Browse Models
Phi-4 is a 14-billion parameter open-weights language model developed by Microsoft Research, featuring a decoder-only Transformer architecture with a 16,000-token context window. The model utilizes synthetic data generation and multi-stage fine-tuning for enhanced reasoning capabilities, achieving competitive performance on benchmarks like MMLU (84.8) and HumanEval (82.6) while specializing in mathematics, code generation, and scientific reasoning tasks.