Microsoft has announced the newest addition to its Phi family of generative AI models: Phi-4. It's in limited research preview.

The new small language model prioritizes efficiency and reasoning capabilities, leveraging Microsoft's focus on responsible AI development.

The new model has 14 billion parameters with improved performance in mathematical reasoning. Microsoft attributes Phi-4's improved performance to the use of high-quality synthetic datasets alongside human-generated content,

Phi-4 surpasses both similar-sized and larger models in mathematical problem-solving, including high-profile competitors including Gemini Pro 1.5

The algorithm’s performance is notable mainly because of the way it was built: Microsoft trained Phi-4 mostly on synthetic, or machine-generated, data rather than web content as is the usual practice.

Microsoft has launched Phi-4, a small language model with 14 billion parameters, excelling in complex reasoning and math tasks. It outperforms larger models like Gemini Pro 1.5, thanks to high-quality datasets and innovative techniques,