[BUG] Context Usage Hints Are Inaccurate When Using Sonnet w/ 1M Context

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?

I am using Claude Code with Sonnet 4.5 1M context window, but CC warning me to compact or switch to Sonnet 1M even though I'm already using it and should be nowhere near 100% context usage.

In the screenshot below you can see I got the hint 7% context left · try /model sonnet[1m], and the status line is warning me Context low (7% remaining) · Run /compact to compact & continue, but the status line also (accurately) shows that I'm using Sonnet 4.5 with 1M context window because I have "ANTHROPIC_DEFAULT_SONNET_MODEL": "claude-sonnet-4-5-20250929[1m]" set in my settings.json. The hint message is instructing me to switch to the 1M context variant of Sonnet 4.5, but you can see that's what I'm already using.

I then ran the /context slash command which accurately shows my current session's usage based on 1M context window, which contradicts the hint and status line messages that claim I only have 7% context remaining

<img width="1107" height="518" alt="Image" src="https://github.com/user-attachments/assets/30d8768a-3b2a-4cbe-a0f5-9c5ed709853a" />

It seems like when I use "normal" Sonnet 4.5 (200K context window) the hints and status line are accurate, it's just with Sonnet 4.5 1M context window that the calculations are wrong

What Should Happen?

The hints and status line warnings that instruct me to compact or use Sonnet 1M context should only be generated when the session's context usage is actually approaching 100% as displayed by /context

Error Messages/Logs

Steps to Reproduce

This can be reproduced by using Sonnet 4.5 (with 1M token context) for a long-running session, and once the context usage gets to 200K+ tokens the hints and status line will start warning you to compact even though you're only using <25% of the 1M context window

Claude Model

Not sure / Multiple models

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.0.2

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_