Magnum 72B

Magnum 72B, from alpindale, is a Qwen2 72B-based model specifically fine-tuned with 55 million tokens of roleplay data, aiming for Claude 3-level prose quality. The model exhibits strong reliability, successfully providing usable responses 89% of the time, indicating few technical issues. However, its speed performance is a notable weakness, ranking in the 19th percentile, suggesting longer response times. Similarly, its pricing is positioned at premium levels, ranking in the 15th percentile. In terms of benchmark performance, Magnum 72B achieved perfect accuracy (100.0%) in the Ethics (Baseline) category, making it the most accurate model at its price point and among models of comparable speed. This highlights a significant strength in ethical reasoning. Its Email Classification (Baseline) performance was solid at 96.0% accuracy, demonstrating good contextual understanding. Conversely, its General Knowledge (Baseline) accuracy was 63.5%, placing it in the 18th percentile, indicating a relative weakness in broad factual recall. Overall, Magnum 72B excels in specific, nuanced tasks like ethics, but users should be aware of its slower response times and premium cost.