[BUG] Bug: Extra Usage charged despite available plan capacity + false rate limit errors in Claude Code

Open 💬 15 comments Opened Mar 9, 2026 by skywalker89

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?

Subject: Bug: Extra Usage charged despite available plan capacity + false rate limit errors in Claude Code

Hi Anthropic Support,

I'm experiencing a bug with Claude Code on my Pro plan that is both blocking my usage and incorrectly charging my Extra Usage budget.

The Problem:
Claude Code returns "API Error: Rate limit reached" even though my plan limits are far from exhausted. At the time of this report, my usage dashboard shows:

  • Current session: 5% used
  • Weekly limit (All models): 50% used (resets Friday 7:00 AM)
  • Sonnet only: 3% used (resets Wednesday 8:00 AM)

Despite all these limits being well within range, Claude Code refuses requests with a rate limit error.

Extra Usage incorrectly consumed:
What concerns me most is that my Extra Usage budget has been drained to €0.00, with €5.91 spent against a €5.00 monthly limit (118%). This appears to have happened automatically — my plan limits were not fully exhausted, yet the system charged Extra Usage credits anyway. Auto-reload is off, so once the budget was consumed, I started getting hard rate limit errors with no way to continue.

This does not make sense: Extra Usage should only kick in when plan limits are reached, but my plan limits clearly had remaining capacity.

Possible cause:
I found several recent GitHub issues describing the exact same behavior:

  • Issue #29579: Rate limit reached despite Max subscription and only 16% usage
  • Issue #25778: Misleading rate limit error when safety filter is triggered
  • Issue #27603: Rate limit error on sonnet-4-6[1m] model despite 27% session usage — the API internally returns "Extra usage is required for long context requests" even when plan capacity is available

It's possible that the long-context model or an internal error triggered Extra Usage billing when it should have used my included plan allocation.

My request:

  1. Please investigate why Extra Usage was charged while my plan limits had remaining capacity.
  2. Please refund the €5.91 in Extra Usage charges, as these appear to be caused by a known bug.
  3. Please clarify whether the 1M context model requires Extra Usage on the Pro plan, as this is not clearly documented.

Account region: Germany (EU)
Plan: Pro ($20/month)
Claude Code version: latest
OS: Linux (Ubuntu)

Thank you for looking into this.

Best regards,
Dan

What Should Happen?

refunding the wrongly charged USD

Error Messages/Logs

Steps to Reproduce

  1. Have an active Claude Pro plan subscription
  2. Use Claude Code with both Pro plan credentials and Console/API credentials configured
  3. Work normally in Claude Code — at some point, Extra Usage credits are consumed

(€5.91 charged) even though plan limits show remaining capacity:

  • Current session: 5% used
  • Weekly limit (All models): 50% used
  • Sonnet only: 3% used
  1. Extra Usage budget is fully drained (€0.00 balance, 118% of €5 limit spent)
  2. Claude Code then returns "API Error: Rate limit reached" on every request
  3. /status confirms plan limits are NOT exhausted
  4. The user was never explicitly prompted to consent to API credit usage,

which contradicts the documented behavior that "All transitions to
API credit usage require explicit user consent"

Workaround:
Running claude logout followed by claude login with only Pro plan
credentials (no Console credentials) resolved the rate limit error.
The incorrectly charged €5.91 in Extra Usage remains unrefunded.

Claude Model

Opus

Is this a regression?

I don't know

Last Working Version

2.1.71

Claude Code Version

2.1.71

Platform

Other

Operating System

Ubuntu/Debian Linux

Terminal/Shell

PowerShell

Additional Information

_No response_

View original on GitHub ↗

This issue has 15 comments on GitHub. Read the full discussion on GitHub ↗