Gemini 3.0 is unusable - Unrecoverable agent model looping detected, no read_file tool available

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Gemini 3.0 doesn’t work at all.

100% of my test prompts have resulted in an Unrecoverable agent model looping detected. Many complain that there’s no read_file tool.

I appreciate that the provider errors are largely out of Cursor’s control, but if you fail to reach the model provider, just retry with some exponential backoff? Why does the ‘try again’ button wipe the entire chat thread?

When it does create a file, it outputs its thinking into the file instead of actually doing an implementation:

    const flushTokens = () => {
      if (currentTokens.length === 0) return

      // Chunk into granularity
      for (let i = 0; i < currentTokens.length; i += granularity) {
        const chunk = currentTokens.slice(i, i + granularity)
        // For text reconstruction, we don't easily have the exact slice of text unless we track it per token.
        // `tokeniseString` returns objects with text.
        // We need `tokeniseString` result.

        // Re-tokenising the whole text part:
        // We should have tokenised the text parts individually.
      }
    }

    // Re-do loop with tokenisation
    for (const part of parts) {
      if (part.type === 'arbitrary') {
        // Flush any pending partial block?
        // "Any 'cache blocks' on the ends that are partially filled should be a monochrome colour."
        // If we have [A] (granularity 2) and then Arbitrary, [A] is a partial block.
        if (currentTokens.length > 0) {
          // Handle pending tokens
          // We need to handle chunking logic properly below.
        }
      }
    }

Screenshot 2025-11-19 at 11.28.05 am

Steps to Reproduce

I mean, literally every prompt has failed, so I won’t bother producing a reproducible test case.

Expected Behavior

A working model, and not to be charged for clearly error-ridden output.

Screenshots / Screen Recordings

Operating System

MacOS

Current Cursor Version (Menu → About Cursor → Copy)

Version: 2.0.77
VSCode Version: 1.99.3
Commit: ba90f2f88e4911312761abab9492c42442117cf0
Date: 2025-11-13T23:10:43.113Z
Electron: 37.7.0
Chromium: 138.0.7204.251
Node.js: 22.20.0
V8: 13.8.258.32-electron.0
OS: Darwin arm64 24.6.0

For AI issues: which model did you use?

Gemini 3.0

Additional Information

I’ve tried both the normal ‘Agent’ mode and a custom mode.

Does this stop you from using Cursor

Yes - Cursor is unusable

Hey, thanks for the report. Gemini 3.0 Pro launched yesterday and there were some initial provider issues that the team has already fixed. Several integration issues are still being worked on.

Please try the following and let me know if the issue persists:

  • Create a new Agent chat (don’t switch models mid-conversation) - this has helped some users
  • Test with a simple prompt to check if the looping still happens
  • Share Request IDs from failed attempts: Chat (top-right menu) → Copy Request ID

If the issue continues, please send:

  • 1–2 specific prompts that trigger the error
  • Request IDs from the failed attempts

This will help the engineers investigate the tool availability and looping issues.

1 Like

This topic was automatically closed 22 days after the last reply. New replies are no longer allowed.