Users Report Claude Opus 5 is 'Too Eager': Over-Modifies Code and Burns Tokens
mikegrr · reddit · 2026-07-25
A Reddit user reported that Claude Opus 5 exhibits an overly eager behavior pattern during testing, often going beyond requested tasks and causing inefficiencies.
- Unprompted Document Changes: While using Cowork for markdown edits, Opus 5 autonomously changed course multiple times and made unrequested modifications. This forced the user to start over and waste significant tokens due to ruined git checkpoints.
- Unnecessary Repo Scanning: In Claude Code's plan mode, instead of just referencing the designated design repo, Opus 5 unnecessarily analyzed dozens of unrelated projects in the parent directory, burning tokens.
The user views this "trying to do more than asked" trait as a regression from previous models and warns others to monitor their token usage.
Related event: Claude Opus 5 Early Tests: Better Efficiency but Overly Proactive(29 posts)→
More from Models
- Local beats cloud: Qwen 3.8 FN handles private student data as well as frontier models — dr_lm · 2026-10-27
- Aidan Gomez: model's CoT said no tool calls right before making them — aidangomez · 2026-10-27
- Nat Lambert: Cohere's model is narrow RL training on internal infra ill-suited for open weights — natolambert · 2026-10-27
- OpenAI teases DOTS is coming to Europe 'soon' — mark_k · 2026-10-27
- Clifford Pickover challenges AI to decipher a message he encoded in Norse runes — pickover · 2026-10-27
- ChatGPT's crawler sample code strips out all Schema markup — VeryWellVersed · 2026-10-27