opt-2.7B: benchmark results
Meta OPT is an English-dominant decoder-only pretrained language model using causal language modeling, a GPT-2 byte-level tokenizer and 2,048-token training sequences. Its training corpus contains 180B tokens; this base release remains separate from OPT-IML instruction-tuned models. Provider: Meta. Released 2022-05-03. Access: Open.
Aggregate Rating 1128 ± 16, rank #10368 of 10916 rated models, from 85 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MMLU-by-task - High School Mathematics | 28.15 | Accuracy (%) | 61.8 |
| MMLU-by-task - High School Physics | 31.79 | Accuracy (%) | 61.3 |
| MMLU-by-task - High School Statistics | 40.74 | Accuracy (%) | 57.1 |
| AI Energy Score (Text Generation) | 4 | Energy Score (1-5) | 47.5 |
| MMLU-by-task - Professional Accounting | 29.79 | Accuracy (%) | 38.3 |
| MMLU-by-task - Global Facts | 31 | Accuracy (%) | 36.8 |
| MMLU-by-task - High School Macroeconomics | 34.87 | Accuracy (%) | 34.5 |
| MMLU-by-task - Medical Genetics | 35 | Accuracy (%) | 34 |
| MMLU-by-task - High School Computer Science | 33 | Accuracy (%) | 33.8 |
| MMLU-by-task - Management | 36.89 | Accuracy (%) | 33.6 |
| MMLU-by-task - Philosophy | 31.51 | Accuracy (%) | 32.1 |
| MMLU-by-task - Machine Learning | 28.57 | Accuracy (%) | 30.8 |
Interactive version: theaggregate.ai/model?slug=opt-2-7b · How It Works · Data refreshed daily, snapshot 2026-11-08.