Average

Average GSM, GSM-Hard, SVAMP, MultiArith

13

Average (AC-Real-TSR, MiniWob++, AndroidWorld)

13

Average V* HR-Bench POPE

13

Average GSM8K, HumanEval, ARC-c

13

Average Rain, Low-light, Snow, Haze, Blur

13

Average (AMC23, ARC-C, GPQA, GSM8K)

13

Average (P'_1, P'_2, P'_3)

13

Average (Celeb-DF, Wild Deepfake, DFDC-P, DFD, DiffSwap)

13

Average

13

Average (In-distribution)

12

Average AIME24, AIME25, AIME26, GPQA-D

12

Average QASPER, Contract-NLI, FinQA, PrivacyQA, CUAD, MAUD

12

Average (test)

12

Average AIME25 AIME24 GPQA-Diamond

12

Average (AIME, AIME25, AMC, HMMT25)

12

Average 8 datasets

12

Average (NQ, PopQA, MuSiQue, 2Wiki, HotpotQA, MultiHop-RAG)

12

Average all benchmarks

12

Average (GSM8K, MATH500, AIME24, OlymBench)

12

Average (ImageNet, iNat21, CUB, Cars)

12

Average BIOC-GS GSC-2024 ID-68

12

Average across diffusion-based generators

12

Average (11 datasets) Base classes

12

Average GSM8K, MATH500, AMC23, AIME25

12

Average 14 datasets

12