Aggregate
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Aggregate
63.97WQ
3
Aggregate WebVoyager, ScreenAgent, and In-house suite
74.8Overall Success Rate
3
Aggregate MedQA MedMCQA PubMedQA MMLU
86.7Accuracy
3
Aggregate All DGPs n=100,000 oracle, n=3000 train (holdout)
5.01MAE
2
Aggregate (AIME24, AIME25, ZebraLogic, AutoLogi, GPQA-D)
-1.6ASG
1
Aggregate Overall experimental suite (test)
—Primary metric
0