IBM: Granite 4.0 Micro

IBM: Granite 4.0 Micro, a 3B parameter model from the Granite 4 family, demonstrates exceptional speed and cost efficiency. It consistently ranks among the fastest models and offers highly competitive pricing across all benchmarks. Designed for long context tool calling, its performance varies across different tasks. The model exhibits strong performance in Ethics (97.8% accuracy) and General Knowledge (91.4% accuracy), indicating a solid understanding of ethical principles and a broad factual base. However, it struggles significantly with Mathematics, achieving 0.0% accuracy, and shows limited proficiency in Coding (22.7% accuracy) and Reasoning (44.7% accuracy). Its ability to acknowledge uncertainty in the Hallucinations benchmark is moderate at 57.5% accuracy, suggesting room for improvement in identifying fictional concepts. Email Classification and Instruction Following also present areas for development, with accuracies of 88.5% and 46.2% respectively. Overall, Granite 4.0 Micro stands out for its efficiency and ethical understanding, while mathematical and complex reasoning tasks represent its primary weaknesses.