dolphin-2.2.1-mistral-7B: benchmark results
Cognitive Computations' dolphin-2.2.1-mistral-7b, an uncensored Mistral 7B fine-tune on a filtered Dolphin (Orca-style) dataset with added Samantha empathy data and Airoboros data for long multi-turn conversation. Provider: Cognitive Computations. Released 2023-10-30. Access: Open.
Aggregate Rating 1379 ± 15, rank #1262 of 1635 rated models, from 32 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - Ko-IFEval | 36.95 | Strict accuracy, prompt/instruction mean (%) | 77.1 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 29.87 | Exact match, flexible extract (%) | 64.5 |
| Open LLM Leaderboard v1 - WinoGrande | 78.61 | Accuracy (%) (5-shot) | 64.3 |
| Open LLM Leaderboard v1 - ARC Challenge | 63.23 | Normalized accuracy (%) (25-shot) | 61.9 |
| Open LLM Leaderboard v1 - HellaSwag | 83.8 | Normalized accuracy (%) (10-shot) | 61.6 |
| Open LLM Leaderboard v1 - GSM8K | 48.14 | Accuracy (%) (5-shot) | 61.4 |
| Open LLM Leaderboard v1 - MMLU | 63.16 | Accuracy (%) (5-shot) | 58.7 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 53.14 | MC2 (%) (0-shot) | 58.6 |
| Open Korean LLM - Ko-EQ-Bench | 12.34 | EQ-Bench score | 58.2 |
| Open Korean LLM - KorNAT-Helpful | 47.43 | Normalized accuracy (%) | 56.7 |
| BenchBench | 48.11 | Aggregate Score (%) | 50.7 |
| Open Korean LLM Leaderboard | 34.49 | Average Score (%) | 47.5 |
Interactive version: theaggregate.ai/model?slug=dolphin-2-2-1-mistral-7b · How It Works · Data refreshed daily, snapshot 2026-11-05.