gpt-neo-2.7B: benchmark results
EleutherAI's GPT-Neo 2.7B, an open replication of the GPT-3 architecture trained on the Pile for 420 billion tokens, the largest model in EleutherAI's original GPT-Neo release. Provider: EleutherAI. Released 2021-03-21. Access: Open.
Aggregate Rating 1128 ± 15, rank #10372 of 10916 rated models, from 96 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MMLU-by-task - Moral Scenarios | 26.48 | Accuracy (%) | 54.4 |
| MMLU-by-task - High School Statistics | 39.35 | Accuracy (%) | 52.9 |
| MMLU-by-task - Elementary Mathematics | 29.1 | Accuracy (%) | 52.4 |
| MMLU-by-task - Professional Medicine | 43.01 | Accuracy (%) | 37.9 |
| MMLU-by-task - High School Computer Science | 35 | Accuracy (%) | 37.2 |
| MMLU-by-task - College Physics | 22.55 | Accuracy (%) | 34.3 |
| MMLU-by-task - Abstract Algebra | 27 | Accuracy (%) | 34 |
| MMLU-by-task - High School Macroeconomics | 34.36 | Accuracy (%) | 33.6 |
| MMLU-by-task - Nutrition | 31.7 | Accuracy (%) | 33 |
| MMLU-by-task - Econometrics | 26.32 | Accuracy (%) | 32.9 |
| MMLU-by-task - Philosophy | 31.51 | Accuracy (%) | 32.1 |
| MMLU-by-task - Prehistory | 30.86 | Accuracy (%) | 31.9 |
Interactive version: theaggregate.ai/model?slug=gpt-neo-2-7b · How It Works · Data refreshed daily, snapshot 2026-11-08.