The video compares the work of Cursor (proprietary fork of VS Code) and Cline (open-source VS Code extension) - they edit a React project of 240k code tokens.
Here both use claude-3.5-sonnet.
Cursor also uses some embedding model (either with OpenAI’s embedding API or by a custom embedding model ) and a cloud vector database to vectorize code chunks for semantic search. [Explanation with pictures ]
The first simple task was completed by both in 1 minute, but Cline returned broken code, on task 3 it "looped".
That is, Cursor Composer won 3 out of 3.