Galactica-6.7B: benchmark results
Meta AI Papers with Code Galactica 6.7B is a scientific decoder-only language model trained on a 106B-token scientific corpus, including papers, textbooks and structured scientific data. This is the original base release under CC BY-NC 4.0, separate from OPT and instruction-tuned variants. Provider: Other. Access: Open.
Aggregate Rating 1223 ± 35, rank #9459 of 10916 rated models, from 6 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Portuguese LLM - FaQuAD NLI | 43.97 | Macro F1 (%) | 17.1 |
| Open Portuguese LLM - OAB Exams | 25.19 | Accuracy (%) | 15.6 |
| Open Portuguese LLM - ASSIN2 RTE | 47.66 | Macro F1 (%) | 15.1 |
| Open Portuguese LLM - BLUEX | 21.56 | Accuracy (%) | 12.5 |
| Open Portuguese LLM - ENEM | 19.8 | Accuracy (%) | 11.6 |
| SciKnowEval | 2 | Inverse Overall Rank | 4 |
Interactive version: theaggregate.ai/model?slug=galactica-6-7b · How It Works · Data refreshed daily, snapshot 2026-11-08.