Satya Nadella’s Post
Another big step forward for our SLM Phi family, with new reasoning models that once again redefine what is possible with small and efficient AI.
Happy Birthday, Phi! To celebrate, we are introducing three new models in the Phi-4 reasoning family on Azure AI Foundry: 1️⃣Phi-4-reasoning (14B, SFT on curated reasoning demos) 2️⃣Phi-4-reasoning-plus (14B, SFT + RL, ~1.5x inference-time tokens) 3️⃣Phi-4-mini-reasoning (3.8B, distilled + SFT with DeepSeek-R1-generated math problems) These models leverage distillation, SFT, RL, and synthetic data pipelines to achieve frontier-level capability in compact form. They are optimized for multi-step reasoning, mathematical problem solving, and internal reflection—tasks traditionally dominated by much larger frontier models. News here: https://lnkd.in/gC_56Dfy
