SWE-bench will hit 90% this year (fabraix.com)

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