IBM: Granite 4.0 Micro

IBM's 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 evaluated benchmarks. Created on October 19, 2025, and fine-tuned for long context tool calling, its performance profile reveals a mixed but generally strong showing in certain areas. The model exhibits a notable strength in Ethics, achieving 97.8% accuracy, and performs well in General Knowledge with 91.4% accuracy, though these scores place it in the 33rd and 28th percentiles respectively. Its ability to handle long contexts is suggested by its fine-tuning, but specific long-context benchmarks are not provided. A significant weakness is observed in Mathematics, where it scored 0.0% accuracy, indicating a complete inability to solve the problems presented. Hallucinations are a concern, with only 57.5% accuracy in acknowledging uncertainty, placing it in the 12th percentile. Performance in Email Classification (88.5%), Instruction Following (46.2%), Reasoning (44.7%), and Coding (22.7%) are moderate to low, suggesting areas for improvement in complex task execution and specialized domain knowledge.