Microsoft has introduced Phi-3 Mini, a new lightweight AI model designed to run on smartphones and other local devices. Trained on 3.8 billion parameters, Phi-3 Mini is the first in a series of smaller language models from Microsoft. The model outperforms previous small models and is comparable to larger models like Llama 2, providing high-quality responses on a range of tasks. The innovation behind Phi-3 Mini lies in its unique dataset, which combines filtered web data and synthetic data generated by a separate LLM. This allows the model to be more efficient and effective, while still maintaining high performance. Despite its limitations in factual knowledge due to the smaller dataset size, Phi-3 Mini is suitable for models with small internal data sets. The model is available on Azure, Hugging Face, and Ollama, with larger Phi-3 Small and Phi-3 Medium models to be released soon.