benchmarksfactbearishMultiple-choice medical benchmarks are increasingly saturated, with open-ended clinical performance far from solvedMachine Learning27 Jul 2026http://arxiv.org/abs/2607.02175v1