Microsoft: Phi-4 Reasoning Competes with Larger Models and Achieves DeepSeek-R1 Performance 10:01May 2, 2025