StableBeluga2: benchmark results
Stability AI's Stable Beluga 2, a Llama-2 70B fine-tune instruction-tuned on an Orca-style dataset (July 2023). Provider: Stability AI. Released 2023-07-27. Access: Open.
Unified ELO 1493 ± 10, rank #1313 of 2928 rated models, from 84 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MMLU-by-task - High School Microeconomics | 81.09 | Accuracy (%) | 100 |
| MMLU-by-task - Clinical Knowledge | 75.85 | Accuracy (%) | 99.6 |
| MMLU-by-task - College Mathematics | 44 | Accuracy (%) | 99.5 |
| MMLU-by-task - High School Biology | 83.23 | Accuracy (%) | 99.3 |
| MMLU-by-task - Sociology | 89.55 | Accuracy (%) | 99 |
| MMLU-by-task - Econometrics | 46.49 | Accuracy (%) | 98.9 |
| MMLU-by-task - Moral Scenarios | 60.78 | Accuracy (%) | 98.7 |
| MMLU-by-task - Nutrition | 77.12 | Accuracy (%) | 98.4 |
| MMLU-by-task - Formal Logic | 50 | Accuracy (%) | 98.3 |
| MMLU-by-task - Business Ethics | 75 | Accuracy (%) | 97.9 |
| MMLU-by-task - High School European History | 84.24 | Accuracy (%) | 97.5 |
| MMLU-by-task - ARC Challenge | 65.96 | Accuracy (%) | 97.1 |
Interactive version: theaggregate.ai/model?slug=stablebeluga2 · How It Works · Data refreshed daily, snapshot 2026-09-23.