Breast cancer care requires multimodal reasoning across a clinical workflow spanning screening, diagnosis, and treatment planning. However, existing medical MLLMs are typically limited to isolated modalities or narrow task settings, restricting their ability to support workflow-level clinical reasoning. To address this gap, we introduce BreastStage, a workflow-aligned breast imaging instruction corpus containing 1.86M instruction-following pairs from 17 sub-datasets, 5 imaging modalities, and 136 task templates, together with BreastStage-Bench, a benchmark for evaluating multimodal reasoning across the breast cancer care continuum. Built upon this corpus, we propose BreastGPT, a unified MLLM with a dual-branch visual encoder and concept-preserving token compression to bridge standard radiology and gigapixel pathology. On BreastStage-Bench, BreastGPT achieves 75.66% closed-ended accuracy and 89.92% open-ended score, outperforming both general-purpose and medical-specific MLLMs across clinical stages and task formats. Data, code, and model checkpoints are publicly released at https://anonymous.4open.science/w/BreastGPT_io/.

Tokenizers play a critical role in large language model studies. Despite recent advances, existing tokenizers fail to explicitly leverage historical tokenization results when making subsequent token decisions, nor do they selectively utilize such history based on contextual relevance. We propose SPEAK, a tokenizer that integrates spiking neurons to explicitly leverage historical tokenization results. Furthermore, we introduce an entropy-aware reset mechanism that selectively leverages history based on contextual relevance, which is determined by token-level entropy. High-entropy tokens are treated as contextual boundaries, whereas low-entropy tokens between consecutive such boundaries exhibit strong contextual relevance. Accordingly, we induce hard reset at high-entropy tokens to discard irrelevant historical tokenization results, and soft reset at low-entropy tokens to preserve and leverage relevant history. Experiments on 2 language models and 5 datasets spanning 16 languages demonstrate superior cross-lingual adaptability, with competitive performance and efficiency. Our code is publicly available at https://github.com/zju-bmi-lab/SPEAK.