Chem-R: Learning to Reason as a Chemist
Weida Wang, Benteng Chen, Di Zhang, Wanhao Liu, Shuchen Pu, Ben Gao, Jin Zeng, Lei Bai, Wanli Ouyang, Xiaoyong Wei, Tianshu Yu, Tianfan Fu, Shuzhou Sun, Jiatong Li, Zifu Wang, Yuqiang Li, Shufei Zhang
2025-10-22
Summary
This paper introduces Chem-R, a new artificial intelligence model designed to be really good at chemistry problems, aiming to help with discovering new chemicals and reactions.
What's the problem?
Current large language models, while powerful, don't really *understand* chemistry. They lack basic chemical knowledge, often make mistakes in their reasoning when solving problems, and don't perform well on a variety of different chemistry tasks. Essentially, they can process words related to chemistry, but they can't 'think' like a chemist.
What's the solution?
The researchers created Chem-R using a three-step training process. First, they gave it a solid foundation in chemical facts. Then, they showed it how experts solve problems, teaching it to think step-by-step. Finally, they trained it to perform well on many different types of chemistry tasks at the same time, ensuring it's well-rounded. This careful training allows Chem-R to be more accurate and reliable.
Why it matters?
Chem-R is a significant improvement over existing AI models, even very advanced ones like Gemini-2.5-Pro. It performs much better on both understanding molecules and predicting chemical reactions. This means it has the potential to speed up the process of discovering new materials and chemicals, which could have a big impact on fields like medicine and materials science.
Abstract
Although large language models (LLMs) have significant potential to advance chemical discovery, current LLMs lack core chemical knowledge, produce unreliable reasoning trajectories, and exhibit suboptimal performance across diverse chemical tasks. To address these challenges, we propose Chem-R, a generalizable Chemical Reasoning model designed to emulate the deliberative processes of chemists. Chem-R is trained through a three-phase framework that progressively builds advanced reasoning capabilities, including: 1) Chemical Foundation Training, which establishes core chemical knowledge. 2) Chemical Reasoning Protocol Distillation, incorporating structured, expert-like reasoning traces to guide systematic and reliable problem solving. 3) Multi-task Group Relative Policy Optimization that optimizes the model for balanced performance across diverse molecular- and reaction-level tasks. This structured pipeline enables Chem-R to achieve state-of-the-art performance on comprehensive benchmarks, surpassing leading large language models, including Gemini-2.5-Pro and DeepSeek-R1, by up to 46% on molecular tasks and 66% on reaction tasks. Meanwhile, Chem-R also consistently outperforms the existing chemical foundation models across both molecular and reaction level tasks. These results highlight Chem-R's robust generalization, interpretability, and potential as a foundation for next-generation AI-driven chemical discovery.