MSP-LLM Introduces Unified Material Synthesis Planning Framework

Heewoong Noh (submitted Feb 7, 2026) presents MSP-LLM, a unified large language model framework that frames material synthesis planning as two subproblems: precursor prediction (PP) and synthesis operation prediction (SOP). The method introduces a discrete material class intermediate, hierarchical precursor types, and explicit conditioning to retain precursor information during autoregressive decoding, improving chemical consistency. Experiments show MSP-LLM outperforms prior methods on PP, SOP, and complete MSP.