We just rolled out CursorBench 4.0! It includes new tasks for how well models follow instructions, work on challenging projects over time, and is more difficult than before (so all models score lower).↗