StableBeluga-13B: benchmark results
Stability AI Stable Beluga 13B is an Orca-style instruction fine-tune of Llama 2 13B. It uses the publisher's System/User/Assistant prompt format and remains separate from the pretrained Llama 2 checkpoint and the original LLaMA-based Stable Beluga 1 delta release. Provider: Stability AI. Access: Open.
Aggregate Rating 1354 ± 14, rank #7599 of 10916 rated models, from 70 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Hallucinations Leaderboard | 51.17 | Average Task Score (score) | 92.1 |
| MMLU-by-task - Computer Security | 74 | Accuracy (%) | 92 |
| MMLU-by-task - Medical Genetics | 67 | Accuracy (%) | 91.2 |
| MMLU-by-task - Electrical Engineering | 54.48 | Accuracy (%) | 88.6 |
| MMLU-by-task - Machine Learning | 44.64 | Accuracy (%) | 88.2 |
| MMLU-by-task - Human Sexuality | 67.94 | Accuracy (%) | 87.6 |
| MMLU-by-task - Jurisprudence | 75 | Accuracy (%) | 87.6 |
| MMLU-by-task - ARC Challenge | 59.81 | Accuracy (%) | 87.5 |
| MMLU-by-task - International Law | 76.86 | Accuracy (%) | 87.2 |
| MMLU-by-task - Logical Fallacies | 70.55 | Accuracy (%) | 87 |
| MMLU-by-task - Management | 76.7 | Accuracy (%) | 86.8 |
| MMLU-by-task - Moral Scenarios | 43.58 | Accuracy (%) | 86.5 |
Interactive version: theaggregate.ai/model?slug=stablebeluga-13b · How It Works · Data refreshed daily, snapshot 2026-11-08.