<p>Sequential recommendation systems predict user preferences by mining temporal patterns in behavior sequences, where behaviors are driven by latent intents. However, existing intent-aware methods often inadequately leverage inter-item correlations and fail to capture long-term dependencies in behavior sequences—specifically, the persistent influence of early interactions on current decisions. To address these limitations, this paper proposes global–local intent contrastive learning for sequential recommendation (GLICRec), which, for the first time, bridges the gap between short-term precision and long-term intent modeling through unified intent-aware contrastive learning. First, dynamic window segmentation decomposes user sequences into local subsequences while preserving complete histories to provide global context. Next, a cluster-based semantic transformation module constructs compact item clusters and replaces items with semantically similar counterparts during data augmentation, preserving intent semantics while introducing controlled diversity. Moreover, the proposed intent-aware collaborative contrastive learning module achieves cross-scale alignment between global sequences and local subsequences, complemented by a diverse intent prototype space distilled through hybrid intent contrast, collectively enhancing long-term dependency modeling without compromising short-term accuracy. Extensive experiments on four real-world datasets demonstrate that GLICRec achieves average improvements of 5.57% in HR and 4.36% in NDCG compared to state-of-the-art baselines, underscoring its practical value for delivering more accurate and robust recommendations in e-commerce and content platforms.</p>