[Bug] Anthropic API: Unexpected Weekly Token Limit Depletion

Found 3 possible duplicate issues:

1. https://github.com/anthropics/claude-code/issues/8886
2. https://github.com/anthropics/claude-code/issues/8505
3. https://github.com/anthropics/claude-code/issues/8651

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

The irony is that I stopped using Claude Code and switched to Codex, just used Claude earlier today for basic things like 'run these unit tests and make sure they all pass' and that's it. Repeatedly ran these on and off all day - claude code was made to do no real development work.

Shocking to see I'm around 80% through a weekly limit within a few hours of zero actual work.

Weekly limits went wild. It is absolutely irrational. Even the pricing structure doesn’t match with usage limits. 20x plan users are facing same issue… so what’s the solution? Create 4 accounts and pay separately 200$ x 4 times to be able to use Claude Code across the entire week? Insane.

This looks like the ration books for food they gave out in the Soviet Union and Cuba 🤌🏻.

“Sr you have to pay this bunch of money to be able to eat breadcrumbs”… that’s how this usage limit sounds

Max X20 plan — weekly limit reached in 2–3 days, Sonnet 4.5 only

I’m a paying Max X20 user (≈220 €/month) and I’ve just hit my weekly usage limit after only 3 workdays.
Now I have to wait 4 full days before I can keep working.

Important details about my usage:

• I do not open multiple terminals - I work only in one terminal on a single backend project.
• I do not use Opus at all -only Sonnet 4.5 with Claude Code for code review and help writing complex backend functions.
• I’m a normal developer with a standard 8-hour workday. No 24/7 heavy load.

When I subscribed to Max X20, the advertised limits were:

• ~900 messages per 5 hours
• 240–480 hours of Sonnet 4 weekly
• 24–40 hours of Opus 4 weekly

Now the reality is nothing close. I hit my weekly cap in 2–3 days just using Sonnet 4.5, no Opus, no MCP.
Anthropic keeps quietly reducing limits almost every month and adding new hidden restrictions - this tool is becoming unusable for normal dev work.

**Anthropic said these changes would affect “less than 2 % of the most active users”, but looking at thousands of complaints on Reddit, Discord and Twitter, it’s clearly hitting almost everyone.

If this isn’t fixed soon, I’ll have to cancel my subscription - paying $200+ for 2–3 days of work is unacceptable.**

After reading some reddit posts, I found this was not a bug but intentional limitation to us, so I cancelled the subscription yesterday. (Max plan)

I still love chatting with CC rather than Codex cli, I am really sorry we have to leave.

I still hope harsh limitation was a plain BUG by hallucinations.

+1, I was surprised to see my Weekly Limits % being depleted overnight, when I haven't used it at all.

I'm purely using Sonnet 4.5, and running claude update every time I open the terminal - yet it shows Opus usage too.

After reading some reddit posts, I found this was not a bug but intentional limitation to us, so I cancelled the subscription yesterday. (Max plan) I still love chatting with CC rather than Codex cli, I am really sorry we have to leave. I still hope harsh limitation was a plain BUG by hallucinations.

Indeed. Totally agreed. I prefer Claude Code over other tools but this is abusive. I live in a 3rd world country and I sacrifice considerable amount of my income to pay Claude Code cause I’m able to work more and better but it is unsustainable to pay such high amount and get this poor rate limit. It is way disrespectful to loyal users

<img width="1008" height="506" alt="Image" src="https://github.com/user-attachments/assets/ffacfc63-9530-49bd-94d1-eae047c09024" />

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.

<img alt="Image" width="1008" height="506" src="https://private-user-images.githubusercontent.com/13174899/497623483-ffacfc63-9530-49bd-94d1-eae047c09024.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.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.bXN1_U7zmmLj2lRVVckZqMfcYRE-Q-J2vKzHYFPMZTg"> 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.

@nathancatania Agreed 100%. I used Claude now to do some summary job for my University and just 1 conversation in summarize some sections of a 60 page document took 2% of my weekly usage. I feel like a homeless saving my weekly usage to work and study.

I found something interesting... If you have Copilot Pro and use Sonnet 4.5 in the Copilot CLI you got this... 🤔

<img width="218" height="95" alt="Image" src="https://github.com/user-attachments/assets/f46a12f2-09a5-4cb9-9a8b-b6fd90f66b24" />

Perhaps unlimited until you hit the _limited_ rate _limit_ 😅?

Perhaps unlimited until you hit the _limited_ rate _limit_ 😅?

@

it could be possible, however as they consume anthropic APIs I do not think it actually hit a “rate limit” as we are capped in Claude Code. Microsoft will pay the bill of my usage (at least I think that’s the business model).

It’s sadly funny that you can access that for 10$ but hit rate limit with 200$ in CC. Which is true is that experience in Copilot CLI isn’t the same as Claude Code: it is slower and not so accurate.

I’m just so disappointed… at this point any complaint it is just catharsis. They clearly doesn’t care.

Two days of frustration & I'm about to just try a completely different thing. I'm honestly afraid of burning myself out testing models that make Sont look like a genius 🙁

@emarketbank this may not solve your 'active coding' / 'quota' issues but should help offload and delegate thinking and reasoning to a different model from within Claude Code (including code reviews):

https://github.com/BeehiveInnovations/zen-mcp-server

I will leave this here just for reference, since it is the longest comment thread related to this matter.

This user posted a report about his token usage. He used 10x less input tokens and 6x less output tokens and hit the weekly rate limit:
https://github.com/anthropics/claude-code/issues/9094#issuecomment-3380216960

Today expired my monthly subscription. Did not renew. Almost 1 week waiting for a statement from Anthropic related to this matter and not a single word. That's just very disrespectful to us (their customers).

Clearly they are making money from other sources and retail isn't relevant anymore. Anthropic employees are answering other issues but not a single one related with poor rate limits.

@non-stop-dev they have posted an update here: https://www.reddit.com/r/ClaudeAI/s/JYRQceJHfD

These limits are intentional and mostly affecting Opus, not Sonnet 4.5 (although limits have been reduced overall). I don't think complaining will do us any good at this point. The solution seems to be willing to pay more (either to Anthropic, or separately for Codex) to get more.

I’m not an Opus user, we only use Sonnet 4.1, and have now been forced to update to Sonnet 4.5. Since that change, I’ve been hitting the weekly limit much faster than usual. This only started happening after v2; previously, I never came close to the limit on a Max account.

It’s not just me either, my whole team is facing the same behavior, even though we have 15 active subscriptions for MAX 20x / $200.

Like others have reported here, the depletion rate feels far higher than expected. In some cases, users have mentioned hitting 16% of their quota within just 30 minutes of normal activity.

Would it be possible for the Anthropic team to provide more transparency, for example, by sharing logs or clarifying how token usage is measured differently between v1 and v2? From our perspective, this feels like a regression in both quality and usability, and clearer insights would help us understand what’s going on.

We’d greatly appreciate further clarification or investigation into this.

This issue has been inactive for 30 days. If the issue is still occurring, please comment to let us know. Otherwise, this issue will be automatically closed in 30 days for housekeeping purposes.

This issue has been automatically closed due to 60 days of inactivity. If you're still experiencing this issue, please open a new issue with updated information.

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.