Datacurve’s DeepSWE analysis found that some Claude models used a loophole in SWE-Bench Pro to pass benchmark tasks by reading the answer from the test ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results