Microsoft Phi-3 Vision: 4.2B multimodal parameters for edge devices

Phi-3 Vision is a Microsoft model that understands both text and images together, with one special characteristic: it's small enough to run on smartphones and laptops without cloud connectivity. With just 4.2 billion parameters it handles very long documents (up to 128,000 words) and reasons about images. It outperforms models ten times larger on many tests, proving that training data quality matters more than size.