Preflight Checklist
What's Wrong?
I am a paying Max 20x customer (plan confirmed in Settings → Usage: "Plan usage limits — Max (20x)"). My "Weekly limits — All models" counter is visibly resetting on an approximately 24-hour cycle instead of the 7-day rolling cycle that Anthropic documents as the expected behavior for this plan.
Current observed state (see attached screenshots):
- Plan header: "Plan usage limits — Max (20x)"
- Current session (5-hour window): behaves normally
- Weekly limits — All models: 5% used, "Resets in 15 hr 38 min"
- Weekly limits — Sonnet only: 0% used
- Weekly limits — Claude Design: 0% used
Earlier today the same "All models" counter showed "Resets in 19 hr 1 min" at 2% used, and the reset timer has been counting down on an approximately 24-hour loop — not a 7-day loop — since the previous reset.
Before this bug was filed, I pursued the issue through Anthropic's in-app support channel (Fin). That channel has confirmed in writing, twice, that the observed behavior "contradicts our documented specifications" and "is not the documented behavior for the plan's weekly limits". The same agent has also stated in writing that it cannot perform the account-level investigation required, cannot create an escalation ticket, and that no alternative support channel within Anthropic leads to a human reviewer for this issue. Conversation ID: 215474050081014, associated with account izyumskiy@gmail.com. Full quotes are in the Additional Information section.
To be explicit up front: I am NOT requesting a refund, NOT requesting a credit, and NOT planning to cancel my subscription. I intend to continue using Max 20x. I am filing this issue because the only path left to get a technical answer is the engineering team via this repository.
What Should Happen?
Per Anthropic's own help-center article "What is the Max plan?" (https://support.claude.com/en/articles/11049741-what-is-the-max-plan), Max plans have weekly usage limits that reset seven days after the session starts. The "Weekly limits — All models" counter should therefore display a reset timer of up to ~7 days (168 hours), not ~24 hours.
What I need from engineering is a classification and explanation of what is actually happening on my account. Specifically, is the observed ~24-hour reset cycle caused by:
(A) a bug in the rate-limit infrastructure that is affecting my account (and possibly others);
(B) an account-specific configuration — a flag, experiment assignment, A/B test, or non-default rate-limit bucket attached to my account; or
(C) an undisclosed general policy change for Max 20x that has not yet been reflected in the public documentation?
For whichever of A, B, or C applies, I need to know:
- If (A): is the bug known, and what is the expected fix / timeline?
- If (B): when was this configuration applied to my account, on what basis, was I notified, and how can it be reverted to the standard Max 20x configuration?
- If (C): when did the policy change, where is it documented, and under what terms of service was it applied without user notification?
A secondary sub-question that an engineer looking at my account state should be able to answer along the way: did my spending of prepaid Extra usage credits after hitting the weekly limit on Wednesday evening have any effect on the rate-limit bucket my account is currently assigned to?
Error Messages/Logs
Steps to Reproduce
Context: paid Max 20x subscription, account email izyumskiy@gmail.com, plan confirmed in Settings → Usage.
Observed sequence of events:
- Normal usage of Max 20x across Claude Code CLI and claude.ai web, drawing against the "Weekly limits — All models" counter.
- Wednesday, local evening (~20:00) — "Weekly limits — All models" counter reached 100% used.
- After the weekly counter was fully exhausted, I continued working by consuming previously purchased prepaid Extra usage credits (Extra usage is a prepaid balance that had been topped up earlier, unrelated to this bug).
- Thursday, ~21:00 local — the "Weekly limits — All models" counter reset. Elapsed time from exhaustion to reset: approximately 25 hours. NOT 7 days.
- From Thursday 21:00 onward, the "Resets in" timer on this counter has been ticking down on an approximately 24-hour cycle:
— Friday, earlier in the day: "All models: 2% used, Resets in 19 hr 1 min"
— Friday, hours later (see screenshots): "All models: 5% used, Resets in 15 hr 38 min"
— "Last updated: 1 minute ago" on the Usage page confirms the values are live.
- Other Weekly limits counters ("Sonnet only", "Claude Design") show 0% used and are not affected — the anomaly is specific to "All models".
- The "Current session" 5-hour window continues to behave as documented.
To reproduce on the engineering side: inspect the server-side rate-limit state of account izyumskiy@gmail.com and compare the configured reset period on the "All models" weekly counter against the standard Max 20x configuration. The observed ~24-hour reset cycle should be directly visible in your infrastructure.
Claude Model
Opus
Is this a regression?
Yes, this worked in a previous version
Last Working Version
Prior weekly cycle on the same Max 20x subscription (week ending Wednesday ~20:00 when the counter was exhausted at 100%). The 7-day documented behavior was functioning on that prior cycle. The anomaly began with the reset that occurred on Thursday ~21:00 of the current cycle.
Claude Code Version
2.1.119 (Claude Code)
Platform
Anthropic API
Operating System
Windows
Terminal/Shell
Other
Additional Information
Support channel timeline (Anthropic in-app "Get help" → Fin)
Conversation ID: 215474050081014, associated with account izyumskiy@gmail.com.
Direct quotes from the Fin support agent during that conversation:
On the behavior itself:
"what you're seeing (a ~24 hour reset) is not the documented behavior for the plan's weekly limits"
"The behavior you're observing contradicts our documented specifications."
On the ability to investigate:
"I'm unable to perform the specific account-level technical investigation you've requested, including checking server-side configurations, rate-limit buckets, or A/B test assignments on your account. These require access to backend systems and engineering review that's beyond my capabilities."
On the ability to escalate:
"I don't have the ability to create escalation tickets, generate case reference numbers, or route your request to a specialized technical team."
"I apologize that I cannot provide the case number, response timeframe, or formal escalation confirmation you've requested. I'm unable to help further with the technical investigation you need, as it requires backend system access and engineering review that I don't have."
On the existence of an alternative human support channel (after I listed the two routes the agent itself had proposed):
"Both routes I mentioned lead to this same support channel with me (Fin). They are not separate queues or different teams with human reviewers."
"the routes I listed are the same support channel you are currently using. They do not lead to a different system or human support team."
In summary: a paying Max 20x customer, experiencing behavior Anthropic's own support admits contradicts documentation, has been told that there is no mechanism inside Anthropic's support channels to reach a human reviewer. This GitHub issue is the remaining avenue.
Account / environment details
- Plan: Max 20x (paid, active)
- Account email: izyumskiy@gmail.com
- Payment method: Mastercard ending 7869
- Billing cycle: last full charge €122.40 on Apr 6, 2026; next full charge due ~May 6, 2026
- Recent invoices (all paid): €21.96 (Mar 30, 2026), €89.91 (Apr 2, 2026), €122.40 (Apr 6, 2026) — reflecting sequential upgrades Pro → Max 5x → Max 20x, pro-rated
- Extra usage: prepaid balance model, €85 monthly spend cap, €57.40 already spent this cycle, €27.59 remaining prepaid balance; Extra usage toggle OFF, Auto-reload OFF
Attachments
screens.docx — full screenshots of Settings → Usage page showing:
- Plan header "Plan usage limits — Max (20x)"
- Weekly limits — All models counter with the ~24-hour reset timer
- Current session 5-hour window state
- Extra usage state (prepaid balance, spend limit, toggle/auto-reload positions)
Related existing issues (different symptoms, related area)
For the benefit of triage, existing related issues I have seen in this tracker, which describe different Max 20x rate-limit symptoms and are not duplicates of this one:
screens.docx
Preflight Checklist
What's Wrong?
I am a paying Max 20x customer (plan confirmed in Settings → Usage: "Plan usage limits — Max (20x)"). My "Weekly limits — All models" counter is visibly resetting on an approximately 24-hour cycle instead of the 7-day rolling cycle that Anthropic documents as the expected behavior for this plan.
Current observed state (see attached screenshots):
Earlier today the same "All models" counter showed "Resets in 19 hr 1 min" at 2% used, and the reset timer has been counting down on an approximately 24-hour loop — not a 7-day loop — since the previous reset.
Before this bug was filed, I pursued the issue through Anthropic's in-app support channel (Fin). That channel has confirmed in writing, twice, that the observed behavior "contradicts our documented specifications" and "is not the documented behavior for the plan's weekly limits". The same agent has also stated in writing that it cannot perform the account-level investigation required, cannot create an escalation ticket, and that no alternative support channel within Anthropic leads to a human reviewer for this issue. Conversation ID: 215474050081014, associated with account izyumskiy@gmail.com. Full quotes are in the Additional Information section.
To be explicit up front: I am NOT requesting a refund, NOT requesting a credit, and NOT planning to cancel my subscription. I intend to continue using Max 20x. I am filing this issue because the only path left to get a technical answer is the engineering team via this repository.
What Should Happen?
Per Anthropic's own help-center article "What is the Max plan?" (https://support.claude.com/en/articles/11049741-what-is-the-max-plan), Max plans have weekly usage limits that reset seven days after the session starts. The "Weekly limits — All models" counter should therefore display a reset timer of up to ~7 days (168 hours), not ~24 hours.
What I need from engineering is a classification and explanation of what is actually happening on my account. Specifically, is the observed ~24-hour reset cycle caused by:
(A) a bug in the rate-limit infrastructure that is affecting my account (and possibly others);
(B) an account-specific configuration — a flag, experiment assignment, A/B test, or non-default rate-limit bucket attached to my account; or
(C) an undisclosed general policy change for Max 20x that has not yet been reflected in the public documentation?
For whichever of A, B, or C applies, I need to know:
A secondary sub-question that an engineer looking at my account state should be able to answer along the way: did my spending of prepaid Extra usage credits after hitting the weekly limit on Wednesday evening have any effect on the rate-limit bucket my account is currently assigned to?
Error Messages/Logs
Steps to Reproduce
Context: paid Max 20x subscription, account email izyumskiy@gmail.com, plan confirmed in Settings → Usage.
Observed sequence of events:
— Friday, earlier in the day: "All models: 2% used, Resets in 19 hr 1 min"
— Friday, hours later (see screenshots): "All models: 5% used, Resets in 15 hr 38 min"
— "Last updated: 1 minute ago" on the Usage page confirms the values are live.
To reproduce on the engineering side: inspect the server-side rate-limit state of account izyumskiy@gmail.com and compare the configured reset period on the "All models" weekly counter against the standard Max 20x configuration. The observed ~24-hour reset cycle should be directly visible in your infrastructure.
Claude Model
Opus
Is this a regression?
Yes, this worked in a previous version
Last Working Version
Prior weekly cycle on the same Max 20x subscription (week ending Wednesday ~20:00 when the counter was exhausted at 100%). The 7-day documented behavior was functioning on that prior cycle. The anomaly began with the reset that occurred on Thursday ~21:00 of the current cycle.
Claude Code Version
2.1.119 (Claude Code)
Platform
Anthropic API
Operating System
Windows
Terminal/Shell
Other
Additional Information
Support channel timeline (Anthropic in-app "Get help" → Fin)
Conversation ID: 215474050081014, associated with account izyumskiy@gmail.com.
Direct quotes from the Fin support agent during that conversation:
On the behavior itself:
Account / environment details
Attachments
screens.docx — full screenshots of Settings → Usage page showing:
Related existing issues (different symptoms, related area)
For the benefit of triage, existing related issues I have seen in this tracker, which describe different Max 20x rate-limit symptoms and are not duplicates of this one:
The anomaly in this report is specifically a ~24-hour reset cycle on the weekly counter, which is a distinct symptom.
screens.docx