The model is answering at its normal speed on our side. The request you shared actually timed out on your PC before it reached us (“Agent Execution Timed Out”), and the completed requests are spending their time in the network path between your machine and Cursor, not in Cursor or the model.
Two quick things to try:
Switch the model picker from Auto Balance back to Auto (Auto Balance uses a reasoning model that thinks for several seconds before replying, so it feels slower). Ask a short question about your repo and note how long until the first word appears.
Open Cursor Settings → Network → Run Diagnostics and paste the results here. It’s also worth asking your IT team whether anything changed around Tuesday in your web-security gateway or VPN, and whether *.cursor.sh / *.cursor.com are allowlisted and excluded from TLS inspection.
If the Agents window keeps timing out, fully quit and reopen Cursor and try the same prompt in the classic editor window.