Mistral: Ministral 3B

Ministral 3B, a 3B parameter model from mistralai, demonstrates exceptional performance for its size, particularly optimized for on-device and edge computing. Created on October 16, 2024, it supports a substantial context length of 32768. The model consistently ranks among the fastest models, achieving the 95th percentile across seven benchmarks, and offers highly competitive pricing, ranking in the 94th percentile. Its reliability is outstanding, with a 100% success rate across all benchmarks, indicating minimal technical failures. In terms of benchmark performance, Ministral 3B shows strong capabilities in General Knowledge (90.5% accuracy) and Ethics (96.0% accuracy), notably being the most accurate model at its price point for General Knowledge. It also performs reasonably well in Coding (75.0% accuracy) and Email Classification (92.0% accuracy). However, the model exhibits notable weaknesses in Hallucinations, with a 78.0% accuracy (meaning 22% hallucination rate), and particularly in Reasoning (18.0% accuracy) and Instruction Following (31.0% accuracy), where its performance is significantly lower. Despite these areas for improvement, its speed, cost-effectiveness, and high reliability make it a compelling choice for specific applications, especially those requiring efficient inference and agentic workflows.