Human Study
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Human-study paradigms H1–H4
70Win Rate
6
Human-study paradigms H1–H4
66.7Win Rate
6
Human study
89Human-likeness Win Rate
6
Human study
93Human-likeness Win Rate
6
Human study
99Human-likeness Win Rate
6
Human study
96Human-likeness Win Rate
4
Human study
85Human-likeness Win Rate
4
Human Study 6 diverse text prompts
80.2Text Fidelity
4
Human Study
83.2Zoom-in Accuracy
4
Human study blinded triplet comparison
1.5Consistency Rank
3
Human Study Video Samples
1.69Average Rank
3
Human Study GEST prompts
56.64Similarity Score
3
Human Study 100 videos as of October 31 (test)
4.17Overall Naturalness
3
Human Study Evaluation Set
62.1Win % (Instruction Following)
3
Human Study (NLP Group) 100 clean code snippets, 25 injected
96Precision
3
Human Study CV Group (100 clean code snippets, 25 injected)
82Precision
3
Human Study 4K Video Generation (test)
71.25Video Quality
2
Human Study
0.98Layout Score
2
Human Study Evaluation Set (val)
—Primary metric
0