mdeberta-v3-base: benchmark results
Microsoft's multilingual DeBERTa V3 model, pretrained on CC100 for cross-lingual language understanding. Its published backbone and vocabulary-embedding parameter counts describe separate components, rather than interchangeable total sizes. Provider: Microsoft. Released 2021-12-01. Access: Open.
Aggregate Rating 1507 ± 33, rank #2240 of 10916 rated models, from 117 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EuroEval Hungarian NLU - Szeged NER | 91.21 | Micro F1 excluding MISC (%) | 99 |
| EuroEval Catalan NLU - ScaLA CA | 65.91 | Matthews correlation coefficient (%) | 98.1 |
| EuroEval Croatian NLU - WikiANN HR | 76.59 | Micro F1 excluding MISC (%) | 97.6 |
| EuroEval Spanish NLU - CoNLL ES | 81.6 | Micro F1 excluding MISC (%) | 97.6 |
| EuroEval Serbian NLU - UNER SR | 89.82 | Micro F1 excluding MISC (%) | 97.5 |
| EuroEval Lithuanian NLU - WikiANN LT | 72.43 | Micro F1 excluding MISC (%) | 97.4 |
| EuroEval Bosnian NLU - WikiANN BS | 78.51 | Micro F1 excluding MISC (%) | 97.2 |
| EuroEval Romanian NLU - RONEC | 71.57 | Micro F1 excluding MISC (%) | 96.6 |
| EuroEval Czech NLU - PONER | 63.87 | Micro F1 excluding MISC (%) | 95.6 |
| EuroEval Greek NLU - ELNER | 79.82 | Micro F1 excluding MISC (%) | 95.6 |
| EuroEval Greek NLU - ScaLA EL | 55.91 | Matthews correlation coefficient (%) | 95.1 |
| EuroEval Catalan NLU - WikiANN CA | 74.21 | Micro F1 excluding MISC (%) | 93.4 |
Interactive version: theaggregate.ai/model?slug=mdeberta-v3-base · How It Works · Data refreshed daily, snapshot 2026-11-08.