AI Paper Atlas

Research map / Large language models / Agents, code and math

CatalystBench: A Comprehensive Multi-Task Benchmark for Advancing Language Models in Catalysis Science

Xueqing Chen, Jian Xu, Ludi Wang, Yang Gao, Huihan Zhu, Yuanchun Zhou, Yi Du, Cheng-Lin Liu

ICLR 2026 · Large language models › Agents, code and math · approach: LLMs for scientific discovery

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add task-specific output heads to multi-task fine-tuning (MFT vs MT) (ablation)ChemLLM · CatalystBench-testavg improvement over single-task: 9.24% -> 12.44%Improved
add multi-task training to task-specific output heads (MFT vs heads only) (ablation)ChemLLM · CatalystBench-testavg improvement over single-task: 5.13% -> 12.44%Improved
fine-tune ChemLLM on catalytic data (CatalystLLM vs ChemLLM)ChemLLM · CatalystBench FEPR2: 0.45 -> 0.73Improved
fine-tune ChemLLM on catalytic dataChemLLM · CatalystBench FPMAE: 2.75 -> 1.34Improved
use CatalystLLM instead of gpt-4.1gpt-4.1 · CatalystBench APR2: 0.61 -> 0.81Improved
use CatalystLLM instead of gpt-4.1gpt-4.1 · CatalystBench SEF1: 0.95 -> 0.94Hurt
use CatalystLLM instead of gpt-4.1gpt-4.1 · Experts score @100expert score: 0.56 -> 0.75Improved
use CatalystLLM instead of gpt-4.1gpt-4.1 · LLM score (gpt-4o judge)LLM score: 0.85 -> 0.82Hurt
add regression, classification or extraction task data to training (ablation)MFT · per-task performancetask performanceImproved
add RMC data to multi-task training (ablation)MFT · FEP and RSC taskstask performanceImproved
use structured complete input instead of unstructured core information (ablation)CatalystLLM · CatalystBenchtask performanceImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy