REAL: Benchmarking Abilities of Large Language Models for Housing Transactions and Services