It is unusable and performs very poorly; it frequently times out and fails to work

As @Tom_Coustols mentioned, giving the run a fresh start (stop it and re-send, or start a new chat) is the right first move. A bit more on what you’re seeing:

“Taking longer than expected” isn’t an error. It’s a client-side indicator that appears when the response hasn’t started streaming within a short threshold, and it almost always points to a network or local-environment delay rather than the model itself.

Worth trying, in order:

  1. Run Cursor Settings > Network > Run Diagnostics - this shows exactly where the connection is getting stuck.
  2. In the same Network settings, switch the HTTP protocol to HTTP/1.1. This clears up a lot of slow or hanging responses, especially on restrictive networks or proxies. If you’re behind a proxy, test without it too.
  3. If one request keeps stalling, grab a Request ID (three-dot menu in the chat) with Privacy Mode off and post it here so we can trace it.

On the AUTO quality side: Auto picks whichever model is fastest and available at the moment, so results can vary on harder tasks. For complex work, try selecting a specific model from the picker instead of Auto.

Same indicator with the full set of steps here: Taking longer than expected… Cursor v. 3.7.19.