MTBench

MTBench is a large-scale benchmark that contains paired time series and textual data from financial and weather domains, used to evaluate large language models on their ability to understand and reason across both structured numerical trends and unstructured textual narratives.

Papers using MTBench (3)

Masked Language Flow Models2026IMTBench: A Multi-Scenario Cross-Modal Collaborative Evaluation Benchmark for In-Image Machine Translation2026MTBench: A Multimodal Time Series Benchmark for Temporal Reasoning and Question Answering2025