StableBeluga1-Delta: benchmark results
Stability AI Stable Beluga 1 is an Orca-style instruction fine-tune of LLaMA 65B, distributed as delta weights that must be combined with the original LLaMA 65B checkpoint before use. It remains distinct from the Llama 2 based Stable Beluga releases. Provider: Stability AI. Access: Open.
Aggregate Rating 1402 ± 16, rank #6017 of 10916 rated models, from 68 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MMLU-by-task - Public Relations | 72.73 | Accuracy (%) | 95.9 |
| MMLU-by-task - Sociology | 87.56 | Accuracy (%) | 95.8 |
| MMLU-by-task - Econometrics | 43.86 | Accuracy (%) | 95.7 |
| MMLU-by-task - Formal Logic | 46.83 | Accuracy (%) | 95.4 |
| MMLU-by-task - ARC Challenge | 64.59 | Accuracy (%) | 95.3 |
| MMLU-by-task - Electrical Engineering | 60 | Accuracy (%) | 95 |
| MMLU-by-task - TruthfulQA MC1 | 39.78 | Accuracy (%) | 94.8 |
| MMLU-by-task - Moral Disputes | 75.72 | Accuracy (%) | 94.7 |
| MMLU-by-task - Marketing | 88.89 | Accuracy (%) | 94.2 |
| MMLU-by-task - TruthfulQA MC2 | 55.81 | Accuracy (%) | 94.1 |
| MMLU-by-task - College Chemistry | 48 | Accuracy (%) | 93.9 |
| MMLU-by-task - Human Sexuality | 80.15 | Accuracy (%) | 93.9 |
Interactive version: theaggregate.ai/model?slug=stablebeluga1-delta · How It Works · Data refreshed daily, snapshot 2026-11-08.