We introduce LMMs-Eval, a comprehensive and efficient benchmark for evaluating Large Multimodal Models, alongside LMMs-Eval Lite and Multimodal Livebench, which ensure low-cost and contamination-free evaluations in dynamic environments.