[BUG] "Rate limit reached" on 1M context mode despite plenty of usage remaining

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?

Claude Code immediately returns API Error: Rate limit reached when using Opus 4.6 (1M context) mode, even though usage dashboard shows plenty of remaining capacity.

Weekly limit (all models): 36% used
Weekly limit (Sonnet only): 3% used
Session limit: 39% used

After switching to the standard (non-1M context) mode, the error disappeared and everything works fine.
The 1M context mode was not heavily used — the error persisted even after not using Claude Code for over 24 hours, performing /login, and starting a completely new session.

What Should Happen?

Requests should succeed since usage dashboard shows plenty of remaining capacity. At minimum, the error message should indicate that the 1M context mode has a separate rate limit, and the usage dashboard should display 1M context usage separately.

Error Messages/Logs

API Error: Rate limit reached

Steps to Reproduce

1. Open Claude Code with Opus 4.6 (1M context) mode
2. Send any message (e.g., "hi")
3. Immediately receive API Error: Rate limit reached
4. Run /login → Login successful → send message again → same error
5. Close session entirely and start a new session → same error
6. Switch model to standard Opus 4.6 (non-1M context) → works fine

Claude Model

Opus

Is this a regression?

I don't know

Last Working Version

No response

Claude Code Version

v2.1.71

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

PowerShell

Additional Information

Usage at time of error (from claude.ai/settings/usage):

Current session: 39% used
Weekly limit (all models): 36% used (resets Friday 1:00 PM)
Weekly limit (Sonnet only): 3% used (resets Friday 4:00 PM)
Plan: Claude Max

The core UX issue is that 1M context mode appears to have a separate, stricter rate limit that is:

Not visible anywhere in the usage dashboard
Not indicated in the error message (generic "Rate limit reached")
Not clearly documented for end users

This makes it impossible to diagnose the problem. Users see plenty of remaining capacity on the dashboard but are completely blocked in 1M context mode with no explanation.