Scientific QA Datasets in the Era of Multimodal and Long-Context Reasoning: A Survey of 2022–2025 Corpora

This comparison highlights recently released scientific question answering (QA) corpora from 2022 to 2025, covering diverse modalities (text, images, multimodal) and domains (e.g., general science, chemistry, astronomy). These datasets represent emerging trends in QA research, including real-world sources like peer reviews and preprints, multimodal reasoning over figures and tables, and the integration of large language models (LLMs) for dataset creation and evaluation. The comparison summarizes each corpus by modality, domain coverage, task focus, and scale, illustrating the shift toward deeper reasoning, long-context comprehension, and domain-specific benchmarks.