Phi 2 has been added to the hallucination leaderboard, which measures the factual consistency of LLM-generated summaries from sets of underlying facts in the retrieval-augmented generation (RAG) architecture. The model is seen as a significant improvement over larger models, with its novel training strategy allowing it to outperform them despite having fewer parameters. This has made it easier for Microsoft to incorporate Phi 2 into commercial software and enterprise systems, putting pressure on proprietary LLM vendors to accelerate innovation and reduce prices, ultimately benefiting consumers.