[BUG] Claude Code shows 'Usage limit reached' despite session at 2% and weekly at 32%
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?
What's wrong?
Claude Code is blocking all input with a 'Usage limit reached' message, but the claude.ai usage panel shows I am well under both limits.
Plan: Pro
Usage at time of error:
- Current session: 2% used (resets in ~2h)
- Weekly (all models): 32% used (resets in ~2h 23min)
- Claude Design: 0% used
Actual behavior:
Input is completely blocked. The enforcement and the displayed usage panel clearly disagree.
Screenshots:
<img width="850" height="171" alt="Image" src="https://github.com/user-attachments/assets/0d3509fd-0b86-4ca2-8d48-7783f775cb89" />
<img width="1234" height="786" alt="Image" src="https://github.com/user-attachments/assets/a17a8194-8fde-4c06-b089-6b58ec2d81fe" />
Additional context:
The claude.ai web interface works normally on the same account at the same time, confirming the issue is isolated to Claude Code's internal limit tracking.
What Should Happen?
Expected behavior:
Claude Code should accept prompts normally, since both session and weekly limits are far from exhausted.
Error Messages/Logs
API Error: Usage credits required for 1M context · turn on usage credits at claude.ai/settings/usage, or use --model to switch to standard context
Steps to Reproduce
- Open Claude Code
- Attempt to send any prompt
- Input is blocked with 'Usage limit reached'
Claude Model
Sonnet (default)
Is this a regression?
Yes, this worked in a previous version
Last Working Version
_No response_
Claude Code Version
2.1.145 (Claude Code)
Platform
Anthropic API
Operating System
Windows
Terminal/Shell
VS Code integrated terminal
Additional Information
_No response_
15 Comments
I am facing the same issue. My plan is pro and limits are well under. Also getting this on Sonnet 4.6 Medium.
It looks like a bug that was introduced after the latest update. Go and have a look at your JSON file, and you might see a model of Sonnet 1M on it.
It looks like they're tightening up the Sonnet model.
So found something interesting, in my desktop app, Claude Code
Opus 4.7: Works
Sonnet 4.6: NOT working
Haiku 4.5: Works
I didn't change anything, Sonnet has been working for more than a month or so.
Confirming this bug still occurs on Claude Code binary 2.1.149 (the latest at time of writing), via Claude Desktop on Windows 11.
Same exact error message:
Repro pattern from my logs (2026-05-23):
Bug fires consistently within ~2s of sending any message on Sonnet 4.6, even in a fresh session with empty context. 8 consecutive failures across 24 minutes.
Workarounds that work:
This suggests Claude Code is force-enabling the 1M context flag for Sonnet specifically, without a graceful fallback to 200k when the user does not have usage credits enabled. The fix is probably to either:
Environment:
I'm experiencing the same issue.
Same issue on max plan as well
Same issue on max plan Sonnet 4.6 as well
Same issue on plus plan
For those still hitting this issue, here's what worked for me:
Root cause: The recent update silently switched Sonnet 4.6 to 1M context by default. This triggers billing via "usage credits" even if you're on a Pro plan — and if you've activated usage credits to unblock Sonnet, it will consume your extra credit balance fast.
Fix: Disable 1M context manually to go back to the standard 200k window.
Add this to your
~/.claude/settings.json:Then restart Claude Desktop. Sonnet 4.6 works normally again with no credit consumption.
⚠️ Do NOT activate usage credits as a workaround — it will bill you for every session at 1M context rates.
For those still dealing with this issue, this is completely unacceptable at this point.
I was forced to switch away from Sonnet and use Opus/other models just to continue working because the standard workflow became unusable. The limits are being hit insanely fast and it completely blocks ongoing projects. This has been happening for days now and there is still no clear communication or proper fix.
What makes this worse is that many of us did not even realize that the recent changes apparently enabled much larger context usage by default. If that is true then users should have been clearly warned because token usage and limits suddenly behave very differently. A lot of people are burning through credits or hitting restrictions far earlier than expected.
This is not a minor inconvenience. People rely on these tools for actual work and deadlines. Silent changes to usage behavior combined with unstable availability is a terrible experience for paying users.
At minimum there should be:
transparent communication about what changed,
clear warnings when high-context modes are enabled,
temporary compensation or refunds for affected users,
and a proper fix instead of leaving users stuck for days.
I genuinely expected better reliability and communication than this.
same issue
I am facing the same issue, I am in to the Pro plan and it's showing the usage limt reached message and my usage was under 10%
<img width="1131" height="752" alt="Image" src="https://github.com/user-attachments/assets/f78f6d3d-0749-4d48-bcf9-95fc070faa87" />
Hit this exact scenario tonight (Max 20x, v2.1.199, macOS): every Claude Code request blocked with "session limit" at 100–102% in the status line, while the CLI's own
/usagepanel and claude.ai settings both showed the session at 3% — and claude.ai chat worked normally throughout.Found a likely mechanism and an immediate fix. I'd been running 5 parallel sessions that legitimately exhausted a window; 3 surfaced the usage-limit UI but 2 hung in a stuck state. Those stuck processes were holding the enforcement counter — Ctrl+C on them cleared the block instantly, hours before the advertised reset. So before waiting out the clock, check every machine for wedged
claudeprocesses.Full evidence timeline (screenshots of both meters disagreeing in the same minute, cross-model/effort/surface repro) posted on #54750: https://github.com/anthropics/claude-code/issues/54750#issuecomment-4886131876.
Same issue here, even with the
/usagecommand I can see that I still have tokens:<img width="848" height="515" alt="Image" src="https://github.com/user-attachments/assets/f2047af1-0efa-4486-a465-fe67e48c7801" />
But if I ask something it says that I reach the limits:
<img width="663" height="133" alt="Image" src="https://github.com/user-attachments/assets/4296a351-f2ac-475d-91f1-3afd9bf2c585" />
the same here!
<img width="1135" height="375" alt="Image" src="https://github.com/user-attachments/assets/4dadcae7-5bd6-4546-86fe-6d5b27cdb44a" />
<img width="1086" height="749" alt="Image" src="https://github.com/user-attachments/assets/bd6c2255-eeb7-48e4-b740-b9834094a195" />
That's a bummer. I was waiting for about 4 hours to continue, and I simply wrote "please continue", and I got the next forced break right away.
@claude Don't you think you should have a look on bugs like this?? BTW, in Claude (MOBILE) everything works, so your CLAUDE CODE DESKTOP has a PROBLEM!
EDIT: What I did to solve this, LOGOUT and fresh LOGIN. This forced the app to re-read the real limits and the message went away.