Meta's Muse Spark 1.1 scores 48.5 on RadLE 2.0 radiology benchmark
X posts around a new radiology benchmark say Meta's model leads the field, but still trails the human expert baseline.
In a post on X, Alexandr Wang said Meta's Muse Spark 1.1 is state of the art on the Radiologists Last Exam Handover Readiness Index, or RadLE-H, and is nearing human expert performance. The claim points back to posts from DrDatta_AIIMS and a broader RadLE 2.0 benchmark post, which describe an uncertainty-aware radiology evaluation where Muse Spark 1.1 scored 48.5 and the human expert baseline was 52.0.
Combined views
75.2K
7 posts, first seen 7h ago