scarlett-33B: benchmark results
A conversation fine-tune by ajibawa-2023 of Llama-1 on over 10,000 conversation sets covering philosophy, advice and jokes. The card cites Samantha as inspiration; that does not establish checkpoint equivalence. Provider: ajibawa-2023. Access: Open.
Aggregate Rating 1365 ± 15, rank #7298 of 10916 rated models, from 68 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MMLU-by-task - TruthfulQA MC1 | 44.06 | Accuracy (%) | 98.2 |
| MMLU-by-task - TruthfulQA MC2 | 61.05 | Accuracy (%) | 97.7 |
| MMLU-by-task - ARC Challenge | 65.02 | Accuracy (%) | 95.8 |
| MMLU-by-task - College Mathematics | 40 | Accuracy (%) | 95.3 |
| MMLU-by-task - HellaSwag | 67.72 | Accuracy (%) | 94.7 |
| MMLU-by-task - Virology | 51.81 | Accuracy (%) | 91.4 |
| MMLU-by-task - Machine Learning | 46.43 | Accuracy (%) | 90.4 |
| MMLU-by-task - Medical Genetics | 66 | Accuracy (%) | 89.9 |
| MMLU-by-task - High School Computer Science | 64 | Accuracy (%) | 89.6 |
| MMLU-by-task - Sociology | 80.6 | Accuracy (%) | 89.5 |
| MMLU-by-task - Professional Medicine | 59.56 | Accuracy (%) | 89.4 |
| MMLU-by-task - Astronomy | 65.13 | Accuracy (%) | 89.3 |
Interactive version: theaggregate.ai/model?slug=scarlett-33b · How It Works · Data refreshed daily, snapshot 2026-11-08.