Phi 4 reasoning

Microsoft releases Phi-4-reasoning, 14B parameter model for structured reasoning tasks

Phi-4-reasoning is a 14 billion parameter dense decoder-only transformer fine-tuned from Phi-4 using supervised fine-tuning on chain-of-thought traces and reinforcement learning. The model targets math, science, and code reasoning tasks with a 32k context window and is optimized for structured two-part responses. It achieves strong results on specialized benchmarks like AIME, OmniMath, and LiveCodeBench, outperforming many larger models in structured reasoning tasks.