Opus 4.6 stuck 'exploring' and never writes code

Support Bug Reports
,
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

I was using Opus 4.6 before this with no issues just a week ago. All of a sudden, any prompt I write goes through the normal process of reading files, but then starts ‘Thinking…’ which then becomes ‘Exploring…’ and never escapes. It will say it’s implementing or changing things as part of the reasoning, but won’t actually begin editing files. I’ve run Diagnostic check with no issues, tried changing from HTTP 2 as recommended in some threads, disabled extensions, restarted the application, restarted my laptop, etc. to no avail. Other thinking models like Codex 5.3 still work.

Steps to Reproduce

Open any project, and send any prompt to Opus 4.6 or Sonnet 4.6.

Screenshots / Screen Recordings

Screenshot 2026-04-08 144216.png
Screenshot 2026-04-08 144216.png2064×1676 464 KB

Operating System

Windows 10/11

Version Information

Version: 3.1.0-pre.21.patch.0 (system setup)
VSCode Version: 1.105.1
Commit: 6baf31caac121b3c954f61c3303b7890619f4be0
Date: 2026-04-08T05:50:06.543Z
Layout: editor
Build Type: Stable
Release Track: Default
Electron: 39.8.1
Chromium: 142.0.7444.265
Node.js: 22.22.1
V8: 14.2.231.22-electron.0
OS: Windows_NT x64 10.0.26200

For AI issues: add Request ID with privacy disabled

Request ID: a5e16b76-95ff-4c98-971c-c582e55c81c2

Does this stop you from using Cursor

Yes - Cursor is unusable

5

Hey, thanks for the report and the Request ID.

I can see in the screenshot that the model really gets stuck on Exploring… and doesn’t move on to edits. I checked the logs for your Request ID. On the backend side, all requests finish successfully, and the model is actively working (reading files, running grep, launching a subagent), but it never switches to the editing phase. So this looks like a model behavior issue, not an infrastructure issue.

Looks like you already tried the main stuff (diagnostics, disabling HTTP/2, extensions, restart). Since Codex 5.3 works fine, the issue is specific to the Anthropic models.

A few things you can try:

  1. Try a more specific prompt that clearly says what exactly should be edited. Sometimes vague tasks make the model keep exploring context forever.
  2. Switch temporarily to Composer 2 or Codex as a workaround while the Anthropic models stabilize.
  3. You’re on a nightly build 3.1.0-pre.21.patch.0, try switching to stable to rule out nightly-related issues.

The team is aware of this. A few users have reported similar Opus 4.6 behavior over the last couple of weeks. Your report with the Request ID helps us prioritize.

Let me know if switching to stable or another model helps.