The latest version 0.50.5 of Cursor seems to be a big step backwards as it runs out of context all the time. When I place the same code and same input into Claude 3.7 directly I get responses with no errors, but in cursor it states that the context has run out, and this occurs even when I am running with the Claude 3.7 model in cursor directly. It also happens in Auto. This makes the product almost unusable in many situations. Switching to VS Code with CoPilot fixes this, as does going directly to Claude 3.7 web interface.
By the way, the same input code, and input prompt which has a lot of data works just fine in Windsurf.ai