The field of multimodal research is moving towards more comprehensive and nuanced evaluation of models' abilities to understand and reason about complex relationships between visual and textual information. This shift is driven by the need for more realistic and challenging scenarios that go beyond surface-level semantic correspondence. Recent work has focused on developing benchmarks and evaluation platforms that can effectively assess models' capacity for logical, spatial, and causal inference. Notable papers in this area include: MRAG-Suite, which introduces a diagnostic evaluation platform for visual retrieval-augmented generation, highlighting the need for more systematic and nuanced evaluation of query difficulty and ambiguity. Q-Mirror, which presents a framework for transforming text-only QA pairs into high-quality multi-modal QA pairs, demonstrating the potential for large-scale scientific benchmarks. MR$^2$-Bench, which introduces a reasoning-intensive benchmark for multimodal retrieval, showing that current state-of-the-art models still struggle with deeper reasoning required to capture complex relationships. OIG-Bench, which presents a comprehensive benchmark for one-image guide understanding, revealing notable weaknesses in semantic understanding and logical reasoning among current models. MDSEval, which introduces a meta-evaluation benchmark for multimodal dialogue summarization, providing a strong foundation for developing effective evaluation methods.