Phi is Microsoft's family of small language models, designed to give strong results while being cheap and light enough to run on phones, laptops and edge devices.
In plain words
Most headline models are huge. Phi goes the other way: Microsoft trains small models on heavily filtered web data plus synthetic data, and says Phi-3-mini rivals much bigger models such as GPT-3.5 on benchmarks. Phi-3-mini, from April 2024, has 3.8 billion parameters and is small enough for a phone. Later Phi models include vision and audio versions, and Phi-4 with 14 billion parameters.
A quick example
Imagine a field technician's tablet with no reliable signal. A small Phi model running on the device could still answer questions about a repair manual without calling a cloud server.
Where you'll meet it
- On-device and offline assistants.
- Low-cost, low-latency apps and research on small models.
Related
Sources
- DocsPhi open models, Microsoft Azure · read 29 Sept 2026
- PaperPhi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone, Abdin et al. (arXiv) · read 29 Sept 2026
- PaperPhi-4 Technical Report, Abdin et al. (arXiv) · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.