List your company

ResearchFormerly WizWand

Model Attribution

Benchmarks

DatasetSOTA methodMetricTrendResultsLast updated

8-model closed-set attribution task Ntest=500 per model (test)

100Top-1 Accuracy

27

Jun 29, 2026

Compositional Prompts (test)

Flux-Turbo-Alpha

0.977Flux-Base Attribution Score

13

Apr 21, 2026

GM-FFHQ to GM-CelebA-HQ

56.9Accuracy

12

Feb 26, 2026

GM-CelebA-HQ to GM-FFHQ

66.3Accuracy

12

Feb 26, 2026

GM-CelebA to CIFAR10

63Accuracy

12

Feb 26, 2026

GM-CIFAR10 to GM-CelebA

67.7Accuracy

12

Feb 26, 2026

66.1Accuracy

12

Feb 26, 2026

64.8Accuracy

12

Feb 26, 2026

GM-CelebA (test)

74.7Accuracy

12

Feb 26, 2026

GM-CIFAR10 (test)

72.018Accuracy

12

Feb 26, 2026

98.04Accuracy

7

Apr 15, 2026

0.43DiffID

6

Jun 23, 2026

Multi-source Attribution (mixed)

100LlamaGen Score

4

Jun 30, 2026

Stable Diffusion direct 2.1 (train)

100TPR@5%FPR (NM/G)

4

Jun 24, 2026

Stable Diffusion direct 1.4 (train)

99.9TPR@5%FPR (NM/G)

4

Jun 24, 2026

Section 5.2 corpus 6 AIs x 3 domains (sample-level leave-one-out)

Reference-anchored (patch_mean, ssim)

51.36-way LOO Accuracy

2

Jun 16, 2026