PhiBench

PhiBench is Microsoft's internal reasoning benchmark that evaluates language models' performance in complex reasoning tasks and language understanding. This benchmark serves as a key evaluation metric for Phi model series development, testing models' capabilities in logical reasoning, problem-solving, and sophisticated language comprehension across various domains and difficulty levels.