I’ve recently been starting to use Cursor Pro, with the free and fixed rate Claude Opus responses. However, Claude Opus is still quite buggy compared to the GPT-4 implementation.
- Claude Opus doesn’t support images.
- Claude Opus can’t accept code modifications from the chat interface
Otherwise, very much happy with the intuitiveness of Cursor.sh. I prefer it over other similar code assistant plugins and tools because it takes in so much context so granularly. The ability to tag (and exclude) the web, other files, and even inline quotations is what sets this apart. Awesome work!
2 Likes
Fixing these, letting us specify token limits via invoice system or our own API key, and adding Claude to the interpreter mode (or a mix of Claude, gpt-4 and other LLMs depending on how it’s working currently) would make Cursor OP.
Also the accept should be handled by gpt-4, even when using Claude, as gpt-4 is already good at applying the difference.
Yep Cursor would go to another level. I personally would love to set a ridiculously large context limit for each chat message, and pay for it myself. Cursor could even add a little spread on the Anthropic API token costs and make a bit of profit off of expanded token windows.
I also second being able to set the accept and CMD+K editing to use a different model to the chat. I find myself switching to GPT-4 for small edits since it’s basically perfect for that, but then switching to Claude Opus for larger architectural discussions.
1 Like
