29.1%
AutomationBench
27.2%
53.3%
Humanity's Last Exam
49.8%
57.2%
PerceptionBench
47.2%
41.4%
PostTrain Bench
34.1%
76.8%
Program Bench
71.9%
80.3%
SWE-bench Pro
69.2%
34.7%
SpreadsheetBench 2
31.55%
84.6%
Terminal-Bench 2.1
84.6%
77.9%
Toolathlon-Verified
76.2%