dolphin-2.9.1-mixtral-1x22B: benchmark results
Cognitive Computations' Dolphin 2.9.1 Mixtral 1x22B, a single dense expert extracted via SLERP from the 8x22B Mixtral mixture, fully fine-tuned from Dolphin-2.9-Mixtral-8x22b with a 64K base context. Provider: Cognitive Computations. Released 2024-05-22. Access: Open.
Aggregate Rating 1296 ± 16, rank #1492 of 1656 rated models, from 29 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - GSM8K | 67.48 | Accuracy (%) (5-shot) | 83.7 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 54.75 | MC2 (%) (0-shot) | 63.4 |
| Open LLM Leaderboard v1 - MMLU | 59.14 | Accuracy (%) (5-shot) | 44.7 |
| Open Portuguese LLM - FaQuAD NLI | 59.48 | Macro F1 (%) | 43.5 |
| Open Chinese LLM - TruthfulQA MC | 52.87 | Accuracy (%) | 42.8 |
| Open LLM Leaderboard v1 - ARC Challenge | 58.53 | Normalized accuracy (%) (25-shot) | 40.1 |
| Open Portuguese LLM - ASSIN2 RTE | 86.89 | Macro F1 (%) | 39 |
| Open Portuguese LLM - OAB Exams | 37.31 | Accuracy (%) | 35.4 |
| Open Portuguese LLM - ENEM | 50.87 | Accuracy (%) | 34.9 |
| Open LLM Leaderboard v1 - WinoGrande | 73.48 | Accuracy (%) (5-shot) | 32.3 |
| Open LLM Leaderboard v1 - HellaSwag | 77.78 | Normalized accuracy (%) (10-shot) | 31.8 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 6.67 | Exact match, flexible extract (%) | 28.5 |
Interactive version: theaggregate.ai/model?slug=dolphin-2-9-1-mixtral-1x22b · How It Works · Data refreshed daily, snapshot 2026-10-20.