Chemical reasoning in LLMs unlocks strategy-aware synthesis planning and reaction mechanism elucidation

Here, we show that large language models (LLMs) can serve as effective chemical reasoning engines when integrated with traditional search algorithms, offering a complementary approach to computer-aided chemistry that mirrors human expert thinking. Rather than using LLMs to directly generate chemical structures, we leverage their ability to evaluate chemical strategies and guide search algorithms toward chemically meaningful solutions. We demonstrate this through two fundamental challenges: strategy-aware retrosynthesis planning and mechanism elucidation. In retrosynthesis planning, our framework allows chemists to specify desired strategies in natural language, identifying routes that satisfy these constraints from large candidate pools with 71% expert agreement. In mechanism elucidation, LLMs guide search toward plausible reaction mechanisms by evaluating elementary electron-pushing steps. Our approach demonstrates how the strategic understanding of LLMs can be combined with the precision of traditional chemical tools, suggesting a path toward more intuitive and powerful chemical reasoning systems.