Microsoft Unveils Phi-4, a Compact AI Model Focused on Math
Microsoft introduces Phi-4, a 14-billion-parameter language model that excels at math due to high-quality synthetic data. The model is currently available only for research through the Azure AI Foundry platform.
Microsoft introduces Phi-4, the latest addition to its family of small language models, featuring 14 billion parameters. The tech giant claims this new model shows significant improvements over previous versions, particularly in solving complex math problems. This performance boost comes from mixing high-quality human-generated content with advanced synthetic datasets.
Phi-4 is currently available in very limited access exclusively on the Azure AI Foundry platform. Developers and researchers can only use the model for research purposes under a strict Microsoft research license agreement. As a small language model, it competes directly with other efficient options like GPT-4o mini and Claude 3.5 Haiku, offering a faster and cheaper alternative to massive AI systems.
The release highlights a growing trend in the AI industry as labs hit a "pre-training data wall" and turn to synthetic data and post-training innovations to improve performance. Notably, Phi-4 is the first model in the Phi series to launch since Sébastien Bubeck, a key figure in the model's development and a former Microsoft AI vice president, leaves the company to join OpenAI.