Imagine asking AI a question like, "Did the director of 'Pulp Fiction' win an Oscar for another movie before 1994?" This requires pulling together facts from multiple sources and reasoning over them—a challenge for even the smartest AI. A new benchmark called DEXTER is putting today's AI models to the test. Researchers have created a dataset of complex questions covering diverse topics, including multi-hop reasoning, comparisons, numerical reasoning, and even deciphering ambiguous queries. DEXTER then evaluates how well AI can retrieve relevant information from a vast knowledge base and synthesize an accurate answer. The findings are intriguing. While AI excels at simple question answering, DEXTER shows that complex questions are a whole different ballgame. Traditional keyword search methods, like BM25, surprisingly hold their own against more modern neural networks, especially when dealing with different data types like tables and text. Advanced AI models, however, struggle with intricate reasoning, revealing the gap between current AI capabilities and true human-like understanding. Even when given the correct information, AI sometimes fails to connect the dots, especially with tasks involving ambiguity or complex numerical reasoning. This benchmark is more than just a test—it's a roadmap for future AI development. DEXTER highlights the need for more sophisticated reasoning and retrieval mechanisms, especially when information is spread across multiple sources. This research pushes us closer to AI that can genuinely grasp nuanced questions and navigate the complexities of information, opening doors to groundbreaking applications in fields from medicine to finance.