claude ai-agents developer-tools bug-reports

Excessive token usage in Claude Code

github.com

February 21, 2026

1 min read

🔥🔥🔥🔥🔥

45/100

Summary

Claude Code version 2.1.1 exhibits excessive token usage, consuming tokens at a rate more than four times faster than previous versions. This issue has been reported specifically on Linux platforms.

Key Takeaways

The update to Claude Code version 2.1.1 has resulted in excessive token usage, with users reporting consumption rates over four times faster than previous versions.
Users have noted that they are hitting usage limits significantly quicker than before, with one user stating they reached their 5-hour limit multiple times in a single day.
The issue appears to be specific to the Linux platform, as users on other platforms, like Heiku, are not experiencing the same excessive usage.
Errors related to lock acquisition and aborted requests have been reported, indicating potential stability issues with the new version.

Read original article

View story →

Coverage & discussions

Hacker News · github.com

Excessive token usage in Claude Code ↗

Community discussions

Hacker News discussion ↗

Historical count: 5 · 5 saved in sample · Historical sample

Analysis ready

Community Sentiment

Negative

Positives

Professionals appreciate Claude's ability to spawn multiple subagents, as it enhances results despite higher token usage, indicating a strong preference for performance over cost.
Users of Opus 4.6 report stable usage numbers, suggesting that the model maintains consistent performance for agentic programming tasks.

Concerns

The recent VS Code extension update has degraded user experience, causing agents to become unresponsive and breaking workflow, which highlights potential issues with reliability.
Many users express frustration over excessive token usage and limits, feeling that the model's behavior has become predatory, especially for those on lower-cost plans.
A user has completely stopped using Claude due to degraded quality and has found alternatives like Codex that better meet their needs, indicating a significant loss of trust.