While Vision-Language Models (VLMs) have demonstrated strong capabilities in general multimodal tasks, they face significant challenges in low-resource, cross-temporal settings such as ancient Chinese documents, where models must overcome complex visual noise and leverage deep linguistic priors. The functional organization of task-relevant components in existing VLMs remains understudied from a computational interpretability perspective. To address this, we introduce AncientVision-3T, an intervention-oriented diagnostic benchmark designed for mechanistic interpretability experiments on VLMs, organized across OCR, image classification, and image understanding tasks of increasing abstraction. Based on this benchmark, we propose the Ancient Chinese Target-Conditioned Gradient-based Neuron analysis framework (AC-TCGN), which identifies and validates task-relevant neurons for dissecting functional organization in general and domain-specific VLMs. Our experiments demonstrate that AC-TCGN effectively identifies neurons that are critical for model predictions. Moreover, our experimental results reveal a transition from separation to entanglement in the functional organization of decoder feed-forward networks: general VLMs show stronger task-wise separability, whereas the ancient Chinese domain-specific VLM shows greater cross-task neuron sharing and reduced task-wise separability. This suggests enhanced coupling between visual form and historical-semantic processing in ancient Chinese document image-text tasks.

Ancient Chinese books have great values in history and cultural studies. Named en-tities like person, location, time are cru-cial elements, thus automatic Named En-tity Recognition (NER) is considered a ba-sic task in ancient Chinese text processing. This paper introduces EvaHan2025, the first international ancient Chinese Named Entity Recognition bake-off. The evalua-tion introduces a rigorous benchmark for assessing NER performance across histori-cal and medical texts, covering 12 named entity types. A total of 13 teams par-ticipated in the competition, submitting 77 system runs. In the closed modality, where participants were restricted to us-ing only the training data, the highest F1 scores reached 85.04% on TestA and 90.28% on TestB, both derived from his-torical texts, while performance on medi-cal texts (TestC) reached 84.49%. The re-sults indicate that text genre significantly impacts model performance, with histori-cal texts generally yielding higher scores. Additionally, the intrinsic characteristics of named entities also influence recogni-tion performance. These findings high-light the challenges and opportunities in ancient Chinese NER and underscore the importance of domain adaptation and en-tity type diversity in future research.