**QuanTemp: A Real-World Open-Domain Benchmark for Fact-Checking Numerical Claims** This paper introduces QuanTemp, a large-scale, diverse dataset focused on numerical claims, including comparative, statistical, interval, and temporal aspects. The dataset is compiled from various fact-checking domains and includes detailed metadata and an evidence collection. The authors evaluate the effectiveness of different methods, such as claim decomposition and natural language inference (NLI) models, in verifying numerical claims. They find that claim decomposition significantly improves verification accuracy, and models pre-trained for numerical understanding outperform generic models. The study also assesses the impact of model size and different NLI models on performance, concluding that smaller, fine-tuned models often outperform larger models in zero-shot and few-shot scenarios. The research highlights the challenges and complexities of verifying numerical claims and provides insights into the development of more effective fact-checking systems.**QuanTemp: A Real-World Open-Domain Benchmark for Fact-Checking Numerical Claims** This paper introduces QuanTemp, a large-scale, diverse dataset focused on numerical claims, including comparative, statistical, interval, and temporal aspects. The dataset is compiled from various fact-checking domains and includes detailed metadata and an evidence collection. The authors evaluate the effectiveness of different methods, such as claim decomposition and natural language inference (NLI) models, in verifying numerical claims. They find that claim decomposition significantly improves verification accuracy, and models pre-trained for numerical understanding outperform generic models. The study also assesses the impact of model size and different NLI models on performance, concluding that smaller, fine-tuned models often outperform larger models in zero-shot and few-shot scenarios. The research highlights the challenges and complexities of verifying numerical claims and provides insights into the development of more effective fact-checking systems.