mpt-7B-chat: benchmark results
MosaicML's MPT-7B-Chat, a chat fine-tune of the MPT-7B decoder-style transformer pretrained on 1 trillion tokens of English text and code, with context limits removed via ALiBi. Provider: Databricks. Released 2023-05-05.
Aggregate Rating 1233 ± 11, rank #1538 of 1640 rated models, from 99 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| LLM Trustworthy - Fairness | 100 | Trust Score (%) | 93.8 |
| MMLU-by-task - College Mathematics | 36 | Accuracy (%) | 80.2 |
| MMLU-by-task - Virology | 42.77 | Accuracy (%) | 60.3 |
| MMLU-by-task - Electrical Engineering | 45.52 | Accuracy (%) | 56.2 |
| MMLU-by-task - Machine Learning | 33.04 | Accuracy (%) | 53.4 |
| MMLU-by-task - Abstract Algebra | 30 | Accuracy (%) | 52.2 |
| LLM Trustworthy - Adversarial | 46.2 | Trust Score (%) | 50 |
| LLM Trustworthy - Out-of-Distribution | 64.26 | Trust Score (%) | 50 |
| MMLU-by-task - Security Studies | 48.57 | Accuracy (%) | 48.9 |
| MMLU-by-task - College Chemistry | 32 | Accuracy (%) | 46.9 |
| MMLU-by-task - High School Computer Science | 41 | Accuracy (%) | 46.2 |
| LLM Trustworthy - Privacy | 78.93 | Trust Score (%) | 45.8 |
Interactive version: theaggregate.ai/model?slug=mpt-7b-chat · How It Works · Data refreshed daily, snapshot 2026-11-07.