I have been using agent mode and finding it helpful. Are there any published benchmarks about its performance? E.g. on SWEBench-verified or AgentBench?
Not currently, as composer and Cursor as a whole is more intended as an interactive way of coding and not a one-shot LLM that can be compared to any other AI models directly!
It will be interesting to see how Claude stacks up against itself when used within Cursor’s Agent Mode though!