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

[emcd]

Preflight Checklist

• I have searched existing issues and this hasn't been reported yet
• This is a single bug report (please file separate reports for different bugs)
• I am using the latest version of Claude Code

What's Wrong?

(Treating this as a bug, since I do not know if it is a misconfiguration with the Anthropic rate limiter or a failure of Anthropic to communicate with its customers.)

Unannounced Reduction in Weekly Rate Limits

Since the release of Claude Sonnet 4.5 on September 29th, I (and many others) have experienced reduced usage limits on our subscriptions. As a Claude Pro subscriber, I was previously able to use Claude Code for about 40 to 50 hours per week, in terms of elapsed time when Claude is actually doing work, accounting for user interaction breaks in the work. This was in line with the email that you sent on 2025-07-28, which stated:

Most Pro users can expect 40-80 hours of Sonnet 4 within their weekly rate limits. This will vary based on factors such as codebase size and user settings like auto-accept mode. Users running multiple Claude Code instances in parallel will hit their limits sooner.

Now, due to a rate limiter misconfiguration or unannounced policy change, I am only able to use Claude Code for about 6 to 8 hours per week in terms of elapsed time when Claude is actually doing work, accounting for user interaction breaks in the work. This has had a severe impact on my established workflow and overall productivity.

Widespread Reports of Problem

• Claude AI usage limit reached|1760000400 #9046

• [Bug] Unexpected Rapid Opus API Usage Limit Depletion #9069

• Claude API Usage Limit Exceeded, Blocking Code Generation #8967

• [BUG] My claude code usage limit is not reset still. #9048

• [Bug] Anthropic API: Rapid Usage Limit Exhaustion with Claude Opus #8644

• [BUG] Limits not resetting #9005

• [Bug] Unexpected Rapid Usage Depletion on Claude Code 20x Plan #8505

• [BUG] claude limit reached in less than one hour on opus 4.1, windows, v2.0.1 #8835

• [BUG] /usage shows 47% weekly used in 3 hours of Opus. Max plan.. should be 40 hours per week #8604

• [BUG] Opus usage limit reached unusually fast with Max 20x plan since Claude Code v2 #8449

• [Bug] Anthropic API: Unexpected Weekly Token Limit Depletion #8918

• [BUG] Is the weekly quota for the Pro version only 20 hours? #8913

• Anthropic API Billing: Persistent Model Access Restriction #8923

• [Bug] Claude Code is not working since Sept 30 | Anthropic API Error: Unexpected Usage Policy Violation Blocking Existing Session #8491

• [Bug] Anthropic API: Unexpected Usage Depletion on MAX20 Plan #8651

• Anthropic API Usage Limit Error Triggered Incorrectly #8594

• [Bug] Anthropic API Usage Limit Blocking Access Unexpectedly #8607

• [Bug] Unexpected Weekly Usage Limit Tracking for Max Plan User #8886

• Anthropic API Usage Limit Not Resetting After Scheduled Reset Time #8675

• [Bug] Usage limit shows 6-day reset instead of 5-hour reset (Claude Max 20x, VS Code extension) #8620

• https://www.reddit.com/r/ClaudeCode/comments/1nvc6i7/usage_limits_we_just_need_replies_not_in_3_weeks/

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

What Should Happen?

Action from Anthropic

• If this is a rate limiter misconfiguration, then acknowledge it and provide proper updates to your paying customers while you fix it.
• If this is an unannounced usage limit change, then communicate it to your paying customers and explain why you could not provide advance notification, similar to what you did in late July of this year.

Error Messages/Logs

Steps to Reproduce

Acquire a Claude Pro subscription. Work with Claude for a day and see how much weekly usage you have left at https://claude.ai/settings/usage .

Claude Model

Sonnet (default)

Is this a regression?

Yes, this worked in a previous version

Last Working Version

1.0.126

Claude Code Version

2.0.9

Platform

Anthropic API

Operating System

Ubuntu/Debian Linux

Terminal/Shell

Windows Terminal

Additional Information

No response

[non-stop-dev]

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":

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:

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).

[non-stop-dev]

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

[tony-pythony]

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..

[sati-bodhi]

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 │
└──────────┴────────────┴──────────┴──────────┴────────────┴──────────┴────────────┴──────────┘

[scottalanturner]

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?

[emcd]

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

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

[non-stop-dev]

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

[emcd]

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:

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

[non-stop-dev]

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.

#9191
#9196

[emcd]

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):

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