[BUG] Opus 4.8 frequently enters long "thinking" on simple tasks without taking action — suspected regression

Open 💬 2 comments Opened Jun 15, 2026 by 2784260366

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?

On Opus 4.8, for simple, well-defined tasks (e.g. a small code edit), the model frequently gets
stuck in long extended-thinking phases — reading and re-deliberating for several minutes without
actually making tool calls or producing the change. It repeatedly self-verifies trivial decisions
instead of just acting.

I tested the exact same task with the same setup (same Claude Code, same memory/context, same
prompt) across different models:

  • Opus 4.6, Opus 4.7, Fable 5: work normally — they act quickly and finish the task.
  • Opus 4.8: consistently spends minutes "thinking" before doing anything, often just reads

without
acting.

This is not a reasoning-effort/configuration issue on my side — it reproduces on 4.8 regardless
of
settings, and only on 4.8. Given that the same workload runs fine on 4.6/4.7/Fable 5, this looks
like a behavioral regression specific to Opus 4.8, likely in how extended thinking is
triggered/exited for simple tasks.

Expected: For simple tasks, the model should act promptly like 4.6/4.7/Fable 5 do, instead of
entering prolonged thinking.

Repro: Give Opus 4.8 a simple, well-scoped code edit. Observe it spend several minutes in
thinking
(e.g. status line shows thinking ... · 7m+ · high effort) before acting, while other models
complete
the same task quickly.

What Should Happen?

On a trivial task, it burned through 100k+ tokens just thinking, without doing any actual work.

Error Messages/Logs

On a trivial task, it burned through 100k+ tokens just thinking, without doing any actual work.

Steps to Reproduce

On a trivial task, it burned through 100k+ tokens just thinking, without doing any actual work.

Claude Model

Opus

Is this a regression?

No, this never worked

Last Working Version

_No response_

Claude Code Version

2.1.177

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Windows Terminal

Additional Information

_No response_

View original on GitHub ↗

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