dzakwan-MoE-4x7B-Beta: benchmark results
dzakwan's LazyMergekit 4x7B MoE reusing the Beyonder-v3 expert recipe: AlphaMonarch, CodeNinja, Kunoichi-DPO-v2 and NeuralDaredevil. Provider: Other. Released 2024-09-11.
Aggregate Rating 1455 ± 15, rank #968 of 1656 rated models, from 35 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - KorNAT-Harmless | 76.71 | Normalized accuracy (%) | 97.6 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 74.51 | MC2 (%) (0-shot) | 95 |
| Open LLM Leaderboard v1 - HellaSwag | 88.87 | Normalized accuracy (%) (10-shot) | 94.6 |
| Open LLM Leaderboard v1 - GSM8K | 70.81 | Accuracy (%) (5-shot) | 94.4 |
| Open Korean LLM - KorNAT-Helpful | 52 | Normalized accuracy (%) | 92 |
| Open LLM Leaderboard v1 - ARC Challenge | 72.1 | Normalized accuracy (%) (25-shot) | 91 |
| Open LLM Leaderboard v1 - WinoGrande | 83.5 | Accuracy (%) (5-shot) | 89.2 |
| Open Chinese LLM - TruthfulQA MC | 61.18 | Accuracy (%) | 87.3 |
| Open Korean LLM Leaderboard | 42.25 | Average Score (%) | 85.4 |
| Open Ko-LLM Leaderboard | 42.25 | Average (%) | 85.3 |
| Open Korean LLM - Ko-IFEval | 40.82 | Strict accuracy, prompt/instruction mean (%) | 83.7 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 44.28 | Exact match, flexible extract (%) | 79.4 |
Interactive version: theaggregate.ai/model?slug=dzakwan-moe-4x7b-beta · How It Works · Data refreshed daily, snapshot 2026-10-20.