Maybe unpopular opinion but I think at this point SWE-Bench has done its part and we need a new benchmark because Gemini being on/near the same level as Claude is obviously wrong
Not sure which version of Gemini are you using but Claude is so much better for me. Gemini is generally overeager to make a code change even when I am just asking conceptual questions, among other issues.
