Hey there! During launch week gpt5 and gpt5 high were awesome. In this last several weeks though the models have gotten noticeably worse at tool calling, specifically in making file edits. An edit that takes 2 minutes using sonnet 4.5 or another model, takes 5 minutes on gpt5 and normally ends with “The model made no changes to this file” several times in a row.
Steps to Reproduce
Ask gpt5 to make changes.
Expected Behavior
It is supposed to call the edit tool properly and not call the same tool over and over again,.
I am having the same issue, mines with Deep seek, claude 4, gpt5. It calls the edit tool functions but then fails to edit anything, launching consoles or doing anything other than just posting it into the chat as plane text now.
Here is a Quick Fix for GPT-5 in the Cursor that did work for me :
GPT-5 differs from Sonnet in how it discovers and utilizes available tools. Since GPT-5 doesn’t automatically recognize its operating within Cursor IDE, you need to explicitly inform it through a global or project-level rule.
Recommended Rule:
"You are currently operating inside Cursor IDE. You have access to numerous integrated tools provided by the Cursor environment, including:
MCP (Model Context Protocol) servers
Terminal command execution
Task/to-do creation and management
File system operations
And other IDE-specific capabilities
Begin each session by identifying all available tools in your current environment, then leverage them proactively throughout our interaction to provide the most effective assistance.
