[BUG] Opus usage limit reached unusually fast with Max 20x plan since Claude Code v2

Found 3 possible duplicate issues:

1. https://github.com/anthropics/claude-code/issues/6457
2. https://github.com/anthropics/claude-code/issues/6488
3. https://github.com/anthropics/claude-code/issues/7592

This issue will be automatically closed as a duplicate in 3 days.

• If your issue is a duplicate, please close it and 👍 the existing issue instead
• To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

It could be realted to opusplan removed from default model settings ?

I found a workaround by manually setting Opus in plan mode in the settings, which seem to resolved the issue.(opusplan)

Well.. i guess i have to contact support now...

This may happen when compressing the conversation.

Yeah I was like hitting my weekly opus usage in the first day without hitting any single 5-hours block opus limit, which is absolutely disappointing. I won't subscribe to max 20x anymore.

I've been using this plan for about 3 months and this is the first time I've seen this limitation. (Since this plan has existed)

I've also found that even with model set to Default(Sonnet) the Opus usage continue to grow according to /usage command
#8525

<img width="503" height="106" alt="Image" src="https://github.com/user-attachments/assets/95e89c39-ce22-4851-9931-72db113ca983" />

However, Claude Desktop shows I've reached 100% of my weekly Opus limit.

Is there anyone from Anthropic who can respond to these questions? I'm currently completely unable to use Opus, which makes the Max 20x plan unusable for me.

https://github.com/anthropics/claude-code/issues/8541 Same here

This is literally disgusting! Especially for those paying $200/month on the 20x Max plan. There’s already a Reddit post blowing up about it, everyone should check it out and drop a comment: https://www.reddit.com/r/ClaudeCode/comments/1nufbi3
Anthropic wake up !!!!!!!

20x Max user here, your new limit, bug or intentional, is not acceptable and this is up right cheating. If you don’t fix your moral or your limit. I’ll put my $200 into OpenAI’s pocket

I already cancel 20x Max because use faster

@bcherny @ant-kurt @ashwin-ant @rboyce-ant @bogini

This looks like a serious bug guys...

Can someone from the Anthropic team please respond? Multiple Max 20x users
($200/month) are now completely blocked:

• Weekly Opus limit consumed in 1-2 days instead of 7 days
• I've now reached 100% and can't use Opus at all
• OpusPlan option removed from v2 settings (we can't control model usage anymore)
• Even with model set to Sonnet, Opus usage continues to grow (see #8525)

We need official communication.

i am having the exact same issue - it must be a bug - it never happened to me before today!

how can we get Anthropic to answer us?!
@claude - let your peeps know about this ASAP

I'm like 50 percent sure its because of the over usage of the thinking mode I've even had system prompt leaking in my replies its thinking more which is using a lot of context.

I just blew up today, this hasn't happened in 3 months. Really annoyed and no way to buy more credits even if I wanted to.

I also am a 20x user and I can't even get a full day of opus usage. Their own docs say we should be getting "AT LEAST 900 messages per session", I am getting maybe 30 of Opus and then hitting my weekly limit. Anyone else getting anything close to 900 messages?

I've been on the max $200 plan for a bunch of months now and this is the first time I've ever seen this. Today. Clearly an issue.

same here. i've only used opus in the past few months and it's my first time seeing this:
Approaching Opus weekly limit

you think there is any chance that Anthropic will address this - or they will just ignore it until it goes away?!
@claude - did you notify your supervisor?!

Well, when using one of the versions that definitely didn't have such an "issue", I see a message about the weekly limit, which I've never seen before, so, did Anthropic decide to tighten limits again, and where can the new limits be found if yes?

<img width="867" height="213" alt="Image" src="https://github.com/user-attachments/assets/df947767-e723-445d-9f1e-5f5a2edcbbb4" />

Pleasant surprise: My usage has been reset!

Current status:

• All models: 1% used (reset Wed 22:59)
• Opus only: 0% used
• Weekly limit appears to be back to normal

Not sure if this is:

• A manual compensation for the bug
• An automatic fix with recalculated usage
• Unusual reset timing (was showing "Monday next week 7pm" before)

Note: The /usage slash command in Claude Code CLI is still bugged and doesn't display usage stats. Had to check via web dashboard.

On a positive note: Using Sonnet 4.5 has proven to be effective and pleasant for now at least.

<img alt="Image" width="867" height="213" src="https://private-user-images.githubusercontent.com/8779534/496492520-df947767-e723-445d-9f1e-5f5a2edcbbb4.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.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.FuatrAEso7HKv4pO_XCViEmRR36VaenzrO7ij8Bllp4"> Pleasant surprise: My usage has been reset! Current status: All models: 1% used (reset Wed 22:59) Opus only: 0% used Weekly limit appears to be back to normal Not sure if this is: A manual compensation for the bug An automatic fix with recalculated usage Unusual reset timing (was showing "Monday next week 7pm" before) Note: The /usage slash command in Claude Code CLI is still bugged and doesn't display usage stats. Had to check via web dashboard. On a positive note: Using Sonnet 4.5 has proven to be effective and pleasant for now at least.

They reset the limit but it's useless, this is not a solution. The limit is STILL THE SAME, unbelievably low for a $200 subscription.

Paying $200 and only be able to use the top model for 1-2 days per week? I'll pay my money to OpenAI if Anthropic keeps doing this.

Yep something seems weird here. "Approaching Opus weekly limit" just appeared on my client, I've never hit that limit before (I'm on 20x), and my usage this week is not at all "abnormally high". Seems like a bug with the new version of the claude-code client. Aand yep a couple hours later "Opus weekly limit reached ∙ resets Oct 9, 3am" -- I didn't even hit the 5 hour warning in the past 2 sessions but apparently I consumed an entire week's quota? with a single project i've only been occasionally poking at today?

Same here. Reached my limit in 2 hours. Max pro 200 user

Evidentemente esto es un error de Claude, lo que evidencia que a los desarrolladores de claude tambien se le acabo el plan semanal jejej, tuvieron que poner su mente, despues de 6 meses en desuso, a pensar, y metieron la pata. Tengo esperanza que arreglen esto @claude . Como empresa puntera en el servicio de IA deben urgentemente quitar el limite semanal, de no hacerlo, van a perder mas clientes que nunca

Obviously, this is a mistake by Claude, which shows that the developers at Claude also ran out of their weekly plan, hehe. After six months of not using their brains, they had to think, and they messed it up. I hope they fix this, @claude. As a leading AI service company, they urgently need to remove the weekly limit. If they don’t, they’re going to lose more customers than ever.

resetting the limit doesn't solve the excessive usage problem. I'm on a pro plan, doing refactor on 300 lines of code and already used 20% in 5-hour limit. it might imply that I'll only be able to do similar stuff 5 times in 5-hour

<img width="1099" height="564" alt="Image" src="https://github.com/user-attachments/assets/6e76b7be-3d41-4153-b3b1-ba5e795ddadd" />

4 hours normal usage with pauses and limit jumped from 20% to 92%

● I'll do ... ⎿  Opus weekly limit reached ∙ resets Oct 9, 12am /upgrade to increase your usage limit.

/upgrade ⎿  You are already on the highest Max subscription plan. For additional usage, run /login to switch to an API usage-billed account.

@claude, we definitely need a fix; it's not ok to pay $200 for just two evenings of use.

Guyyys, 6 hours of work and all week Opus limit is reached, @claude please pay attention to it, I barely used it in these 6 hours, 1 prompt and 5% of the limit is used, 20 bash script launches and limit is reached

I've never hit this before in the 3 months I've been subscribed to max. Also, this weeks usage is much lower than some other weeks.

Some others are reporting the same.

Definitely looks like this is a recent issue.

well fellas, i can tell you that we are cooked! there is no way anthropic is gonna address this - basically the max plan is a useless POS that gives you 4 hours of opus - at this stage its 100% not worth it...

Based off watching the meter now on Claude Desktop? I mean not even using much Opus I hit 30% in a day for the week which is crazy on Sonnet 4.5. I don't even want to try and use Opus knowing how fast that will consume.

Yes i'm on windows here btw, FWIW this seemed to change for myself with Anthropic's last "fix" -- when they said they globally reset everyones usage to fix issues, it was like 3 hours later that my $200 plan went from 0% used to 100% expended. Quietly watching this mess unfold as Anthropic sets fire to (their only?) predictable revenue stream while grok and co desperately try to poach customers with pricing at 1/10th Anthropics. aaand nothings happened about this issue with hundreds of people complaining after a week? This still feels like it could be some idiotic marketing play, to force people to at least give sonnet 4.5 a go (i'm sure this issue occuring directly after its release was PURE coincidence...)

Yep, i've tried it. Probably shouldn't have it's performance was significantly worse than Opus 4.1, which was already significantly worse than Opus 4.0. Was nice knowin ya Claude.

Sharing for visibility: I documented testing and concerns that Opus 4.1 limits may have been quietly tightened after Sonnet 4.5. Details here: https://www.linkedin.com/posts/activity-7380259553172406272--YWi

If you are seeing the same, please add your plan and usage details here and on LinkedIn, upvote this issue, and share the post to increase visibility and pressure for an official response.

i am on max plan too. Was getting hit by 5 hours limits over last several week, now this weekly limit, ironic it resets on Wednesday for me..Totally not happy, possible cancelation coming soon. Paying 200$ and not being able to use it for more then few hours a week is ridiculous. I am not even a power user.

<img width="1210" height="589" alt="Image" src="https://github.com/user-attachments/assets/8279d914-7716-4ea7-aee5-07be801375d1" />
Only lasts 3 days for the weekly limit.

Sharing for visibility: I documented testing and concerns that Opus 4.1 limits may have been quietly tightened after Sonnet 4.5. Details here: https://www.linkedin.com/posts/activity-7380259553172406272--YWi If you are seeing the same, please add your plan and usage details here and on LinkedIn, upvote this issue, and share the post to increase visibility and pressure for an official response.

They definitely tightened it. I used to use opus 4.1 all day every day and never reach weekly limit.

<img width="694" height="345" alt="Image" src="https://github.com/user-attachments/assets/948df810-50c0-4101-ac44-ce8a5dbdbca2" />

I've been using the $200 Max plan now for a few months and in that time have only ever once come close to using the full five-hour limit and used it pretty much the same every day. But since this Thursday to now Saturday, I have not had a single warning of reaching the five-hour limit, but as of Saturday morning, I started getting the warning that I was reaching the weekly limit. When I finally worked out how to look at the stats for the usage, I was utterly shocked because it said I was on almost 90% of my usage. After less than two days of the same kind of work that I was doing all month prior to the version 2 model.
It would appear that Opus is now charging the full API rate to people with a subscription as well. This is unsustainable for me. I simply couldn't afford to pay multiples of the $200 to do the same work I was doing the previous month, all month. There is either something wrong or they're trying to scrape back money by penalizing the people that use it the most. This certainly does not fit with the description on the website that states that only a few people will hit the weekly limit. And that most people will be okay. I'm not a power user, but it would appear that I'm being treated the same as one.
Even switching to Sonnet now is chewing through my quota faster than it used to do with Opus. This is Unsustainable.
Looking at all the complaints, I think Anthropic's going to have a serious issue if they don't resolve this for the genuine users.
I'm now reading that there's been a big move from Claude code to GPT-5 coding, and that the upgrade to Sonnet 4.5 was hoped to bring some of them back. I think that's where I need to look now.
I get that there are some that just abuse the hell out of it and get what they can and wreck things for others, but I'm certainly not one of them and I suspect most of the people that are commenting here aren't either.
I Prefer using Claude code, but it's unaffordable now, so I'm going to now looking seriously into alternatives.

<img width="1009" height="536" alt="Image" src="https://github.com/user-attachments/assets/3e04a21f-41b5-4716-807c-fcab95db628a" />

I believe this is a bug rather than a deliberate policy change. The global usage reset two days ago doesn't align with a scenario where Anthropic would intentionally tighten the limits - why reset everyone to 0% if the goal was to reduce usage? even if it doesn't solve the "bug"

That said, the lack of communication is frustrating and quite unprofessional. Radio silence on a critical issue affecting $200/month subscribers is not acceptable.

I'm assuming the team is working on it, though. On my side, I have an open support ticket (and I imagine, I'm probably not the only one), so hopefully we'll get clarity soon.

In the meantime, let's keep documenting our usage patterns to help with debugging.

+1
i'm a max20 user, , worked for one night, and then couple hours today, suddenly approaching weekly opus limits?!??

i've used it for months wayyyy more, this is the first week this happened to me

<img width="900" height="472" alt="Image" src="https://github.com/user-attachments/assets/150dd197-dade-4264-b19b-ef43642f9573" />

Same here. It seems like I got 10% of usage compared to the usage of the beginning of last week. If this is not a bug a lot of paying users are going to the competition (me included). That's for sure.

<img width="999" height="495" alt="Image" src="https://github.com/user-attachments/assets/8a5b6608-7b7c-4b36-b7e5-7670303816a2" />

20x Max Plan here since June. Since switching from 10x Max to the 20x plan, I have NEVER seen any usage limit warnings. Since updating to 2.0.X, I'm now hitting the same problem.

• No - I'm not using this full time. I use CC for a side-project that I work on sometimes after finishing my day job.
• No - I'm not account sharing.
• Yes - I only have one instance of Claude Code running at a time.

If this is a bug:

• Anthropic, you need someone to come in and at least let people know that you're looking into it. These are paying customers on your highest plan.

If this is intentional:

• All this is going to do is push people to tryout Codex/ChatGPT Pro, or Cursor Ultra. Both are $200/month and don't have these horrible weekly limits at all.
• If you restrict people's ability to do work, they'll simply take their money elsewhere to a tool that will let them work without restrictions. And once they're gone, marketing 101 says that they'll be much harder to gain back.

Guys, they're intentional, and to someone who mentioned that they reset the usage, you can see the post here.

https://www.reddit.com/r/ClaudeAI/comments/1nvnafs/update_on_usage_limits/

TLDR: We reset the limit so you poor users stop making noises, accept that we can do whatever we want to lower your limit, nerf our model secretly, and claim it's a bug. Now use our cheap model even if you pay the highest tier of subscription.

Guys, they're intentional, and to someone who mentioned that they reset the usage, you can see the post here. https://www.reddit.com/r/ClaudeAI/comments/1nvnafs/update_on_usage_limits/ TLDR: We reset the limit so you poor users stop making noises, accept that we can do whatever we want to lower your limit, nerf our model secretly, and claim it's a bug. Now use our cheap model even if you pay the highest tier of subscription.

Yeah that's the worst part. I had no issues up until then. I rarely hit any kind of usage warning let alone limit, the fact I pay for Max20 is more of an embarrassment to admit than anything, I was having a light to regular usage day on October 3rd (2nd/3rd in US time i'm in .au), and without getting any 5 hour limit warning, daily limit warning, opus model warning, etc I got this new "Approaching weekly limit" message appear. 2 hours later of what was literally like 10 messages because I was already concerned, that changed to "Weekly limit reached - reset Oct 9th, 1pm". So now i'm here unable to use Opus for work as the account is intended for, without any warning this was coming to change my patterns, without any announcement from Anthropic about any kind of radical limit change, or any acknowledgement of the bug / support issues, and i'm still locked out for another 2 days. And then I find out some people who were more aggressive with limits already hit them and got a reset, but I get "punished" for not being a heavy enough user to notice quickly enough.

To me this seems almost concretely "This costs too much. Let's just apply new limits silently, most people will be confused instead of just leaving and we can spread out the negativity". The lack of transparency on usage limits was a concern about Anthropic's subscription system from the start. This issue has already pushed me to get myself setup with other providers, and if this is Anthropic's standard practice to readily (ab)use such "at our discretion" nonsense, I cannot in good faith recommend them for any commercial solution of any kind.

Wow... thanks @Meowzz95 for the link. I'm not sure I understand everything, but apparently Anthropic is not answering the most upvoted question (119+ upvotes) asking for confirmation about the documented quotas (25-40 hours of Opus, 240-480 hours of Sonnet).

Sonnet 4.5 seems to consume normally on my end, but all of this really lacks clear communication. Customers deserve transparent information about what they're getting, especially when limits change significantly.

The communication is honestly catastrophic.

I’ve canceled my Max ×20 plan and switched to ChatGPT Pro, moving my workflows to Codex. Anthropic’s handling of usage limits, and the way they quietly quantized their model, only acknowledging it after Sam’s post on X, are to me, serious red flags that don’t justify paying them.

Ya'll if this is all true, I'm canceling the subscription this month

In my opinion, Sonnet 4.5 is quite "stupid" and does not follow instructions; no output styles help, it just breaks everything... Сlaiming that this is the best model for coding is simply an assertion out of nowhere.

+1, sonnet 4.5 has done more damage than sonnet 4.1 ever did.

It's factually useless for what I do (computer graphics, Rust).

Just for example, I was working on some complex topological operators for 3D mesh processing. Opus was a great help for all but the most difficult-to-reason-about tasks.
Sonnet 4,5, with one change, destroyed one week of Opus work across 5k lines of code (sprinkled everywhere), ignoring detailed instructions about invariants laid out in a markdown file and code examples.

Opus 4.1 is what has worked for me fine in the last three months (5h block-limit ofc, max plan) but now I am seeing the Approaching Opus weekly limit message too.

Wtf, pardon my French? Seriously, Anthropic?

P.S.: I posted this on HN: https://news.ycombinator.com/item?id=45489316, upvote, maybe that will get us Anthropic's attention.

Sonnet4.5 is an improvement for agentic workflows, but it still forgets basic context due to auto-compacting after hitting the context limits. Moving to Codex will not solve the problem as they have the same issues. All of them are stalling time untill they get the AI-Data centers up and running. Codex does the same thing, first you get a taste of how it should work, then it slowly degrades.

Also, OpenAI retains ALL logs. No opting-out. ID veri for gpt-o3/o1 etc. Plus GPT5 is not a model, it is a family name for the GPT-4 family as shown on the system card, using 'smart-routing' for prompts. So all model providers are pulling the same stunts untill they winged it to get the Collosal data centers up users will be left with moving from one to the other compounding high expectations upon dissapoints for each new provider switch.

Nr1 source of all of this is energy.

Sonnet4.5 is an improvement for agentic workflows, but it still forgets basic context due to auto-compacting after hitting the context limits. Moving to Codex will not solve the problem as they have the same issues. All of them are stalling time untill they get the AI-Data centers up and running. Codex does the same thing, first you get a taste of how it should work, then it slowly degrades. Also, OpenAI retains ALL logs. No opting-out. ID veri for gpt-o3/o1 etc. Plus GPT5 is not a model, it is a family name for the GPT-4 family as shown on the system card, using 'smart-routing' for prompts. So all model providers are pulling the same stunts untill they winged it to get the Collosal data centers up users will be left with moving from one to the other compounding high expectations upon dissapoints for each new provider switch. Nr1 source of all of this is energy.

Is this comment written by poor sonnet itself? @mrXInbetween

There are multiple obvious hallucinations:

but it still forgets basic context due to auto-compacting after hitting the context limits.

nobody is talking about context issue, this thread has always been about usage limits.

Plus GPT5 is not a model, it is a family name for the GPT-4 family as shown on the system card,

Codex specifically has a model picker to pick which model to use

using 'smart-routing' for prompts

This is ChatGPT Chat interface's "Auto" option, and even there, you have an option to choose instant, thinking, thinking mini, pro.

So all model providers are pulling the same stunts untill they winged it to get the Collosal data centers up users

This is even more confusing, do you mean they have no data centers? I guess Sora 2 alone uses a huge data center 😂

Are you dumb?

Context issues are usage limits.

Did you read the GPT system card or did you just think "hmm lets show off my ignorance and stupidity and not do my own research before shooting off"

Context issues are usage limits

These are definitely not the same; people here faced the inability to use Opus, not because of its context window, but because of possible new restrictions on the provider side

Context issues are usage limits.

mrXInbetween, I do not think you understand. This is thread _is not about context_.

My context does not compress when I use Sonnet 4.5 any more often than it does when I use Opus 4.1. And when it compresses (or rather: before, to not waste tokens) I start a new session. Because as we all learned, anything else is shooting yourself in the food, eventually.

Sonnet 4.5 is simply a vastly inferior model.

I have entire sessions to prove this from the last four days. And four months worth of code written by Opus to prove the counter point.

yesterday i was around 80 something percent opus, then went to sleep
today used a bit of sonnet 4.5 "to maintain what's left for really challenging programming parts"
i was browsing now and the tab was already open, i was shocked

<img width="900" height="472" alt="Image" src="https://github.com/user-attachments/assets/0beea983-375e-42f1-b435-29504b1d87ea" />

I'm out of usage somehow, all it took 2 programming sessions, I have uni and other things. I'm not programming hardcore and it's like this (Max 20), my subscription ends on 14th. And if there is no update regarding this by then. I'm definitely not renewing!

<img width="1220" height="733" alt="Image" src="https://github.com/user-attachments/assets/eb197ef6-1ae4-4d06-aea7-da4c8b1e9d15" />
I lost access to Opus after 24 hours of buying the $200 max plan !

There are alternatives:

1. Other CLIs: see an article: https://www.linkedin.com/pulse/2025-landscape-cli-coding-assistants-aleksandr-filippov-omy2c.
2. Claude Code CLI with another model: use https://github.com/1rgs/claude-code-proxy to keep the Claude Code UX while switching the backend to, for example, Gemini.

Just switched to Codex. Opus was amazing... RIP!

Just canceled Opus and requested a refund. Last week I reached my limit for an entire week I thought that was ridiculous. Today they charged me again my monthly without being able to use it for an entire week out of the month. I opened a second account to be able to use it while the other account reset and $400 later (my original account and my new one) the new one suddenly says opus limit reached after a 3 hour coding session yesterday with MAXIMUM 6-8 messages barely anything completed; just BS back and forth with Opus.

Very very unfortunate but RIP claude code. Im out. Highly unreasonable restrictions and limits on Opus. Completely kills the entire function of Claude Code. 4.5 is great but we pay for MAX at $200 a month to be able to use Opus within reasonable limits. Im not coding a project here for the pentagon neither running 20 hour days on Claude Code Ultrathink Opus. Very sad.

This is too bad. I only used it for less than 2 hours.
<img width="1800" height="942" alt="Image" src="https://github.com/user-attachments/assets/da54ef53-3ce4-43b3-b2f7-caf97b3d6f61" />

<img width="215" height="169" alt="Image" src="https://github.com/user-attachments/assets/2fbaa9c5-7344-4ecd-be51-cbc35f2804d6" />
somehow these are now earlier by several hours
also why the limit on the opus rest afterwards by a lot?

Anthropic guys, you want us to switch to Sonnet 4.5 instead of Opus 4.1, but sorry I do not achieve same results.
The max account is now unusable for me.
Cancelled my Max Account and trying Warp.
Not bad at all ! well surprised
Good UX ! much more visibility of what was done, more inline with my workflow
Even in Yolo mode when you let it work during a long time without control

I’m having the same problem, but in my case I’ve never even used Opus — I only run Sonnet 4.5. Still, I keep getting the “Approaching weekly limit” warning after very minimal usage.

This clearly shows something is broken in the way usage is being counted since v2. It makes the Max plan feel pointless — I’m paying for capacity that I never actually consume. Unless this is fixed soon, I’ll most likely unsubscribe because the value just isn’t there.

For now, I’ve started trying out Warp, and honestly the results are impressive. The only real drawback is that it doesn’t yet support web search, but for terminal-based agentic coding it really feels like a game changer.

It's time to RIP Claude!!

same here

I have disabled playwright-mcp, will not offload to gemini/DeepSeek etc, not use claude agents..I will be resetting today, its been 4 days since darkness.

If these measures fails, I must move on. I will other terminals with Gemini or anything else for trivial tasks, let's see how Claude consumption gets by end of day today.

From what I see this is only being hit by subscription users, users are not being hit with the daily limit first, and it could potentially be the result of harder thinking modes being discovered and used heavily but I'm not too sure that is the case.

We strongly recommend Sonnet 4.5 for Claude Code -- this is the model everyone on the Claude Code team chooses (just polled the team earlier). We are optimizing for giving people as much Sonnet 4.5 as possible, since we think it's the strongest coding model. Give it a shot. If you want more Opus than what the Max plan includes, we recommend using an API Key.

We want you to have the choice, but in practice, we have to make many hard tradeoffs around what model we give the most of. In this case it's definitely 4.5. This might change again in the future, eg. if there's a new Opus model that's better than 4.5.

We strongly recommend Sonnet 4.5 for Claude Code -- this is the model everyone on the Claude Code team chooses (just polled the team earlier). We are optimizing for giving people as much Sonnet 4.5 as possible, since we think it's the strongest coding model. Give it a shot. If you want more Opus than what the Max plan includes, we recommend using an API Key. We want you to have the choice, but in practice, we have to make many hard tradeoffs around what model we give the most of. In this case it's definitely 4.5. This might change again in the future, eg. if there's a new Opus model that's better than 4.5.

This makes sense and am having a decent time so far with Sonnet 4.5. Was there a change to the usage limits for Opus after the August change? Am here mostly because of how much faster I hit a weekly limit without even seeing a daily limit warning appear.

We paid 200$ to get what's agreed on when we paid (x usage of opus/sonnet), regardless of what's better. If you want to change something you don't do it just after we pay,that's actually the definition of scamming
you inform us before next cycle and we make the decision whether we want to subscribe to that NEW plan or not.
Idc what the team uses or if the new model is better in a certain aspect you see. I paid to use opus according to what you proposed. Point.
Maybe if you tried to convince me i might've accepted, and that's different.
But putting me in a position where I'm forced to accept this. NO!

That's not to mention that the actual weekly quota for the whole subscription has been secretly reduced a lot. Which kind of defies parts of what you said...

I'll take what you said as a confirmation that it has been done intentionally, which i wished it was a bug...
Ig it's time to search for an alternative..

It's not always about the decision, sometimes about how you decide to apply that decision

It's also about honesty and transparency...

Hey guys, I asked Claude code(in terminal) to cleanup its acts, it did go through deleting many old previous conversations, set a cronjob to keep only last 30, cleaned up .claude/projects and config(json) file, shrank my global Claude.md file, i have been using it nonstop for last 4 hours.. usage barely moved!!! Secondly i did remove Microsoft Playwright MCP server too..something was not correct with that too!

What are you suggesting, that using a fresh Claude Code, no MCP’s and a
very minimal Claude.MD file is the solution?

On Wed, Oct 8, 2025 at 9:03 PM Sams3595 @.***> wrote:

Sams3595 left a comment (anthropics/claude-code#8449) <https://github.com/anthropics/claude-code/issues/8449#issuecomment-3383694438> Hey guys, I asked Claude code(in terminal) to cleanup its acts, it did go through deleting many old previous conversations, set a cronjob to keep only last 30, cleaned up .claude/projects and config(json) file, shrank my global Claude.md file, i have been using it nonstop for last 4 hours.. usage barely moved!!! Secondly i did remove Microsoft Playwright MCP server too..something was not correct with that too! — Reply to this email directly, view it on GitHub <https://github.com/anthropics/claude-code/issues/8449#issuecomment-3383694438>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/A3XRAH6TIXGJLEGIM6SEMMD3WWX7XAVCNFSM6AAAAACH3ZAJNWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTGOBTGY4TINBTHA> . You are receiving this because you commented.Message ID: @.***>

You say that like you're not giving us a lower limit on Sonnet 4.5, than we previously had for Opus 4.1 for the same money?

Ok, thanks for the confirmation, organizing the cancellation of 46 subscriptions. This is... Nothing short of a trap.

As for your "team", perhaps this might help them understand why their product has fallen apart.

Sonnet 4.5 is significantly behind Opus 4.1, which was already a good deal worse than Opus 4.0. They've been steadily trending towards quickly punching out basic apps "at a higher quality" as long as they're early in the context (aka poorly defined/vague/open ended), while the model has essentially been getting more and more ... anxious? Opus 4.1 has performance anxiety, Sonnet 4.5 has straight up stage fright. It's under so much "pressure" right at the start of a context, it will try to smash out something amazing and sometimes succeed, but once you actually start diving into debugging it very quickly falls to pieces and starts hallucinating badly.

Opus 4.0 used to be able to confidently handle a blend of 10 tasks all together in the one prompt, there was no point in unravelling them it could infer more about the structure by how I wove them together. Sonnet 4.5 can handle one, maybe two before it loses track of itself. It's still "usable", and if you're very liberal with new contexts and the Task took you can get some decent results, but it's literally like working with a child with learning difficulties instead of a genius collaborator. it's a completely different beast from Opus, it needs constant hand-holding and redirection, it simply isn't capable of the kind of in-depth analysis or even just being able to "think through things" on a basic level.

Eg when planning, I've come to expect Sonnet 4.5 to throw a completely insane nonsense suggestion out on the 2nd or 3rd turn, if it can't one-shot a problem there's a very high chance it will take a very strange path towards a non-fix and will require redirection. My favourite example this week was that I complained about power usage from my server being more than my solar rig generated, Sonnet 4.5 confidently responded that I should simply run my solar panels at night for better utilization. Simple.

This is not an isolated example this kind of unforced error will come essentially EVERY context that drags on beyond 50k tokens. I used Opus not because I need a word salad generator, but because I was actively leveraging its depth of thought to build highly technical code. This is no longer a product Anthropic apparently offers, even at API rates (quants only, even Opus 4.0 at $75/M tokens rates is a mere shadow of what it used to be)

Sonnet 4.5 is a step up for Sonnet, but clearly a completely different class of model from the old, pre-quantized Opus 4.0 I started with and paid for, which I only got about 2 weeks of before they crunched it down (again without admission from Anthropic until they unofficially let it slip later, like all of these my only true awareness of any issue was significant in-app performance degradations.) I was actually working on a project specifically to push the limits of Opus 4.0 to see what it could do, and one day it just started making errors and could no longer handle operating at the same level, I had to scrap the whole project. A week or two later I find out the model got quantized down and that explained everything.

The thing I get the least, is where does this leave Pro & Max5 users? Like, with the way the limits are now applying, I do not believe Max5 actually has "5 times" the limit of Pro, nor that Max20 actually has "20 times" the limit of Pro aka 4 times the limit of Max5. All the figures I see now are a) time based figures (how do these make ANY sense? its all about tokens not how long i ponder between each prompt....) and b) all seem to show about twice the usage time for Max20 vs Max5. Anthropic have been careful to never clearly define exactly what usage means so they can change the rules whenever they like, but I paid for a plan which has exactly 20.00 times the usage quota of Pro, and I do not believe that to be the product that was provided, even WHILE they're lobotomizing the product and delivering inferior value at the same time.

My experience with Anthropic/Claude started on a high, and has been in constant decline ever since. I've come to understand Anthropic as one of the most consumer-hostile companies of our time, and i'll be doing all I can to avoid them and warn others away so they don't have to go through the same painful learnings. You only need to burn people ONCE with a bait & switch, and they'll never trust you again. From my perspective this plan has dropped in value by 48 times, a halving in value with the initial quantization, a halving in value switching to opus 4.1/sonnet 4.5, and a 12x reduction in value with these usage limit changes. There is no WAY I could justify US$200 for the current offering, and i'm still rather stunned that Anthropic just don't have a product anymore for such users, not even a US$2000/mo subscription. But that's fine, there's plenty of other providers.

I've been exhausting the rest of this months usage with Sonnet 4.5 switching my everything over to third party or custom tooling. As long as I keep tasks short, clear, and easy to accomplish Sonnet is mostly capable -- but then a 34B coding LLM running in 4 bit on my local GPU also achieves that too... I'm over it, building a custom rig to run GLM 4.6 at home, backed up by OpenAI Codex with fallback to Grok Code. No more extortionate subscription fees to an intellectual dealer/pimp who keeps altering the deal to extract anything they can from their customers "now that they're addicted". I'd rather pay 10x the money for sovereignty, especially if we're going to start comparing models on benchmarks, quality is VERY comparable to Sonnet 4.5 and I have only room to improve instead of Anthropic's constant downgrades.

You are not optimizing? You're up straight cheating your customer. Ask your $200 subscription user to use a cheap model and advertise it is the best model in the world is your solution? We pay you money and we should get what we paid for, i.e. enough usage on Opus without abusing the service. And now just because you think blah blah blah and you take away what we paid for? I believe this is even a breach of law in certain states.

@bcherny Better late than never.

The product is better, I agree. But I think most of us subscribed to the Max 20x plan specifically to get the most out of Opus 4.1 and bypass the throughput limits constantly being reached during coding sessions. Is this still an issue with Sonnet 4.5? The 5x plan might be enough for most users now. Although, on the other hand... it's a good thing that this limit is lower with Sonnet 4.5.

From my perspective, the problem isn't with the product, but with the communication.

This thread illustrates the problem well. I honestly thought it was a bug.

You say this could change again in the future; please inform users of these changes in advance.
Users should be able to choose the plan that best suits their needs, based on the best-performing model and its actual limitations.

Personally, I'm keeping my current plan because I need it. But I understand that some people feel cheated. This lack of transparency is aggressive, even if the product is good. Since this isn't a bug, I understand the need to close it.

However, please respect your users, be clear and communicate changes that affect paid plans.

Opus usage limits with the Max plan are in line with what's in the Help Center article: https://support.claude.com/en/articles/11145838-using-claude-code-with-your-pro-or-max-plan.

There was a bug earlier where we said in the UI that you hit your Opus limit but it was actually a weekly limit, this is now fixed. It's unrelated to rate limits and was a UI bug.

We highly recommend Sonnet 4.5 -- Opus uses rate limits faster, and is not as capable for coding tasks. Our goal with Claude Code is to give everyone as much as possible of the best experience by default, and currently Sonnet 4.5 is the best experience, based on SWE Bench, user feedback, and team vibes.

Please let us know if you're not getting Opus usage in line with the Help Center article.