Hyperspectral image(HSI) denoising addresses noise impact during image acquisition. Transformers have gained notable prominence in the field of denoising, but their quadratic self-attention complexity poses computational challenges, hindering global information processing. Classical window-based self-attention limits non-local information flow, hampering large-scale object capture in HSI. Furthermore, spectral variations among neighboring bands introduce redundancy, which burdens the model and diminishes token variability, resulting in over-smoothing in the attention map. To address these issues, we propose a novel method, named Spectral Aggregation Cross-Square Transformer(SACT). We introduce a cross-square self-attention mechanism to enhance information exchange between windows, capturing long-range dependencies within spatial intra-spectrum from multiple perspectives. Spectrally, it extends the attention region horizontally, surrounding, and vertically, exploring omnidirectional spatial correlations among different receptive windows. Additionally, a spatial-spectral aggregation self-attention module is designed to capture global contextual dependencies across spatial and spectral dimensions, reducing spectral redundancy computation. Our method has evaluated synthetic and real hyperspectral datasets and shows SACT’s effectiveness in enhancing both quantitative and qualitative HSI denoising performance.