Microsoft Releases Phi-4-Reasoning-Vision-15B Multimodal Model

Microsoft Research has released Phi-4-reasoning-vision-15B, a compact yet powerful multimodal AI model. As an 'open-weight' model, its architecture and trained weights are publicly available, fostering broader research and development. Despite its relatively small size of 15 billion parameters, it claims to match or exceed the performance of much larger systems while consuming significantly less computational power and training data. A key innovation is its ability to intelligently decide when to employ explicit, chain-of-thought reasoning processes versus providing a direct response. This 'reasoning-on-demand' approach optimizes efficiency, using complex logic only when necessary for accuracy. The model represents a push toward more sustainable and accessible high-performance AI, proving that smaller, smarterly designed models can be highly capable in understanding and reasoning across both text and visual inputs.