[BUG] API Error: Server is temporarily limiting requests (not your usage limit) · Rate limited

Found 3 possible duplicate issues:

1. https://github.com/anthropics/claude-code/issues/53905
2. https://github.com/anthropics/claude-code/issues/53850
3. https://github.com/anthropics/claude-code/issues/53858

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

이 문제 너무심해. 무슨 3번요청하면 1번꼴로생겨

I have the same

Same here. It's rate limiting me to only 1 instance/ chat at a time

It's quite annoying that claude code works very unreliably last few days, but https://status.claude.com/ shows everything as green. Are these errors not even considered to be a problem by Anthropic?

It often rate-limits me few times an hour. Every time interrupting subagents, and making me waste tokens on restarting them (for some weird reason there is still no "retry" option in claude CLI). And it often makes hours wasted (because I was not at the keyboard to properly resume the process).

same error here

Same error here burned about 10% of "Current session" tokens on just this issue in one case.

Adding a data point, plus a diagnostic note and triage flag.

My instance: May 15, 2026, ~17:00 UTC, Opus 4.7, Max 20x plan, ~10-15% of session usage (so well clear of quota). Status page green. Support chat confirmed Claude Code retries 10 times with exponential backoff before surfacing this error, which puts the sustained throttle window at roughly 15-20 minutes, not a brief blip.

Diagnostic note: https://github.com/anthropics/claude-code/issues/53927 (closed as a duplicate of this issue) includes the raw error log dump, and the underlying response is HTTP 429 with error.type = "rate_limit_error". That's the same error class the API returns for actual quota hits. The "not your usage limit" copy is a client-side override on top of an identical-shape API response, which is presumably why the surfaced UX collapses two different root causes (infra capacity throttle vs. quota hit) into one error string.

Triage flag: the dedupe bot keeps flagging this thread against issues that are different error surfaces (https://github.com/anthropics/claude-code/issues/53850 and https://github.com/anthropics/claude-code/issues/53858 are quota burns, https://github.com/anthropics/claude-code/issues/53905 is a 500 with rate_limit_check_failed). It also flags new reports of this exact error (https://github.com/anthropics/claude-code/issues/59730, https://github.com/anthropics/claude-code/issues/53927) against unrelated threads. Worth disambiguating in any internal labeling so server-side capacity throttles don't keep getting bucketed with quota complaints and network errors.

UX gaps worth naming:

1. Identical surfaced error for "infra capacity event" vs. "your quota was hit." Two different root causes, same user experience, no way for the user to tell which one they're in.
2. No retry or resume option after the 10 backoff attempts exhaust. The session is just gone.

Echoing @xaionaro on the retry: an in-product retry would cap the damage when the throttle window outlasts the backoff schedule.

I have the same issue.

It's appearing to get worse, it's so bad today!

This is daylight robbery. Running into this issue for more than three hours now on four separate Cowork instances.

Happens more and more for me too 😕

Even worse today. One single failure of this type will ruin all running sessions and your all day.

If I initialize 2 Claude Code (Max Plan) sessions simultaneously, this issue will almost always occur.

Please fix!

fixxx

API Error: 400 Output blocked by content filtering policy
API Error: Server is temporarily limiting requests (not your usage limit) · Rate limited

I'm currently on a Claude Max 5x plan opened under a company account. I never ran into this problem on my old personal Claude Max 5x account, so it may be some kind of restriction Claude applies to newer accounts.
I also suspect it's tied to IP address reputation. The errors started showing up right after our company began routing through its own IP block over the past few days, instead of Comcast's. Switching to a T-Mobile hotspot while working in the office resolved the issue.

API Error: Server is temporarily limiting requests (not your usage limit) · This request would exceed your account's rate limit. Please try again later

but it actually burned my usage limit

<img width="1036" height="117" alt="Image" src="https://github.com/user-attachments/assets/356c0e41-29e1-4b66-a4d7-1d9d9132be0f" />

That was 0% to 100% in less than 30mins and the task was just removing two files and some additional logic it wasn't even able to complete the second one.

I would choose to use DeepSeek if this happens again, too many restrictions with Claude.

Deepseekv4 now has a 1M context window.

API Error: Server is temporarily limiting requests (not your usage limit) — I'm on the Max Plan 20x. These are pretty large limits for 5 hours — to actually burn through them you'd need to run many parallel tasks — but I can't do that because of this error. So it turns out that on the $100 plan I don't have enough token limits, and on the $200 plan I don't have enough server request limits. What kind of nonsense is this?

I'm on the $200 plan, first time ever getting this same issue. Its definitely a fatal bug, you can run into another chat and you won't run into this issue.

I think it's pretty obvious no actual human is involved with this project.

Same thing happened to me. Waste of a $100 subscription. I can't even use my subscription, and the support bot just closes the subscription

<img width="704" height="346" alt="Image" src="https://github.com/user-attachments/assets/d862e067-985a-430e-bdab-121816a309e1" />

<img width="796" height="269" alt="Image" src="https://github.com/user-attachments/assets/60c30330-1a97-4611-84e9-d592b8bf5065" />

Mine has not worked in 2 days

Same issue here

what is the point of having claude max when u can't run multiple instances ? are devs and sales retarded or what

I am using separate VMs per project, so have multiple sessions PLUS they are on different IPs. If i use teams of subagents, it just starts spewing these errors. Very frustrating. Plus spewing them on different sessions on different machines.
Also claude max.

<img width="1246" height="579" alt="Image" src="https://github.com/user-attachments/assets/61f8acf4-44ce-4a0c-ae8f-e8269bb2bad2" />

<img width="1283" height="273" alt="Image" src="https://github.com/user-attachments/assets/1ea1fe2a-10e7-4af7-b7cd-c185ad99d254" />

I noticed it happens more when i use a team of subagents in at least one session.

API Error: Server is temporarily limiting requests (not your usage limit) · Type 2b rate limited. Please try again later.
I received this for the whole day. It makes me crazy

I have discovered a workaround - if you sign up for the GPT Codex Plan, you get a better coding agent and limits that are actually reflective of use. Solved my problem 100%.

<img width="1162" height="215" alt="Image" src="https://github.com/user-attachments/assets/a459fde0-4f94-4a3d-b1f9-422b3e77540b" />

what can I say

I am experiencing the same issue. I have NEVER in my history of being a claude poweruser for about a year now, using it every single day with massive context windows and multiple terminal tabs open experienced this bug.

But today I can't even use claude at all. It completely locks me out. I couldn't even submit a bug report. This is a SERIOUS bug that needs fixing immediately.

<img width="785" height="101" alt="Image" src="https://github.com/user-attachments/assets/a66623aa-c5cc-4e80-aa20-0064356c18f6" />

<img width="781" height="226" alt="Image" src="https://github.com/user-attachments/assets/6df50366-7014-49ae-85be-763ca784a747" />

I have absolutely no brand loyalty and will happily take my $200 monthly subscription to another provider if this issue is not resolved

Followup on this issue:

<img width="941" height="382" alt="Image" src="https://github.com/user-attachments/assets/a53e1b56-155a-44da-b277-777b4c524c0f" />

According to claude, the issue is likely caused by the "high effort" feature. Claude recommends medium effort and only using opus when necessary. I disagree completely. Anthropic needs to slow down the rate claude makes requests to prevent the server side rate limit from being hit rather than relying on the server limit and essentially breaking everyone's projects and punishing their computers/internet for being fast. This makes ZERO sense from a product standpoint.

<img width="1110" height="721" alt="Image" src="https://github.com/user-attachments/assets/c299b0c3-d605-430a-81b2-43c590a9fae2" />

Very annoying

@tamerdahdul : Before giving up on Claude Code (due to this and other ignored issues), https://github.com/henryaj/autoclaude used to mitigate the issue for me.

workaround : /model claude-opus-4-6[1m]

same issue, trying to use a team of agents; what's the point of agentic if we can't delegate to multiple sessions on Max??

same issue here, any possible solution?

fix this issues, what kind of business you are running tell me

<img width="509" height="395" alt="Image" src="https://github.com/user-attachments/assets/067ad82e-3830-402a-9899-c27f445db0f8" />

i am on max plan 20x , fix this issue fast

<img width="598" height="275" alt="Image" src="https://github.com/user-attachments/assets/b014cb22-dec9-4a5b-bf73-80da721486e1" />

0%-93% usage limit in 15mins just re-reading

<img width="697" height="77" alt="Image" src="https://github.com/user-attachments/assets/596b571a-359f-49a6-b011-4e1ec73e052a" />

+1

<img width="1492" height="144" alt="Image" src="https://github.com/user-attachments/assets/1b3bbcca-228f-448c-879d-2f9ed1f46e5b" />

<img width="507" height="220" alt="Image" src="https://github.com/user-attachments/assets/dba4b8bb-60bc-41da-bd08-e78b53fc3416" />

unusable!

I'm getting this too. I'm nowhere near my session limit or my 5d limit, and yet everything I'm trying to do gets cancelled with "API Error: Server is temporarily limiting requests (not your usage limit) · Rate limited".

is there anyone here from Anthropic to try to address? trying to orchestrate across several agents, this isn't working. even with Medium effort still getting throttled, and it's becoming impractical to continue. Am I being forced / pushed to use Codex?

Claude model is not available, but don't worry it's not because you hit the limit messaging is just pure joke. It was huge waste of 200USD try to do something with and forced to deal with only Try again button during the sessions.

Hi, getting the same here

<img width="2205" height="402" alt="Image" src="https://github.com/user-attachments/assets/4b456bcd-edc1-4e6c-8b13-8e76f1f063b8" />
<hr/>

Usage data:

<img width="988" height="745" alt="Image" src="https://github.com/user-attachments/assets/114fb59c-05b3-461c-ad9c-98415600ec2c" />
<hr/>

Your current status:

<img width="904" height="862" alt="Image" src="https://github.com/user-attachments/assets/398a94b6-4090-4f86-bec8-68873a7d1db8" />

any issue to inform?

94% ¿wtf?

see my previous comment (47% and now magically used 94%)

<img width="985" height="745" alt="Image" src="https://github.com/user-attachments/assets/3b0b3685-53f6-4b92-a0ab-0d12528a09b8" />

can someone (support) explain this very strange case?

is there anyone here from Anthropic to try to address? trying to orchestrate across several agents, this isn't working. even with Medium effort still getting throttled, and it's becoming impractical to continue. Am I being forced / pushed to use Codex?

Well, you can complain to deaf ears, pause your coding for a month, or just switch. After a week of use, I can say GPT 5.5 is better at coding anyway.

Well, you can complain to deaf ears, pause your coding for a month, or just switch. After a week of use, I can say GPT 5.5 is better at coding anyway.

I second this. And the CLI is less buggy. And quotas are larger.

Anthropic has so far been amazing, but extremely unreliable at times. Personally, I will be switching. Anthropic support is completely unavailable and unreliable. I have an AI bot loop that never gets me to support.

I have for months been getting the same API Error: Server is temporarily limiting requests (not your usage limit) · Rate limited error. And, my consumption goes from 0% to 70-80% used after receiving these messages 10-20x. Every time I send a single message and receive the above response, my consumption rockets 3-5% or so. I have no idea why. So, if I check in multiple chats whether its working or not, I will burn through all my consumption over the course of an hour. If someone knows why, Id love to know. In the interim, I will be switching to chatgpt.

Opus, effort high: API Error: Server is temporarily limiting requests (not your usage limit) · Rate limited. Just burned through $10 of usage credit although current session was only at 20 % -- now is at 81%. Lost $10 and got no answer. Same as @vsgro

To all: Make sure to 👍this issue (OP) and👎the automatically-generated 3-day notice to close this issue. This issue is real and cost us all money. It should not be closed.

This is the most disruptive and user-hostile rate-limiting approach Anthropic could build:

1. Faceplant every agent tab with "API Error: Server is temporarily limiting requests (not your usage limit) · Rate limited"
2. Make zero effort to retry - if you've stepped away from your computer, work has ground to a halt
3. User manually tells each agent to continue
4. Rate limits kick in again, immediately return to 1.

Anthropic employees should try using the product the way their customers do and gain some empathy for what that experience is like. Boris is out there saying "you should be using loops doofus", but in practice the harness constantly pukes like this and any loop would be short-lived unless you have the magic internal Anthropic Claude that doesn't have these issues.

we got new model heavier then opus, and still have rate limitting with opus model. this is joke

API Error: Server is temporarily limiting requests (not your usage limit) · Rate limited got this again

API Error: Server is temporarily limiting requests (not your usage limit) · Rate limited got this again and again

This is the most disruptive and user-hostile rate-limiting approach Anthropic could build: 1. Faceplant every agent tab with "API Error: Server is temporarily limiting requests (not your usage limit) · Rate limited" 2. Make zero effort to retry - if you've stepped away from your computer, work has ground to a halt 3. User manually tells each agent to continue 4. Rate limits kick in again, immediately return to 1. Anthropic employees should try using the product the way their customers do and gain some empathy for what that experience is like. Boris is out there saying "you should be using loops doofus", but in practice the harness constantly pukes like this and any loop would be short-lived unless you have the magic internal Anthropic Claude that doesn't have these issues.

Use ChatGPT instead. I'm waiting for my sub to expire.

If this continues, I'm thinking of switching to Codex.

man this is anoying API Error: Server is temporarily limiting requests (not your usage limit)

<img width="918" height="355" alt="Image" src="https://github.com/user-attachments/assets/a65841b0-0b64-4581-9925-773376cb1d12" />

same, unusable

same issue here

managed to edit a single line. Max x20 plan. is this a joke?

<img width="907" height="700" alt="Image" src="https://github.com/user-attachments/assets/9201ccb7-fb2d-41ad-b8a4-60792b0c7526" />

oh and the classic "(not your usage limit)" followed by "You've hit your usage limit" (I haven't)

same issue. and each time i get this 20% of my 5x session gets eaten up! this is outrageous

This is hitting me today, wasting time and tokens on killing long running sessions, and making me frustrated.

I would like to request a more customer friendly way of handling low capacity on anthropic's end.

• You could give us information about when during the day and week you want us to use claude for long running tasks to avoid this.
• You could give us a setting we can turn on to auto-resume work when the server is ready to take our session.
• You could give us free cold cache revivals on a cache YOU killed with a rate limit error that we did not earn, instead of making us pay for it, and then instantly killing our sessions again.

Just some ideas.

Same issue since yesterday.
This morning I haven't used Claude Code (Max x5) and when I tried to use, the usage monitor said that I used 10% of 5 hour limit.
How?
I only got this error: API Error: Server is temporarily limiting requests (not your usage limit) · Rate limited.

this shit is now instead of throwing error just not working ? i can't run 3 agents at the same time, u know i never reach my 5h usage with only 2 agents. 3rd agent is just keep fucking churning for 10 minutes

getting the API error with just 1 agent running, what the actual F.....

<img width="422" height="85" alt="Image" src="https://github.com/user-attachments/assets/74c08e8b-c8ad-4977-bc10-604d1fbbbe25" />

Open 4.8 / High. As a Max user, I'm not able to run a high effort single agent? Why?

where is any resemblance of customer support here?

Look, even regular companies have outages, and we all know Anthropic has some extreme problems with keeping their servers up. It happens: it's not realistic to expect a business to always perfectly provision capacity 100% of the time.

BUT EVEN SO, YOU CAN STILL CONTROL HOW YOU COMMUNICATE WITH YOUR CUSTOMER!!! This does not need to be a secretive mysterious message (relayed at the exact moment Claude itself is incapable of explaining it)!

It could _absolutely_ provide more info, or _at very least_ tell the user when more info is coming. Again, you can't perfectly provision servers ... but you _can_ have the decency to communicate honestly/openly. That _IS_ something you can control, so not doing it is incredibly rude and disrespectful!

Yeah, the Anthropic mgmt reading this are gently weeping, while wiping away their tears with $100 bills...

Having the same with 1-2 parallel sessions on the 20x plan. I will probably be switching to byteplus because of that.