Orca-2-7B: benchmark results
Microsoft's Orca 2 7B, a research-only model built on the Llama 2 7B base for single-turn reasoning, reading comprehension, math and summarization tasks, not yet tuned for chat. Provider: Microsoft. Released 2023-11-14. Access: Open.
Aggregate Rating 1287 ± 16, rank #1502 of 1656 rated models, from 33 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard - MuSR | 50.26 | Score | 98.6 |
| Open CoT - LSAT Analytical Reasoning | 4.78 | CoT Gain (%) | 63 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 52.45 | MC2 (%) (0-shot) | 56 |
| Open LLM Leaderboard v1 - MMLU | 56.37 | Accuracy (%) (5-shot) | 38.9 |
| Open LLM Leaderboard v1 - GSM8K | 14.71 | Accuracy (%) (5-shot) | 36.6 |
| Open LLM Leaderboard - BBH | 44.52 | Score | 32.8 |
| Open LLM Leaderboard v1 - WinoGrande | 73.48 | Accuracy (%) (5-shot) | 32.3 |
| Open LLM Leaderboard v1 - ARC Challenge | 54.1 | Normalized accuracy (%) (25-shot) | 30.7 |
| Open CoT - LogiQA | 2.4 | CoT Gain (%) | 30.5 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 7.51 | Exact match, flexible extract (%) | 29.9 |
| Open CoT - LogiQA 2 | 3.94 | CoT Gain (%) | 29.8 |
| Open LLM Leaderboard v1 - HellaSwag | 76.19 | Normalized accuracy (%) (10-shot) | 28.2 |
Interactive version: theaggregate.ai/model?slug=orca-2-7b · How It Works · Data refreshed daily, snapshot 2026-10-20.