Microsoft open-sources Phi-4-Reasoning-Vision-15B visual reasoning model

Microsoft has released the Phi-4-Reasoning-Vision-15B open-source model, licensed under the MIT license. It is the first small language model in the Phi-4 series to simultaneously possess high-resolution visual perception and selective reasoning capabilities. The model can automatically switch between "reasoning mode" and "non-reasoning mode" according to task requirements. It employs multi-step thought chains for mathematical reasoning and directly outputs results for simple OCR tasks to reduce latency.