[BUG] Unexpected high usage (49%) for minimal session work
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?
Description
Experiencing unexpectedly high usage consumption for a session with minimal actual work performed.
## Session Details
- Model: Claude Opus 4.5 (
claude-opus-4-5-20251101) - Session age: ~30 minutes since reset
- Usage reported: 49% of session quota
## Actual Work Performed in Session
The session was a continuation from a previous conversation (context was summarized). The actual work in the NEW session was:
- Answered 3 short clarifying questions about a workflow plan
- Updated one markdown plan file (~300 lines)
- Created one YAML workflow file (~170 lines)
- A few TodoWrite calls
Total: ~4-5 tool calls, minimal back-and-forth
What Should Happen?
## Expected Behavior
Given the minimal work (essentially creating one file after brief Q&A), usage should be significantly lower than 49%.
## Possible Factors
- Large context summary passed from previous session continuation
- Pre-loaded file contents in system reminders at session start
- Opus model pricing
Even accounting for these, 49% seems disproportionate for ~10-15 minutes of light work.
Error Messages/Logs
Steps to Reproduce
Steps to Reproduce
- Start a Claude Code session with Opus model (
--model opus) - Have a long/complex conversation that eventually hits context limits and gets compacted/summarized
- Continue working in the same session after compaction (or the session auto-continues from the summary)
- Perform minimal work in the new context window:
- A few Q&A exchanges (3-5 messages)
- Create/edit 1-2 small files
- Check usage with
/costor observe the usage indicator
Observed: Usage shows ~49% consumed despite minimal actual work post-compaction
Expected: Usage should reflect only the work done, not disproportionately high for simple file creation
## Additional Context
The session continuation included:
- A
<system-reminder>block with a long conversation summary (~500+ lines) - Pre-loaded file contents from previous
Readtool calls shown in system reminders - An existing plan file (~300 lines) that was read and updated
It appears the large input context from the continuation may be getting counted repeatedly or disproportionately against the usage quota.
Question
Is this expected behavior for Opus with session continuations, or is there potentially a bug in usage calculation?
Claude Model
Opus
Is this a regression?
I don't know
Last Working Version
_No response_
Claude Code Version
2.1.4 (Claude Code)
Platform
Anthropic API
Operating System
Windows
Terminal/Shell
PowerShell
Additional Information
_No response_
This issue has 5 comments on GitHub. Read the full discussion on GitHub ↗