Researchers have created a benchmark called Clarify-Then-Search to evaluate how well large language models (LLMs) can improve the results of deep search by asking clarification questions. The benchmark uses real-world query data from Baidu's search engine and tests various LLMs, including GPT-5.2 and ERNIE-4.5-Turbo-128K. Results show that clarification improves search utility, but models often over-ask questions that are unanswerable, leading to unknown responses. The benchmark provides a way to evaluate the effectiveness of clarify-then-search pipelines in deep search. --- Why it matters: This matters because it provides a standardized way to test and compare the performance of large language models in improving search results through clarification questions. This can help researchers and developers improve the accuracy and reliability of search engines, which is crucial for applications where accurate information is critical. Source: https://arxiv.org/abs/2608.20357