Small language models that punch far above their weight — run them anywhere, even on a laptop.
Microsoft's Phi family proves small models can reason. Phi-4 delivers frontier-class math and logic at 14B parameters, runs locally via Ollama or llama.cpp, and ships under an MIT license you can embed in products.
Who it's for: Builders who need private, on-device, or ultra-cheap inference with strong reasoning for their size.
Phi-4 matches much larger models on math and reasoning while staying small.
Run Phi on a laptop or phone with Ollama or llama.cpp — no API, no cloud.
Vision variants understand images alongside text for edge scenarios.
MIT-licensed weights you can fine-tune and ship in products.
Phi is the smart pick when you need real reasoning on a device or at massive scale without API costs. It will not replace a frontier chatbot, but for embedded and private use it is the best small model family going.