CSCR embeds both prompts and LLMs into a shared space using fast logit or perplexity fingerprints. A cost‑banded InfoNCE loss trains the space to balance quality against cost. It generalizes to unseen models and out‑of‑distribution prompts.

TRIDENT integrates molecular structures, text, and taxonomic annotations via volume-based global alignment and local correspondence learning, achieving state-of-the-art performance in molecular property prediction tasks.