Token usage limit reached significantly earlier than in previous versions

Status Open
Maintainer reply None cached
Activity 2 comments · opened Mar 24, 2026

Description

The token usage limit is being reached much earlier compared to previous versions of Claude Code. In this session, the limit was hit during a moderate-complexity bugfix task (NPE crash fix with diagnostic logging + test writing), which would have completed comfortably within limits in prior versions.

Details

  • Model: Claude Opus 4.6 (1M context)
  • Task: Bugfix with ~3 file edits, 2 gradle builds, 1 GitLab ticket creation, and spawning a test-writing agent
  • Reset window: ~4 hours — this feels disproportionately long for the amount of work completed
  • The limit was hit when spawning a subagent (android-unit-test-expert), suggesting agent spawns may consume tokens more aggressively than expected

Expected behavior

A session of this complexity should complete within token limits, as it did in previous versions. The 4-hour reset window also seems excessive.

Environment

  • OS: macOS (Darwin 25.3.0)
  • Claude Code CLI
  • Date: 2026-03-24

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗