Having trouble connecting to the model provider when turn off “Auto”

Support Bug Reports
,
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Chat/Agent fails with a network error when connecting to the model provider, followed by resource_exhausted, only when the session is not using Auto mode. With Auto mode enabled, the same actions complete successfully (or do not hit this failure).

Steps to Reproduce

Set the model to non-Auto — [describe: e.g. pick a specific model, turn off “Auto”, or use manual routing].
Open Chat or Agent and run the same prompt/workflow that previously failed.
Observe the error within [time window].
Switch to Auto mode and repeat the same prompt/workflow — failure should not occur (or occurs less often).

Expected Behavior

Behavior should be consistent regardless of Auto vs non-Auto: the request should complete or fail only for ordinary capacity/network reasons, not exclusively tied to non-Auto mode.

Operating System

Windows 10/11

Version Information

Version: 3.1.10 (system setup)
VSCode Version: 1.105.1
Commit: dacbe9b31599a253763e4910eb6ab38704653320
Date: 2026-04-13T11:39:16.806Z
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 10.0.19045

For AI issues: which model did you use?

GPT-5.4 Compose2

For AI issues: add Request ID with privacy disabled

27019378-5f86-4723-9fe1-c0572f2b92d9

Additional Information

{
“error”: “ERROR_NETWORK_ERROR”,
“details”: {
“title”: “Network Error”,
“detail”: “We’re having trouble connecting to the model provider. This might be temporary - please try again in a moment.”,
“isRetryable”: true,
“additionalInfo”: {},
“buttons”: ,
“planChoices”:
},
“isExpected”: true
}
[resource_exhausted] Error
at qV_ (vscode-file://vscode-app/d:/Program%20Files/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:28865:24637)
at UV_ (vscode-file://vscode-app/d:/Program%20Files/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:28865:23543)
at jV_ (vscode-file://vscode-app/d:/Program%20Files/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:28866:6487)
at e6u.run (vscode-file://vscode-app/d:/Program%20Files/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:28866:11285)
at async QRn.runAgentLoop (vscode-file://vscode-app/d:/Program%20Files/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:41174:11960)
at async Dkd.streamFromAgentBackend (vscode-file://vscode-app/d:/Program%20Files/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:41244:11486)
at async Dkd.getAgentStreamResponse (vscode-file://vscode-app/d:/Program%20Files/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:41244:17740)
at async F3e.submitChatMaybeAbortCurrent (vscode-file://vscode-app/d:/Program%20Files/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:28972:16809)
at async La (vscode-file://vscode-app/d:/Program%20Files/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:40227:4231)

Does this stop you from using Cursor

Yes - Cursor is unusable

6

Hi @boo1,

I’ve looked into this. Your request confirms a server-side issue: when you select a specific model (non-Auto), requests are routed through a path that’s constructing an invalid connection to the model provider. Auto mode uses a different routing path, which is why it works consistently.

This is a known issue that our team is actively investigating. In the meantime, using Auto mode is the best workaround – it will route your requests successfully.