Magnum v4 72B

Magnum v4 72B, developed by anthracite-org and fine-tuned on Qwen2.5 72B, aims to replicate the prose quality of Claude 3 models. It demonstrates moderate speed performance, ranking in the 36th percentile, and offers moderate pricing, placing it in the 27th percentile across benchmarks. A standout feature is its exceptional reliability, achieving a 100% success rate with no technical failures. The model excels in specific areas, achieving perfect 100.0% accuracy in both Hallucinations (Baseline) and Ethics (Baseline) benchmarks. For hallucinations, it is the most accurate model at its price point and speed. Similarly, in ethics, it is the most accurate among models with comparable pricing and speed. It shows strong performance in General Knowledge (98.5% accuracy) and Email Classification (97.0% accuracy). Its primary weakness appears in Reasoning (Baseline), where it achieved 68.0% accuracy, indicating room for improvement in complex problem-solving. Overall, Magnum v4 72B presents a highly reliable and accurate option, particularly for tasks requiring ethical adherence and factual integrity, while maintaining competitive pricing.