TLDRocket
Sign in

AI & Chemistry

2 summarised stories about AI & Chemistry, each linking back to the original source. Browse all topics →

Thursday, 16 July 2026

Catalyst-Agent: Autonomous heterogeneous catalyst screening with an LLM Agent

arXiv cs.CL 18 hours ago

Catalyst-Agent is an LLM-powered system that autonomously screens heterogeneous catalysts for electrochemical reactions by coordinating materials database searches, slab construction, adsorption energy calculations, and structural modifications. The system converged in 1.40-3.41 trials per successful material on average across oxygen reduction, nitrogen reduction, and CO2 reduction reaction campaigns, identifying five previously unreported CO2RR candidates including Sn3Sc and Sn3Y. This approach demonstrates that LLM agents can replace manual trial-and-error workflows in catalyst discovery with autonomous, reproducible screening pipelines.

Learning Mechanistic Reasoning for Chemical Reactions with Large Language Models

arXiv cs.CL 18 hours ago

Researchers created a large-scale dataset and benchmark to train LLMs on step-by-step chemical reaction mechanisms, addressing limitations where current chemical LLMs focus on name reactions and produce physically inconsistent predictions. The fine-tuned Qwen3-30B-A3B model achieved 8.3% exact pathway match on the FukuyamaBench benchmark, compared to 5.1% for the specialized FlowER model. This mechanism-aware training approach improves the chemical reasoning capabilities of language models for more accurate reaction prediction.