Microsoft Releases New Reasoning Model Phi – 4 – reasoning – plus

Microsoft recently released the Phi-4-reasoning-plus reasoning model. With a parameter count of 14 billion, it demonstrates performance on par with o1-mini, o3-mini, and Sonnet 3.7. The model was meticulously fine-tuned using approximately 1.4 million carefully curated reasoning demonstrations via supervised fine-tuning (SFT), followed by limited reinforcement learning (RL) optimization, achieving exceptional results.​