TopBench: A Benchmark for Implicit Predictive Reasoning in Tabular Question Answering

Large Language Models (LLMs) have advanced Table Question Answering, where most queries can be answered by extracting information or simple aggregation. However, a common class of real-world queries is implicitly predictive, requiring the inference of unobserved answers from historical patterns rather than mere retrieval. These queries introduce two challenges: recognizing latent intent and reliable predictive reasoning over massive tables. To assess LLMs in such Tabular questiOn answering with implicit Prediction tasks, we introduce TopBench, a benchmark consisting of 779 samples across four sub-tasks, ranging from single-point prediction to decision making, treatment effect analysis, and complex filtering, requiring models to generate outputs spanning reasoning text and structured tables. We evaluate diverse models under both text-based and agentic workflows. Experiments reveal that current models often struggle with intent recognition, defaulting to just lookups. Deeper analysis identifies that accurate intent disambiguation serves as the prerequisite for leading these predictive behaviors. Furthermore, elevating the upper bound of prediction precision requires the integration of more sophisticated modeling or reasoning capabilities.

Many people use spreadsheets or databases to answer everyday questions, such as estimating medical costs, comparing financial options, or selecting promising candidates. Today’s AI systems are good at finding facts that already appear in a table, but real users often ask questions whose answers are not directly written down. For example, a user may describe a new person and ask what their future insurance bill is likely to be, which requires learning from past records rather than simply looking up a row. We introduce TOPBENCH, a benchmark designed to test whether AI systems can recognize these hidden prediction needs and reason over historical tables. TOPBENCH covers four practical scenarios: predicting one unknown outcome, choosing between options, estimating the effect of a change, and ranking or filtering many candidates. Our experiments show that even strong AI models often confuse prediction with retrieval and fail to build reliable predictive solutions. By exposing these weaknesses, TOPBENCH provides a clearer path toward AI assistants that can support more realistic data-driven decisions.