Masked Language Modeling
Benchmarks
Dataset NameSOTA methodMetricTrendResultsLast Updated
3.828PPLX
35
Feb 26, 2026
3.8U-PPL
20
Feb 26, 2026
8.57PPL (U)
20
Feb 26, 2026
74.76MLM Accuracy
12
Feb 26, 2026
1.274BPC
8
Feb 26, 2026
69.57MLM Avg (%)
7
Feb 26, 2026
2.89Perplexity
7
Feb 26, 2026
2.587Loss
6
May 8, 2026
6.619Training Loss
6
Apr 24, 2026
6.676Validation Loss
6
Apr 24, 2026
0.002vNMSE
6
Feb 26, 2026
5.813Perplexity
6
Feb 26, 2026
3.499Perplexity
6
Feb 26, 2026
8.906Perplexity
6
Feb 26, 2026
3.5PPL
5
Feb 26, 2026
78Wall-clock Time (17.5 Target PPL)
4
Mar 19, 2026
42.7MRR
4
Feb 26, 2026
1.787Log Perplexity
4
Feb 26, 2026
52.6MRR
3
Feb 26, 2026
1.12BPC
3
Feb 26, 2026
101.91Perplexity
2
Feb 26, 2026
58.43Perplexity
2
Feb 26, 2026
6.941Validation Loss
1
Apr 24, 2026