Phi-4 is a 14-billion-parameter open language model from Microsoft Research that reaches strong reasoning scores for its size by training heavily on high-quality synthetic data.
In plain words
Most models learn mainly from web content and code. Phi-4 was trained with lots of synthetic, textbook-like data made for teaching maths, coding and reasoning, plus filtered public websites and academic books. It is a dense transformer with a 16,000-token context window, released in December 2024 under the MIT licence. It is mainly trained in English and focuses on Python for code.
A quick example
Imagine a student using Phi-4 as a study helper, asking it to explain each step of a maths problem. Because it is a 14-billion-parameter model, it is far smaller than the largest frontier models.
Where you'll meet it
- Reasoning, maths and coding help in English.
- A building block for research and generative AI features.
Related
Sources
- PaperPhi-4 Technical Report, Abdin et al. (arXiv) · read 29 Sept 2026
- Docsmicrosoft/phi-4, Microsoft (Hugging Face) · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.