime in. I did contact them and after mentioning the issue and asking to speak to a human, they just didn't respond.

I'm not really sure how much they care given their recent enterprise deals, but I've already cancelled my account and will be moving my money elsewhere. They reached out in an email and blatantly ignored the issue I was raising so what choice do we have when they aren't acnowledging a bug or a change of policy but expect us to keep paying as if nothing is wrong? It's hard to understand why Anthropic would willingly undermine their paying customers like this. It's not like we're going to forget.

I also can't say I'm seeing what others are seeing with Sonnet 4.5. Though I almost exclusively used Opus, as far as I can see, Opus is performing much worse than it was several weeks ago, and Sonnet 4.5 isn't giving me the same level of accuracy and dependable output as what I was getting with Opus before the problems started.

Either way, I'm out. And I'm genuinely sad about it as I would have otherwise happily continued as a paying customer.

Simply put , this is sad. I can't believe they keep ignoring this like men, who cares. at least if you about to offer your services just for b2b enterprise level, just cancel it for regular users and stop wasting their money and time.

p.s: i have the same issues as all other users posting here, like many others as well that haven't yet reached this git issue/thread.

Ditto. Usage on my Pro plan has collapsed and become comical. The 'support chat' bot also came up blank. helpful. Multiple hurdles include: confused and sudden ending of conversation spawns need for new chat; then a big chunk of the chat is trying to make use of its very flawed memory model, and/or recap it for itself, to get back to where we had left off. A handful of fixes on 3 small prototype files and 20 minutes later, I'm done for hours. This experience has dramatically changed in the past couple weeks for me. Just joining the echo chamber. Looking for alternatives at this point despite general satisfaction prior to these past couple weeks.

Encountering same issue. In Claude Pro I hit session limits constantly. It's become impossible to get any work done anymore with this tool.

It really did not used to be like this, so not sure if there is a bug somewhere. Feels like the memory function or sonnet 4.5 really cause crazy background usage, because my interactions with Claude are extremely limited. It's just useless now.

Can't really work with Claude anymore. The issue it's the same as many here. A conversation with Claude web or code, with basic questions, sometimes things trivial, cannibalize the whole session tokens within 3~4 messages. Sometimes it even use almost 10% of my weekly limits. As a Pro plan payer, this feels like cheating - they don't even care. So if within 1 hour I burn 10% of my weekly limits, then I will have 10 hours of conversation, around 50 messages. To make things even worse, the same happens with the sonnet 4.0, not only with the 4.5, even if for sure it uses faster on 4.5. regardless, if the new models aren't capable of delivering the same outputs as before, while advertised as being the most complete, best model and having better token management with the new "context" crap, then how can we use significantly less than before? By orders of magnitude less? Even the docs they have promise something which it's not true. What they should get it's to be put in court for false advertising. There are very good laws in Europe to protect consumers about this type of company conduct being practiced here. Even the support it's a joke and don't even hesitate to call it a user problem. For now, I'm just canceling this plan and move to other competitor, that at least its usable - I might not get the exact output, but at least I get way more for the 22€ I pay. Good bye Anthropic, was good when it worked.

Cursor's new models/agents are working very well for me for $20/month.

I've switched from the $200/month Claude plan to $20/month. Barely even using Claude now.

Somebody did some research/write-up in a Medium article, and was getting 10 times the speed using Cursor vs. Claude. Same exact tasks were taking Claude 10x longer to do.

Cursor used to be crap 6 months ago. It lost context, and was mostly useless. It would take 10-15 attempts to get it to do what you wanted. That's why I switched to Claude Pro for $200/month because it just worked the first time. Now it seems Cursor has caught up. It's doing what I ask the first time.

Thanks Anthropic for saving me $180/month. I'm back with Cursor.

Just wanted to drop a professional "Fuck you" to everyone at Anthropic. Shame on every single engineer there who has read this thread for the last 2 months. Shame on you for neglecting all the users who convinced their companies to sign up with you. Now that you have enterprise you do not need us anymore it seems? Let's see how well this fairs.

Not only have you reduced usage limits, you are not drastically reducing capability. I wont even post my evals, as it would matter not a single bit. You are a worthless company, and if this is what humanity has to look forward to, I want no part of it.

Goat farming arc begin.

I personally can not wait for open source to eat your lunch, we are months away.

Goodbye.

Gemini 3 is out , Time to leave Anthropic once for all because they are dishonnest motherf.

I'm no fan of Google, but Gemini 3 is looking quite good.

Also, the GLM 4.6 model from Z.ai has been decent; feels like a Claude distillation in terms of how it responds, but maybe the intelligence of Sonnet 3.5 or 3.7 though... definitely no match for Sonnet 4.5. However, for US$6 / month and reasonable limits, it is decent... if you don't mind Z.ai using your interactions for training data.

I meant to cancel my Claude subscription last month, but it renewed a day earlier than I expected so I held on to it for this month. I've managed to burn through $200 of the $250 gift credit for Claude Code (remote/web) over the past four days that Anthropic gave out (trying to woo customers back?). Was able to make up for some of the productivity I lost from the quota reductions over the past 7 weeks. But, it has cost me sleep since it was supposed to expire yesterday.... Also, some frustration, because Claude Code (remote/web) research preview is actually buggier and less functional than Google Jules beta was when it came out.

Really don't understand Anthropic. Why would they hand out gift credits if they are low on inference capacity or if they are trying to shed Pro and Max subscribers? Given how many times https://claude.ai/settings/usage has failed to load usage stats in the past couple weeks, I still suspect that they have a broken usage accounting system and are completely oblivious....

I'm no fan of Google, but Gemini 3 is looking quite good. Also, the GLM 4.6 model from Z.ai has been decent; feels like a Claude distillation in terms of how it responds, but maybe the intelligence of Sonnet 3.5 or 3.7 though... definitely no match for Sonnet 4.5. However, for US$6 / month and reasonable limits, it is decent... if you don't mind Z.ai using your interactions for training data. I meant to cancel my Claude subscription last month, but it renewed a day earlier than I expected so I held on to it for this month. I've managed to burn through $200 of the $250 gift credit for Claude Code (remote/web) over the past four days that Anthropic gave out (trying to woo customers back?). Was able to make up for some of the productivity I lost from the quota reductions over the past 7 weeks. But, it has cost me sleep since it was supposed to expire yesterday.... Also, some frustration, because Claude Code (remote/web) research preview is actually buggier and less functional than Google Jules beta was when it came out. Really don't understand Anthropic. Why would they hand out gift credits if they are low on inference capacity or if they are trying to shed Pro and Max subscribers? Given how many times https://claude.ai/settings/usage has failed to load usage stats in the past couple weeks, I still suspect that they have a broken usage accounting system and are completely oblivious....

Codex has been doing absolutely phenomenal. It's expensive but I was willing to pay for that over a Claude Max plan its that good. And I couldn't hit the weekly limit if I tried. It remembers conversations across various chats in extraordinary detail. Do yourselves a favor and ditch Claude.

Anthropic knows they have an issue. There's no possible way they don't know. Their support channels have been getting absolutely bombarded.

Codex has been doing absolutely phenomenal. It's expensive but I was willing to pay for that over a Claude Max plan its that good. And I couldn't hit the weekly limit if I tried. It remembers conversations across various chats in extraordinary detail. Do yourselves a favor and ditch Claude.

Glad to hear Codex has been working out for you. Last time I used the CLI, it was still missing support for per-project settings and there was no good subagents support. As for the gpt-5-codex model itself, I found it to be slow and not good at instruction-following. Maybe the gpt-5.1-codex model is better, though... have not tried it yet. (Cancelled my ChatGPT subscription several months ago because I was barely using it.)

Anthropic knows they have an issue. There's no possible way they don't know. Their support channels have been getting absolutely bombarded.

Some part of Anthropic definitely knows that there is an issue. But, it does not seem to be the part that can actually do something about it or is willing to do something about it. I've worked in a number of startups and you would be surprised at how dysfunctional they can be. Almost everyone is scrambling to have "an impact" and is avoiding accountability for broken things that would keep them from the scramble. Ignoring people and problems is one way to avoid being pinned with "fix it" duty, sadly.

I've also noticed a drastic change this week. Just passed a weekly reset and already approaching the weekly limit.

It's even more aggravating and obvious that something massively changed because my usage has actually gone down as I'm using claude code a lot more.

Right after I upgraded Claude Code to 2.0.47 tonight, I mysteriously started getting Approaching weekly limit messages in the CLI. This has been a "light" week of usage for me, and I have NEVER been warned before. Went to ask Claude, and you gotta love the irony, as I get shut down right as I've asked the question:

<img width="860" height="776" alt="Image" src="https://github.com/user-attachments/assets/2006650c-4f6a-4ce4-9c09-6d1ea48df038" />

Since when does 76% mean over the limit? Maybe they can get Grok or OpenAI to write them a calculator? This is absolute non-sense, and the fact that we can't even get support is an added insult.

<img width="1248" height="588" alt="Image" src="https://github.com/user-attachments/assets/de62381d-4936-4909-b705-4f18939334e8" />
<img width="536" height="316" alt="Image" src="https://github.com/user-attachments/assets/19bd8e6a-db27-4f7c-a8b9-29e1c0b4baef" />

Let your own model judge you for your actions.
History does not forget.

<img width="1296" height="1294" alt="Image" src="https://github.com/user-attachments/assets/e6d70aaf-7520-47a9-b360-44cf5acd69e2" />

Hi everyone, Thanks for the detailed reports and for your patience. We investigated and confirmed the system is working as intended.

What happened: The usage limits we announced in July were originally planned for August. We made a decision to shift the go-live date to late September (X / Reddit) instead of August alongside the Sonnet 4.5 release, because Sonnet 4.5 was such an efficient model the rate limit experience would be better. However, we know that change coupled with the model upgrade made it hard to understand what was happening with your limits.

Here are some factors that affect actual usage (note that these figures are estimates from our data and vary significantly based on how you use Claude Code):

• Number of parallel subagents
• Number of concurrent sessions
• API latency (lower latency = less time)
• Use of extended thinking or Opus vs Sonnet vs Haiku
• Reading each permission prompt and taking time to type out prompts and give feedback vs auto-accepting all permission prompts

Some usage patterns will hit limits much faster than others, even with similar "wall clock" time.

If you have more questions, happy to clarify but I'm going to close this ticket for now.

"Sonnet 4.5 was such an efficient model the rate limit experience would be better"

You can read through this ticket how much better that experience is, indeed. 😂, i guess it ends here, i finish my project that i started and move on with alternatives.

Btw having to tell Claude to add me a bootstrap class (mb-5) on a div in all the blade templates, because we use bootstrap, just for making fun of me i guess, he serves me : "here i've added this css plain style similar to mb-5" basically re-creating the css of that class in the page as stylesheet, is mind boggling how much more efficient that is.

Thank you for "not an answer", we've got the idea long time ago. hf.

@ThariqS maybe you should go through the entire thread and report to the company you did squeeze the limit a bit too much.

Considering also Claude code 4.5 is not anymore the best and unique tool to code (check benchmark), you are going to lose a lot of users.

@ThariqS : The limits that you announced in July said that I could get 40 to 80 hours of Claude Sonnet 4 usage per week on my Pro plan. How are you calculating that number?

As @gianpaolodn says, this thread shows that you have clearly squeezed the limits far too much relative to whatever efficiency gain that you think Sonnet 4.5 has provided. We are not getting even close to the same productivity/value that we were getting before the end of September.

So, again, how are you calculating the number of hours per week that you advertised for each plan?

@ThariqS : The limits that you announced in July said that I could get 40 to 80 hours of Claude Sonnet 4 usage per week on my Pro plan. How are you calculating that number? As @gianpaolodn says, this thread shows that you have clearly squeezed the limits far too much relative to whatever efficiency gain that you think Sonnet 4.5 has provided. We are not getting even close to the same productivity/value that we were getting before the end of September. So, again, how are you calculating the number of hours per week that you advertised for each plan?

Well put!

I'm canceling my subscription for sure.
My intuition is to naturally stop using claude on a daily basis since i'm reaching my quota so fast that im deterred from "wasting" it.

They become master in the art of patronizing

@ThariqS so this is expected usage? My limits reset last night. I am now at 78% usage, for 72$ of credits? I am on the max 200$ plan, for my last of 6 accounts.

<img width="671" height="334" alt="Image" src="https://github.com/user-attachments/assets/6e5bba0d-17dc-4837-b892-73d58a7bcf59" />

<img width="1125" height="225" alt="Image" src="https://github.com/user-attachments/assets/9f586423-de5c-49f3-8dc3-fffd06e90dfa" />

Are you actually kidding me?

It would only be fair to have been as outspoken as I was to now admit the current usage limits are much more favorable and fair to users. I have resubscribed 1 of my accounts today, and have been having a good usage experience thus far.

I believe it is best to just ban people who abuse your limits, as opposed to going as harsh with restrictions as we had to deal with the last 2 months. I do hope the good faith you have built up with your loyal customers can be repaired from all of this. It sucks with a company you have supported since day 1 to find yourself with 1 foot in and 1 foot out of the door every other month.

If the intention was to deal with customer outlash long enough to retune opus behind the scenes to be more cost effective, a simple post in this thread would have alleviated so much concern. This was not just a dream, there are dozens of screenshots and reports from many users.

I know it will be forgotten about, as we all have hours of work to do every single day, but this was a shame how this was all handled. I do appreciate you retuning things for your paying customers. I know myself and many others have zero issue paying for a service, many times over, as long as the value and respect is there.

Apologies? What an interesting customer client relationship at this point.

## Another Report: Limit reached in 3 hours instead of 5

Subscription: Pro plan
Date: 2025-11-26
Platform: Windows

### Issue Description

For the past 2 months, I worked extensively with Claude Code and never hit the 5-hour limit. Today, I reached
the limit after only 3 hours of work.

### What Changed

• No change in my usage patterns
• Using Claude Opus 4.5 (which should be more efficient according to Anthropic)
• Same type of coding tasks as before

### Expected vs Actual

• Expected: 5 hours of usage before limit (as documented)
• Actual: Limit reached after ~3 hours

This confirms the pattern others are reporting - limits are being consumed significantly faster than before
without any announced changes.

@hermannakruse isn't the new opus just using extra usage? Or goes together with the limits?

Cause if not I did pay for it instead of using my limits.

## Another Report: Limit reached in 3 hours instead of 5 Subscription: Pro plan Date: 2025-11-26 Platform: Windows ### Issue Description For the past 2 months, I worked extensively with Claude Code and never hit the 5-hour limit. Today, I reached the limit after only 3 hours of work. ### What Changed - No change in my usage patterns - Using Claude Opus 4.5 (which should be more efficient according to Anthropic) - Same type of coding tasks as before ### Expected vs Actual - Expected: 5 hours of usage before limit (as documented) - Actual: Limit reached after ~3 hours This confirms the pattern others are reporting - limits are being consumed significantly faster than before without any announced changes.

Okey.... This is new....
I only use Sonnet 4.5 at this moment (Only Claude Code CLI).

All models 52% used
Sonnet only 22%
Who can explain that how can I have 52% while my sonnet use is 22%. So they adjusted the sonnet use and forgot the overal weekly usage? Really wondering how this is done.
Both: Resets Mon 8:59 PM

Even though Claude Code is definitely better than Codex for 90% of the tasks and I like a lot of features, this ever changing token usage is really annoying.
I am on a MAX plan, but website usage is not possible anymore as it burns tokens like hell.
Only Claude Code -for coding and GP 5.1 for brainstorming and ideation.
Where in the past Claude was winner in multiple ways. Really a bummer

It feels this all started to change when they moved their web interface form the OLD artifact system with revisions to I think the Claude Code back-end, from that moment on it went downhill with the tokens...

And customer service at Claude.... well lets not mention that, product is/was great, but customer service.... Non existent.

Just to mention why Claude Code -> It remains a "memory" so on crash you can continue and resume after instead on rereading and instructing again. If any other is able to seriously compete deff wiling to try, but so far codex is not the solution (for me)

<img width="826" height="246" alt="Image" src="https://github.com/user-attachments/assets/12d0cc8e-ecf4-46d1-8b9f-d894d0e0c731" />

Just for info, I think current token usage is equaling more a pro plan than a max plan :-(

All models 52% used Sonnet only 22% Who can explain that how can I have 52% while my sonnet use is 22%. So they adjusted the sonnet use and forgot the overal weekly usage? Really wondering how this is done. Both: Resets Mon 8:59 PM

@tudstudent : Where are you seeing this? https://claude.ai/settings/usage only shows a 5-hour limit and a weekly limit for me. If you're seeing a UI bug on either the website or in Claude Code, then you might want to open a new issue.

Just for info, I think current token usage is equaling more a pro plan than a max plan :-(

Try being on the so-called Pro(fessional) plan... it feels more like the free tier now. Each 5-hour window that reaches 100% usage equates to around 11% to 12% of the total weekly usage. And with Opus 4.5 making the Max plan subscribers a bit happier, the chances of Anthropic doing anything about the Pro plan misery seem small. Feels like they want Pro plan users to "give in" and upgrade to Max. I like Claude, but am not going to pay US$100 / month to get back to the Pro plan usage that I had prior to the end of September, especially not after how (Mis)anthropic has treated their customers.

<img width="1408" height="745" alt="Image" src="https://github.com/user-attachments/assets/8ae42c7e-2ab8-438f-b710-b16baf4849c9" />

@tudstudent : Interesting... must be something that only Max plan users see. Currently, I do not see a separate Sonnet limit:

<img width="2350" height="1424" alt="Image" src="https://github.com/user-attachments/assets/2fe51be4-3961-4aab-af7c-0397c64affcb" />

As far as I know, what you are seeing is something new. Not sure why they would remove the separate Opus limit and replace it with a separate Sonnet limit. This feels backward, given that Opus still has the larger resource footprint. Anthropic must truly believe that Opus 4.5 is so much more efficient than Sonnet 4.5 that they want to push customers to use it instead of Sonnet. That is the only way that I think a Sonnet limit would make sense. They are trying to pressure you to use Opus for efficiency reasons.

Well, just looking back some months ago I could use a tremendous amount more with really stupid tasks (read juggling code over the web interface)
Nowadays everything in Claude Code (yes I admit easier faster, but took time). But now much less interface and much faster done...

Bottomline is that the dynamic change is not transparent and far from ideal.
And being a Max user they advertise with priority to... But customer service is a big nono at Claude.
So I would say AI = 8/10 but customer service is a 1/10.

Hmz, just checked claude code settings and it seems its per default on Opus 4.5.
Can't remember I have put it there, but it could as well explain why its going even faster...
So the message to remove it form limits... Big blabla

@ThariqS : As a reminder, you said here that you would be "happy to clarify" if "[we] have more questions":

If you have more questions, happy to clarify but I'm going to close this ticket for now.

We gave you additional questions but are still waiting for clarification. In particular, I am still waiting for a clarification on how you calculate the expected number of hours of weekly usage for the various plans:

Most Pro users can expect 40-80 hours of Sonnet 4 within their weekly usage limits.

How are you determining these numbers? ^
What Pro users are seeing in practice do not match any reasonable calculation of such numbers.

Sorry for not seeing your request earlier. As I haven’t reached any limits in the last time the whole case is not very important to me just now the only numbers I have for myself as I have been working an awful lot with Claude having fun after update I run into five hours limit, but I guess that is solved as if it didn’t happen again thank you for looking.Gruss H. KruseAm 01.12.2025 um 18:40 schrieb Eric McDonald @.***>:emcd left a comment (anthropics/claude-code#9094)
@ThariqS : As a reminder, you said here that you would be "happy to clarify" if "[we] have more questions":

If you have more questions, happy to clarify but I'm going to close this ticket for now.

We gave you additional questions but are still waiting for clarification. In particular, I am still waiting for a clarification on how you calculate the expected number of hours of weekly usage for the various plans:

Most Pro users can expect 40-80 hours of Sonnet 4 within their weekly usage limits.

How are you determining these numbers? ^
What Pro users are seeing in practice do not match any reasonable calculation of such numbers.

—Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you were mentioned.Message ID: @.***>

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.