Microsoft Releases Phi-3 Lightweight AI Models, Previews Vision Variant
Microsoft makes its lightweight Phi-3 AI model family generally available for developers while previewing a new multimodal Phi-3-vision model. The release targets the growing demand for efficient, on-device artificial intelligence.
Microsoft releases its lightweight Phi-3 model family to developers, offering three sizes: Phi-3-mini, Phi-3-small, and Phi-3-medium. Developed by Microsoft Research, these compact language models provide reasoning capabilities comparable to OpenAI's GPT-3.5 at a significantly lower cost. Phi-3-mini is now integrated directly into Azure AI, making it easily accessible for cloud deployments.
This release aligns with the upcoming arrival of AI PCs, enabling developers to run advanced artificial intelligence directly on laptops, mobile devices, and wearables. As the demand for local, on-device AI grows, Microsoft competes with similar compact models like Google's Gemma 2 and Hugging Face's Zephyr. The smaller footprint allows for broader adoption without requiring massive computational resources.
In addition to the core family, Microsoft previews Phi-3-vision, a 4.2 billion parameter multimodal variant that processes both text and images. This preview model handles general visual reasoning and interprets charts, graphs, and tables based on user prompts. While Microsoft positions this capability as a cost-effective alternative to larger models, the company does not yet specify a public release date for the vision variant.