[BUG] API Error: Rate limit reached despite Claude Max subscription and only 16% usage
Open 💬 152 comments Opened Feb 28, 2026 by CaptainDaredevil
Preflight Checklist
- [x] I have searched existing issues and this hasn't been reported yet
- [x] This is a single bug report (please file separate reports for different bugs)
- [x] I am using the latest version of Claude Code
What's Wrong?
Every request immediately returns "API Error: Rate limit reached" despite Claude Max subscription being active and only 16% usage shown in /status.
What Should Happen?
Requests should succeed within Claude Max plan limits (only 16% used).
Error Messages/Logs
API Error: Rate limit reached
Steps to Reproduce
- Run claude logout, then claude login
- Open Claude Code — shows "Claude Max" in header
- Run /status — Login method: Claude Max Account, 16% session usage
- Type any message (e.g. "hi")
- Immediately get "API Error: Rate limit reached"
Claude Model
Sonnet (default)
Is this a regression?
I don't know
Last Working Version
_No response_
Claude Code Version
v2.1.63
Platform
Anthropic API
Operating System
Windows
Terminal/Shell
VS Code integrated terminal
Additional Information
Prior to this, credentials.json showed subscriptionType: null and rateLimitTier: null.
API credits were added as a workaround — they were consumed instantly (within seconds),
leaving the account in debt. After fresh login subscription is recognized
(subscriptionType: "max", rateLimitTier: "default_claude_max_5x") but rate limit
error persists despite only 16% usage. Session resets at 3pm Europe/Budapest.
152 Comments
I have similar issue , they have some problem with API rate limiting #29584
Same issue — Max 20x ($200/month), Windows, v2.1.63, only 16% session usage
Environment
What's happening
Intermittent
API Error: Rate limit reacheddespite very low usage:/statuscorrectly showssubscriptionType: "max"and login method as Claude Max AccountBehavior pattern
claude logout→claude logintemporarily fixes it (works for a few minutes)ANTHROPIC_API_KEYenvironment variable set/clearand/compactdo not resolve the issue permanentlyAdditional context
Related issues
Please man, fix this shit wtf
Same issue - Max 5x
ℹ️ Update on Wed 4 Mar
I logged in this morning and it was back to normal.
Environment
What's happening
Intermittent
API Error: Rate limit reacheddespite very low usage:/statuscorrectly shows subscription as MaxBehavior pattern
claude logout→claude logindoes NOT fix it/clearand/compactdo not resolve the issueANTHROPIC_API_KEYenvironment variable setAdditional context
here it's opus and sonnet
session was before a lot of upgrades
only haiku works
sometime it's working again after /compact with haiku and sometimes with disabling 1M context with env var (even if i'm not eligible)
claude team here
Still broken on 2.1.63 — $200 Max plan, additional evidence:
I'm on a $200/month Max plan and have been locked out since Feb 28 despite very low usage amounts. I was given a 7 day rate limiting lock out which seems wrong. The $200 Max plan limits should reset in hours, not days. Never had this problem in a year of Claude Code CLI usage...oddly this past week is one of my lowest usage periods in months...
Additional data points:
The prompt caching overcounting + 7-day lockout on a Max plan is a significant regression. If not resolved I will have to switch to Codex or some other platform.
Happy to provide logs if helpful.
Backuped and deleted my ~/.claude/projects and it's working.
edit: not for long grrr
I'm running into the same issue. I tried to switch from api key to pro subscription with claude code for vscode, but after removing my API key and logging out and switching account to subscription, claude code vscode extention keeps reporting:
API Error: Rate limit reachedI am however, able to log out and login using my subscription in with claude code cli, but not in vscode extension,
Also I noticed when logging in with claude code cli, it takes me through the authentication process twice before I can actually input any prompts.
Open your .claude settings json and make sure that you have exact this string: "forceLoginMethod": "claudeai",... Afterwards relogin to your session. It helped me.
@CaptainDaredevil , Thanks for the reply. Unfortunately that didn't work for me.
Happening to me OSX, latest version (2.1.68), Max $200 plan.
Any resolution on this Bug?
Experiencing the exact same issue.
Setup:
Problem:
After hitting the 4h session limit, every request returns "API Error: Rate limit reached" — even after the session resets. Affects all models (Opus and Sonnet).
/status shows:
Root cause identified:
The server sets
cachedExtraUsageDisabledReason: "out_of_credits"in~/.claude.jsonimmediately after every fresh login, despite the account having $182+ in remaining extra usage budget.What I've tried (all unsuccessful):
~/.claude.json,credentials.json,statsig/,cache/out_of_creditsflag → server resets it on next requestConclusion:
This is a server-side state issue. The account gets stuck in a rate-limited state after hitting a session limit, and no client-side action can resolve it. Anthropic support has been contacted but so far only provided automated responses suggesting steps that have already been taken.
This makes a $100/month subscription completely unusable. Please prioritize a fix.
I submitted a ticket and having the same issues, nothing has fixed it.
In my case I was using
opus[1M], didn't realize it was using extra usage credits past 200K. It burned up the $70 CAD we got for free a while back, so I thought it was a bug or an issue with authentication.Switched it back to default Opus 4.6 and it worked again. See https://github.com/anthropics/claude-code/issues/28975#issuecomment-3972206477
That fixed it for me
Same thing here since last weekend (feb.28th). Paying for a service that does not work.
I have same issue with Max 20x too, even tho I just opened new session today.
@edwinckc you saved my day!
/modelfixed the issue (by mean of coming back to "Default"@edwinckc saved my day too! switching the model back to default worked for me as well.
Same issue here. Ran /model and switched back to Default , it works now.
the 1m context window was too OP they had to nerf
Do check if the model you've selected is wrong or outdated
after using other model, it works
<img width="500" alt="Image" src="https://github.com/user-attachments/assets/bbc7aa9d-d074-4d80-880a-e665209026c3" />
Max x5 plan
Same issue right now. Was using Sonnet mostly this week. More Opus today / last night.
Dashboard at https://claude.ai/settings/usage, and local Account & Usage give the same numbers - 23% 5hr, 64% Weekly (77% Weekly Sonnet)
<img width="1184" height="1401" alt="Image" src="https://github.com/user-attachments/assets/f131abce-b3b0-4533-885b-4a9949c7f5dc" />
Same issue.
Same issue
In case you want to add more comments in this X post were I tagged the cowork team
https://x.com/BuiltOutLoud/status/2031798760896942423?s=20
It's strange, my claude code cli is working now but vscode plugin don't .
<img width="678" height="1250" alt="Image" src="https://github.com/user-attachments/assets/887cad3c-3f7f-46be-ac7d-93fe92b8f4e1" />
Same issue. Running simple (2 page doc) build execution at 55% context on Opus and hitting this on max plan…
Same issue
Same issue
<img width="431" height="222" alt="Image" src="https://github.com/user-attachments/assets/fac9f00c-c828-41a4-905c-b40b577b3e12" />
Same Issue. Trying really hard not to let me frustration boil over to unprofessionalism, but how can we make this a key part of workflow when there is no response or fix after several days. It feels irresponsible for the enterprise, or small business, to make this tool a critical part of their infrastcture given the lack of movement on this issue.
Same issue. Same problem as people describe. Checked claude status, just to verify full operational status. Productivity dropped and motivation following fast. Please fix this bug asap
For me changing model in vscode plugin to sonnet and then back to opus fixed the problem!
+++
MAX PLAN X20. For several days Opus and Sonnet giving big API limit and unworkable on cowork, Reinstall doesn't work. Atrocious this hasnt been fixed.
!Image
!Image
Can you explain this? What is vscode plug in?
Same issue, $200 plan, started a new conversation - it's just cowork that is showing me rate limit api reached; I switched from Cursor after spending $1500 on API but cowork and code are so terrible, I've wasted 3 days fighting with the models they won't do anything right - but getting locked out despite paying is toxic.
Same problem here. Solved changing the model from /fast to normal mode.
same issue, switching models in vs code does not solve the problem.
same issue max plan 200 $ a month, and cannot use service since more then 24 h... good job anthropic : )
I think I’ve at least found my mistake regarding the API limit. I had uploaded an HTML file for Claude to review. That apparently was too much. If I close the Claude window in VS Code and reopen it, it works again.
I’ve never used it with vs code
Sent from Proton Mail for iOS.
-------- Original Message --------
On Thursday, 03/12/26 at 21:22 Martin Krebs @.***> wrote:
I fixed it by toggeling extra usage and setting like 10 $ on my account ... for whatever reason that makes it work again
Same issue. Would be amazing if someone from Anthropic replied to this thread. I'd love to keep paying them money rather than cancel my subscription. This is a complete showstopper bug, it's hard to imagine something more critical than "entire program doesn't work."
Same issue, with any model, claude team is vibe coding too?
I don’t know if it was the reason why mine started working again, but switched to 4.5 haiku for a few chats then it worked again for the more advanced models
Sent from Proton Mail for iOS.
-------- Original Message --------
On Friday, 03/13/26 at 12:52 aespielberg @.***> wrote:
Having same issues as of today on my pro plan. Current session say 17% used..
Do you get the same if you switch to the cheapest model?
Sent from Proton Mail for iOS.
-------- Original Message --------
On Friday, 03/13/26 at 19:00 Lawrence Altaffer @.***> wrote:
Yup
Claude desktop lets me use it and told me to give you that
On Fri, Mar 13, 2026 at 2:02 PM Stable Genius @.***>
wrote:
--
Lawrence Altaffer
804 677 6659
Also with haiku 4.5?
Sent from Proton Mail for iOS.
-------- Original Message --------
On Friday, 03/13/26 at 19:03 Lawrence Altaffer @.***> wrote:
actually Haiku worked, how do i get back the other models?
On Fri, Mar 13, 2026 at 2:05 PM Stable Genius @.***>
wrote:
--
Lawrence Altaffer
804 677 6659
I just chatted randomly with haiku for a while, tested with better models again and then it worked 🤷
Sent from Proton Mail for iOS.
-------- Original Message --------
On Friday, 03/13/26 at 19:09 Lawrence Altaffer @.***> wrote:
Getting this error now as well...no workaround.
you updated claude code ? 1m context available for everyone, could help
Just got this error, what worked for me was to switch to a different model and switched back. Not a permanent fix but a good temporary workaround.
<img width="1932" height="630" alt="Image" src="https://github.com/user-attachments/assets/b88117ad-1948-463f-b2f2-568f6c0b7db5" />
Just turn on the extra usage and it works.... After thinking about it, it seems that I haven't been able to use the Claude Code CLI since I turned this option off a few days ago.
I have that on and it's still not working. I might try toggling it and seeing if that breaks something loose.
i have resolve this issue by just log out and log in claude code in terminal... 🤔
Same issue here. What fixed it was changing the models; then things started working as expected.
<img width="732" height="704" alt="Image" src="https://github.com/user-attachments/assets/00a59630-aa82-443b-bff7-551fac647679" />
I just ran into the same issue.
Despite having plenty of free usage left, I kept hitting the API limit.
All of this started right after I updated to the latest version.
The workaround I found was to switch the model to Haiku for a while, doing pretty much nothing else.
Then, I switched to the default (Sonnet 4.6) and increased the effort to high before getting back to work.
After an hour, I switched back to Opus, and it seems to be working fine for now.
Another one here with the same problem.
I'm paying for the Team Premium Seat plan and I can't use Claude Code with Opus since the 1M context update.
Great job, Anthropic...
Just for laugh, did you try to swich to haiku 4.5. Don’t know if this was a coincidence or a fix but it worked for me
Sent from Proton Mail for iOS.
-------- Original Message --------
On Sunday, 03/15/26 at 15:39 arbide @.***> wrote:
I can use haiku 4.5 and sonnet 4.6 but cannot use opus since the 1M context update.
Rate limits even on Max can happen when you cross the 200K input token threshold, which triggers 1M context mode. Some tips to avoid this:
Comprehensive guide to understanding and managing Claude Code costs: https://github.com/Sagargupta16/claude-cost-optimizer
Experiencing the same issue here — "API Error: Rate limit reached" despite having usage remaining.
Workaround: Simply running
claude loginagain resolved it immediately. No logout needed.Hope this helps anyone still stuck on this.
i fixed the error by adding credit on my extra usage. it is weird because claude doenst even consume it but like it fixed the problem for me after 2 days of errors. so now i like hav 15 bucks on extra usage who remain untouched. does it fixes the problem for someone else ?
Same issue. It works in CLI but not in the extension in VS code
I want to advise you not to use models with a parameter of 1 million context window. As I understand, they are used only by credit, not by subscription. Also make sure that in the configuration settings of Claude Code, you have not a free choice of using this tool, but specifically by subscription.
I want to advise you not to use models with a parameter of 1 million context window. As I understand, they are used only by credit, not by subscription. Also make sure that in the configuration settings of Claude Code, you have not a free choice of using this tool, but specifically by subscription.
hi I do not think so, I have a sub and use the 1 M models and my cred's do not go down. do you have any doc that would back up what you are saying ?
VS code giving the same error, with copilot cli or with claude code directly
Same issue, only two sessions. One in a different folder. I attempted solving it by re-login, no luck.
MIGHT HAVE FOUND THE ISSUE
So for some reason the VS Extension only allows opus 1m context at the moment, but that is not provided to pro or max plan to my understanding, therefore it costs extra credit usage and it costs a lot. The way i fixed it is uninstall, remove the permissions form claude and reinstall 5 minutes later, before logging in, set to Opus and log in after. After youre logged in, you are able to change the model to sonnet or haiku, which are permitted for pro and max plan, but you cant go back to Opus, as the only Opus model is the 1m context ( again, on vs code ). On the terminal you can still select between Opus 4.6 and Opus 4.6 1M so no issues there. Let me know if it fixed cause its working for me.
for me, I just logout/login and it works again.
I had the same issue. What worked for we was changing the model from opus to haiku (this model already worked) and then back to opus
Same problem for me when using Claude Code with a Pro account, Linux client (latest version).
I've tried logging out/in multiple times, over 12 hours.
Usage is currently reported at almost 0%.
That worked for me
Just tried this, fixed it for me too.
tried this, haiku works perfectly fine, but opus and sonnet don't want to be fixed
upd: I installed a new version of claude code, and after a few minutes the problem disappeared
same issue!! I still have 10% quota
edit: the method "change mode" really works!
I had a similar "API Error: Rate limit reached" error, but specifically triggered by slash commands that had
model: sonnetin their YAML frontmatter while running Opus 4.6 with 1M context.Removing the
model: sonnetline from the slash command headers resolved it for me. May be a different root cause than the issue reported here.Direct fix: Change the model to the standard opus
then claude logout /login
First chat using Claude Code VS Extension and this happened.. totally borked my chat session.
Switching models in the chat isn't recovering it... and quite hilariously Haiku can't even handle this prompt:
New chats work just fine.
this worked
One thing that's helped me debug these discrepancies: having the 5h and 7d remaining % visible in real time via the statusline. The
rate_limitsfield added in v2.1.80 gives you the actual usage numbers CC is tracking internally, so you can see whether the enforcement matches what the dashboard shows.I built a lightweight statusline that displays both windows with pace tracking: https://github.com/Astro-Han/claude-lens -- it's helped me notice when the 5h window resets vs the 7d, which often explains "impossible" rate limit hits.
I am having this issue too:
Environment
What happens
Workarounds that (don't) work for me
Happy to provide logs or screenshots if needed.
Same issue - Max 5x plan, Opus 4.6 (1M context), extra usage drained abnormally
Environment:
Symptoms:
Steps to reproduce:
Expected behavior:
This confirms the pattern: 1M context + Opus triggers abnormal extra usage billing even when plan limits are far from reached.
shit's broken fr
I had the same problem and changed my model to Sonnet 4.5 and it started to work.
Same issue. Downgrading to
v2.1.72of extension fixed it. Details: https://github.com/anthropics/claude-code/issues/33094#issuecomment-4071638291Hitting this exact issue on a Max plan. What helped me was routing lighter tasks (file reads, simple edits, test runs) through DeepSeek/Gemini instead of burning Opus quota on everything. I set up TeamoRouter (
https://gateway.teamo.ai/skill.md— 2-second install) which auto-switches between models based on task complexity. Keeps Opus/Sonnet for the heavy reasoning, offloads ~60% of routine requests. Brought my effective cost way down and haven't hit rate limits since.same happened to me - "API Error: Rate Limit Exceeded" no matter what i did. moved from sonnet 4.6 to opus and it started working again. This is disappointing
Thanks @Peter-Inge , I did the same workaround and worked as well.
This solution works for me too
Downgrading to v2.1.72 of extension fixed it by @freakynit
switching the model worked for me from sonnet to haiku and then again switching back to sonnet
I've update this morning to last v2.1.85 that was released 12 hours ago. For me it fixed the issue.
Before this release, I tried everything (except a clear reinstalling everything and clearing the cache), and couln't find a way.
I got hit with "API Error: Rate limit reached" immediately after trying to use sonnet[1m] in claude -p for the first time (while opus[1m] and both normal opus and sonnet work normally).
Thanks @Peter-Inge, your fix worked.
It's funny. I did the opposite - switched from the Sonnet to the Opus and it helped 🫠
I wonder if there's an issue when you start off at a Pro Plan and upgrade to a Max plan while still logged into Claude Code, the rate limit issue occurs.
Or if you've left a session open for many days, the rate limit issue occurs.
At the moment, I logged out in Claude Code and then logged back in an I haven't been having the rate limit issue. That rate limit was hit frequently after a few minutes.
I'm on a MacBook Pro and using the terminal "claude" app.
Getting this now, wondering if it's a infra issue with capacity and instead of just slowing down it's throwing errors, either way, CC needs to handle/BE needs to give it "slow down" back off responses
Got it last night after 17hours of grinding (heavy lifting).. Need to finish some work and guess what.. it's not gone..
so frustrating.. If I did too much tell me but don't show me that stats that aren't true. I haven't hit my week limit yet so come on.
sorry I am a bid frustrated at the moment. Any advise would be appriciated. I have switched back and forward. and sometimes its gone but my long session. (the one i need) is on rate limit.. leassons learned. (save more) haha
edit: switching to /fast-mode gave me some room again to work. quick close out.
same issue 20x max plan i only have 1 session of cc and sending message every 5 minutes on avg.
was getting this whole yesterday on 20x max
I have never reached the limits until now even working more than today. Today is sunday!!!
<img width="1191" height="909" alt="Image" src="https://github.com/user-attachments/assets/22da2ef5-e478-47a1-aefe-475b423615da" />
Same, I barely made a request on Claude Code and it consumed 33% usage with Max x20! What are they doing?
As first thing i have deleted my plan that will end in the middle of april, so i hope they understend, and i hope you will do the same. They first provided double the standard limits, and later reverted to what they called “standard,” which was actually half of the original limits — effectively erasing any meaningful reference to the previous baseline.
If they don't fix this ASAP, I'm switching to Codex and they'll never see a cent from me again. That's for sure.
And I'm not alone, a lot of people is tired of spending 200$ per month to be a beta tester.
This is unacceptable for a paid product. The Claude Code from Anthropic is being crippled by constant and aggressive **Rate Limit Errors**, making real usage practically impossible.
This is not about occasional limits. This is about being blocked repeatedly, unpredictably, and without any clear explanation. You start working, and suddenly everything stops with a rate limit error. No transparency, no guidance, no control. Just hard stops.
For a professional tool, this is a disaster. It completely breaks development flow, kills productivity, and makes it impossible to rely on the system for anything serious. A paid user should not feel like they are fighting the tool just to get basic work done.
If rate limits exist, they must be clear, predictable, and aligned with what users are paying for. Right now, the Rate Limit Error behavior feels arbitrary, excessive, and poorly designed.
This needs to be fixed.
same here
same here, Anthropic are really starting to make working with their platform a pain.
Same boat here. Maybe 30-40% of my messages get through. On a 20X max plan since forever. Never had this error ever. I thought maybe servers were a bit overloaded yesterday on Sunday. My usage shows I'm not anywhere near any limits.
One thing I just noticed is if you run /compact it will sometimes recover. My guess is there bug is they have a context limit for requests but are not calling /compact early enough
Seems to have improved this afternoon for me, I'm guessing they under provisioned there older models while they are training the new ones.
Same issue. I have 200$ (actually, 240$ with taxes) Max subscription, 9% of current session usage and I've got
API Error: Rate limit reached. Claude Status page shows all bars green.That's unacceptable for this price
UPD: seem like in my case problem appears when I run some custom slash commands
This is ridiculous. I've hit this issue for 3 days straight. Max 20x. WTF am I even paying for??
Same here, Max x20. Ridicolous
Still broken as of March 31, 2026. MAX plan. No fix. No acknowledgment.
This issue has been open for over a month with 85+ affected users and Anthropic has not fixed it, has not communicated a timeline, and has not offered any compensation.
My current data (today):
ENOTFOUND api.anthropic.comDNS failures with zero self-recoveryThis is a chronic, systemic billing/counter bug. The auto-close bot on my new report #40745 linked this issue as a "duplicate" — which only proves Anthropic has known for over a month and chosen not to fix it.
The bug is actively recurring today, March 31. Not resolved.
Demanding from Anthropic:
85+ participants affected. This is not acceptable.
Same problem here.
When I reached 47%, I was automatically switched to extra credits, and whatever balance I had was consumed within minutes. I couldn’t even log out properly and kept getting the “Credit balance too low · Add funds” message. After signing out and back in, I lost my progress.
Now I’m at 59% used and getting “API Error: Rate limit reached".
Same issue still occurring, am at 36% usage on Max subscription. “API Error: Rate limit reached"
I solved it by switching model /model from Opus to default
Same issue still occuring. I had 2 subscriptions using 2 different emails. When I switch from one subscription to the other which mean the new one has 0% usage, I still got the API Error: Rate limit reached
Yes I have 2 subscriptions as well - and switching causes the issue. But also when both have reset. I also /reset it first. That alone did not do anything.. but maybe together /reset /exit claude /model switch ??
Same error, NEED FIX!
@bcherny Any updates on this?
I recently downgraded my subscription from x20 to x5 because it’s absurd to pay $200 just to be rejected. Additionally, the usage goes crazy; one prompt consumes 10% of the limit. I’m not sure what you guys are planning, but I have a feeling it’s going to be detrimental to consumers.
Works when switching model from Opus to default (Sonnet-4.6). Faced this issue today itself. I think this is a major issue as I can see from the messages earlier in this thread. Please FIX THIS!
same here!! anyone got any reply yet?
the fix that works for me is switching models with /model and going back to the one you were using before!
Same here. Massive impact on my productivity. Anthropic - please FIX THIS!
Same issue... hit the limit then upgraded from PRO to MAX... and still capped to the limit... no matter wat
And this is the play!
Now every request i do claude code web cut 12%, when i start the request soon add more 12% i don't know what is the problem but so cannot work.
Same here.... plese FIX THIS!!!
<img width="439" height="187" alt="Image" src="https://github.com/user-attachments/assets/ae2e71b6-e7d7-465d-a5ba-3d15ddc101dc" />
API Error: Rate limit reached
Edit:
By changing the model to Opus, I was able to use it again, but Sonnet is giving me a "limit reached" error.
For me, 01/04/2026, this morning: rate limit 100% exhausted in ~50 minutes. I waited 4 hours and the rate limit was 100% exhausted again with a single prompt. ><
I get this too with Opus, over and over - very frustrating. Works on default. Guess my usage will go further and I won't need to upgrade, then 🙈
ESCALATION REPORT — April 2, 2026 — 12+ HOURS COMPLETE DOWNTIME
New compound bug discovered: After the rate limiter wrongly fires (Bug #1), Claude Code in VS Code enters a permanently broken state where every single prompt — including a single "." character — is rejected with "Prompt is too long." This is not a prompt length issue. It is session corruption that persists even after the reset timer expires.
My timeline (Claude MAX, $200/mo):
Even automated cron jobs sending minimal messages fail with "Prompt is too long."
@Tito-42 above confirms the same pattern today — rate limit exhausted in ~50 min, then again with a single prompt after waiting. This is clearly a systemic server-side failure, not isolated incidents.
This is now a P0 outage for paying subscribers. Anthropic has still not responded publicly or deployed a fix. Formal complaint filed with Anthropic support. Chargeback in progress if unresolved.
Getting this error as well sadly. I was having claude sonnet automating frontend review as a QA tester via Claude desktop when API rate limit reached even though I still have around 46% left in weekly usage.
Waited 2-3 hours now and retried it, but the API limit reached error is still there.
Claude is unusable now unfortunately despite being a Pro subscriber.
Switching from Sonnet to Opus did not resolve this issue for me
<img width="1323" height="848" alt="Image" src="https://github.com/user-attachments/assets/945fbb86-7857-4e45-a8d6-753caa9b9387" />
<img width="1564" height="870" alt="Image" src="https://github.com/user-attachments/assets/a85dd28c-ba4a-452e-be72-8ed8cf4a184b" />
I had the same fake rate-limit alarm. Fixed it by enabling Extra usage (under billing) with auto-reload on.
The trade-off: you need to add at least $15 in credits — but those credits won't actually be consumed unless you genuinely hit the plan limit. Once I did this, the phantom "Rate limit reached" error disappeared completely.
I have now switched to using Claude code with copilot-api. So might just cancel my subscription.
I haven't tested this with the Claude Code CLI today, so I can't confirm whether the same applies there, but I encountered a nearly identical issue with the VS Code Extension.
My observation is that this bug seems to occur after enabling Extra Usage and then either turning it off or exhausting the Extra Usage credits. (Alternatively, it may be triggered by repeatedly toggling between the 4-hour/weekly rate limit and Extra Usage — something around that cycle seems to be the condition.)
When this happened to me, my tokens were being consumed from Extra Usage rather than from my subscription's 4-hour quota — which was absolutely not "extra" in any meaningful sense.
This occurred after I received a $50 Extra Usage gift from Anthropic, enabled it, and then turned it off.
I asked Claude Code about similar issues reported in the tracker, and it told me the root cause appeared to be an authentication issue specific to the VS Code Extension.
The particularly frustrating aspect of this bug is that once a server-side flag gets flipped, it seems impossible to work around — even repeated logout/login cycles don't help.
Since then, I've been unable to use the VS Code Extension and have had to fall back to the Claude Code CLI.
I hope this helps narrow down the root cause.
One additional note: Anthropic distributed another round of plan-based gifts as a refund just yesterday, so I suspect this issue may be resurfacing for affected users as a result. That said, as long as Extra Usage is enabled and you still have a remaining balance, you should be able to continue working by letting it consume from that balance.
Adding my situation as a data point. My issue is Cowork-specific rather than general Claude Code usage:
Plan: Max 5x
Duration: 48+ hours of complete Cowork failure
What works: Claude Chat and Claude Code CLI on the same account work fine simultaneously
What doesn't: Cowork returns "API Error: Rate limit reached" instantly on every action — no processing delay at all
Symptom: "API Error: Rate limit reached" fires instantly on every Cowork action — including a simple "hello"
Usage: Well within limits per the dashboard
What I've ruled out: Local machine issue (account-specific, not environment-specific), insufficient credits, active API usage
Key differentiator: Claude Chat works perfectly on same account simultaneously, confirming this is not a genuine quota issue
Dashboard: Usage well within limits, no flags on the Limits page
Already tried: Sign out/in, restart Claude Desktop, support ticket (no response)
The error fires before any tokens are processed — identical to what others have described here. This is a server-side ghost rate limit state stuck on my account for Cowork specifically. Requesting a backend reset. My previously filed GitHub issue #34068 was closed as "invalid" which I believe was a misclassification.
Requesting a server-side reset of the Cowork rate limit state for my account. Happy to provide any additional diagnostics. CC: @ThariqS
Having this too on the Pro plan. Turning on extra usage fixes it (cc @haingt-dev ) but then it does charge me
I've had the same problem after updating through CLI and I was able to solve it by going into /config and realizing that for some reason it had put the default model to Opus 4.1 with the 1M token context. Manually changing back to Opus 4.1 fixed the issue. This may also be why @hyper-lode was able to fix this by enabling extra usage because afaik the 1M version is unavailable on the Pro plan.
Claude code in terminal will not use opus despite the fact that the claude code WEB allows for it, which is entirely frustrating and is the ONLY Solution to desktop terminal being a turd.
Also noting that somehow claude code web consistently has to compact after only about 3 fixes in the code.
Opus 4.6 is considered the "USER DEFAULT" according to the advertising when logged into terminal claude (my case powershell) but it onyl responded OUTSIDE Of APi error rate limit reached AFTER i swapped back to sonnet.
I think i'll deal with sonnet for now so i don't run into API errors
same here
same issue here
Same issue here, i will try to check if uninstall works, if it don't im calling PROFECO XD (sorry guys).
Getting rate-limited at 16% usage is infuriating — you're paying for access you can't even use.
I've moved most of my coding agent work to Franklin to avoid exactly this. The core difference is their YOPO (You Only Pay Outcome) model:
Subscription model (Claude Max): Pay $100-200/mo → get rate limited when you need it most → pay for access you can't use
YOPO model (Franklin): Pay only for completed tasks → no rate limits → wallet balance = hard cap, no surprises
How YOPO works practically: you fund a USDC wallet on Base or Solana. Each task that completes costs provider cost + 5%. Failed attempts, retries, dead ends — those are free. When the wallet is empty, it stops. No mystery rate limits.
The smart router inside also saves money by picking the cheapest capable model per task (55+ models). Simple stuff → cheap model. Hard reasoning → strong model. I was spending ~$12/mo vs $100/mo on Claude Max.
Free tier works with zero wallet (NVIDIA Nemotron + Qwen3 Coder):
Worth trying if you're tired of paying for subscriptions that rate-limit you.
Bug Report: VS Code Extension Silently Billing API Credits Instead of Plan Allocation
Plan: Pro
Claude Code Version: 2.1.156
OS: Mac
Date observed: May 28, 2026
What happened:
The extension began billing against API credits without my explicit consent, rather than using my plan session usage limit. I have 19% used on current session limit, but VS Code will not run Claude unless I turn on my auto-reload again. Usage from API was taken without my consent.
Expected behavior:
Claude Code should use my plan allocation and either stop or clearly ask before switching to API credit billing. The transition should never happen silently.
Steps to reproduce:
Use Claude Code via the VS Code extension until the session limit is reached
Observe the auto-reload prompt
Check console.anthropic.com billing -- API credits are being consumed without intentional user authorization
Additional context:
This appears to be the same issue documented in #32286 and #55998. The VS Code extension seems to be the specific trigger -- the CLI behaves correctly. Requesting review of the billing authorization flow in the extension, and a refund of any API credits consumed without explicit consent.
the surface mismatch is /status reads a session-local counter, but the rate limit decision uses a different signal: 5-hour window total + weekly + extra-usage balance + tier. when subscriptionType briefly flipped to null and your API credits drained in seconds, that drain hit the current 5h window. fresh login restored max/default_claude_max_5x in credentials.json but didn't reset the server-side window counter. /status shows local session %, not the window the rate limiter actually checks. real reset is at 3pm Budapest, not sooner.