mpt-7B: benchmark results
MosaicML's MPT-7B, a decoder-style transformer pretrained from scratch on 1 trillion tokens of English text and code, with ALiBi replacing positional embeddings to remove fixed context limits. Provider: Databricks. Released 2023-05-05.
Aggregate Rating 1158 ± 18, rank #1619 of 1656 rated models, from 52 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| HellaSwag | 76.4 | Accuracy (%) | 45.4 |
| PIQA | 80.6 | Accuracy (%) | 45 |
| OpenBookQA | 51.4 | Accuracy (%) | 35.7 |
| BoolQ | 75 | Accuracy (%) | 32.5 |
| WinoGrande | 68.6 | Accuracy (%) | 32.5 |
| Open Portuguese LLM - ASSIN2 RTE | 76.49 | Macro F1 (%) | 28.1 |
| ARC Challenge (AI2) | 42.6 | Accuracy (%) | 27.6 |
| Open Portuguese LLM - FaQuAD NLI | 44.66 | Macro F1 (%) | 25.9 |
| Open LLM Leaderboard - MuSR | 36.72 | Score | 22.3 |
| BizBench - FinKnow | 39.7 | Accuracy (%; 877 multiple-choice business and finance knowle | 20 |
| BizBench - FormulaEval | 48 | Accuracy (%; 50 financial-formula functions to complete from | 20 |
| BenCzechMark - NER | 16.23 | Duel Win Score (%) | 18.2 |
Interactive version: theaggregate.ai/model?slug=mpt-7b · How It Works · Data refreshed daily, snapshot 2026-10-20.