xglm-4.5B: benchmark results
Meta's XGLM 4.5B, a multilingual autoregressive language model trained on a balanced corpus spanning 134 languages. Provider: Meta. Released 2022-01-31. Access: Open.
Aggregate Rating 1133 ± 13, rank #1617 of 1640 rated models, from 79 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MMLU-by-task - Global Facts | 34 | Accuracy (%) | 58.4 |
| MMLU-by-task - Econometrics | 28.07 | Accuracy (%) | 47.3 |
| MMLU-by-task - College Mathematics | 29 | Accuracy (%) | 32.7 |
| MMLU-by-task - Sociology | 31.34 | Accuracy (%) | 32.1 |
| MMLU-by-task - Professional Law | 27.31 | Accuracy (%) | 30.9 |
| MMLU-by-task - Professional Accounting | 27.66 | Accuracy (%) | 30.6 |
| MMLU-by-task - Philosophy | 30.23 | Accuracy (%) | 29.9 |
| MMLU-by-task - High School Computer Science | 31 | Accuracy (%) | 29.4 |
| MMLU-by-task - High School Statistics | 28.7 | Accuracy (%) | 28 |
| MMLU-by-task - Logical Fallacies | 28.83 | Accuracy (%) | 27.8 |
| MMLU-by-task - College Medicine | 27.17 | Accuracy (%) | 27.6 |
| MMLU-by-task - Anatomy | 30.37 | Accuracy (%) | 26.9 |
Interactive version: theaggregate.ai/model?slug=xglm-4-5b · How It Works · Data refreshed daily, snapshot 2026-11-07.