Microsoft released Phi-3-mini, a 3.8-billion-parameter small language model, available through Azure AI Studio, Hugging Face and Ollama in 4K and 128K context variants. Microsoft claimed it matched models ten times its size, with benchmark scores close to GPT-3.5. It carried the Phi line's approach — training small models on textbook-quality data — up to a practically useful scale, and Microsoft said Phi-3-small at 7 billion parameters and Phi-3-medium at 14 billion would follow.