Data Party THU

Jun 8, 2026 · Artificial Intelligence

Can Large Language Models Design Chemical Synthesis? ChemReason‑Bench Exposes AI’s Logic Gaps

The ChemReason‑Bench benchmark, introduced by Shanghai Jiao Tong University, evaluates large language models on six program‑reasoning tasks for chemical synthesis, revealing that while top general models show modest reasoning ability, step‑completion remains difficult and domain‑specific models lag behind, prompting new training datasets for improvement.