Excessive token usage in Claude Code

Source: Hacker News

Bug Description

Starting from this morning with the update to Claude Code 2.1.1, token usage has become excessive. I am working on the same projects and routines that I have been using for months, but today the 5‑hour limits are reached 4+ times faster. During the holiday season I saw a 2× increase in usage, but the current spike is well beyond that. I am already at the maximum limit, whereas previously I would never exceed 50 % of the weekly cap. This sudden surge is making the tool unusable.

Environment Info

• Platform: Linux
• Terminal: gnome-terminal
• Version: 2.1.1
• Feedback ID: 60967402-d2b7-4e28-9779-2720680deeae

Errors

[
  {
    "error": "Error: NON-FATAL: Lock acquisition failed for /home/oleh/.local/share/claude/versions/2.1.1 (expected in multi-process scenarios)\n    at nZ$ (/$bunfs/root/claude:2991:2164)\n    at B6A (/$bunfs/root/claude:2991:1884)\n    at processTicksAndRejections (native:7:39)",
    "timestamp": "2026-01-08T11:58:17.963Z"
  },
  {
    "error": "Error: Request was aborted.\n    at makeRequest (/$bunfs/root/claude:869:3940)\n    at processTicksAndRejections (native:7:39)",
    "timestamp": "2026-01-08T12:20:28.962Z"
  }
]

Additional Observations

Running a plan mode in Opus consumes about 10 % of the token quota almost instantly, without even reading the entire database. In contrast, the Heiku model does not appear to be affected; I was able to use it for roughly an hour and only incurred 2–5 % token usage.