SMDD-Bench: Can LLMs Solve Real-World Small Molecule Drug Design Tasks?

Abstract

LLM agents have incredible potential for scientific discovery applications. However, the performance of LLM agents on real-world, small molecule drug design (SMDD) tasks across diverse chemistries and targets is unclear. Current evaluation methods are either ad hoc, too simple for real-world discovery, limited in scale, or restricted to single-turn question answering. In effort to standardize the evaluation of LLM agents on small molecule design, we introduce SMDD-Bench, a challenging, multi-turn, long-horizon agentic benchmark consisting of 502 guaranteed-solvable task instances spanning 5 task types: 2D Pharmacophore Identification, Interaction Point Discovery, Scaffold Hopping, Lead Optimization, and Fragment Assembly. SMDD-Bench tasks span a wide region of chemical space and involve 102 unique protein targets. Completely solving the benchmark would require having strong chemical and biological reasoning and 3D intuition, understanding specialized tool use, and displaying planning expertise over a limited number of oracle calls. We benchmark 7 frontier open and closed source LLMs and find even the most performant LLM, GPT5.4, solves only 40.2\% of tasks. We hope SMDD-Bench provides a standardized testbed to invigorate the field towards training and evaluating LLM agents for fully autonomous computational drug design. We host a public leaderboard at smddbench.com .

Small-Molecule Data Splitting for Machine Learning in Drug Design
Cao-Minh Truong, Van-Thinh To, Gia-Bao Truong +4
ChemRxiv30 Sep 2026PreprintCC BYFull textGeneral chemistryChemRxiv
Small Organic Molecule-Based Multicolor Fluorescent Smart Materials: From Design Principles and Optical Properties to Versatile Applications
Zhen Yan, Taihong Liu, Liping Ding +1
ACS Central Science18 May 2026Peer-reviewedCC BY5 citedMaterials & polymersOpenAlex
Computational design and prioritization of database-absent small-molecule GLP-1 receptor agonist candidates from public bioactivity data
JOGU MOHAN
ChemRxiv12 Oct 2026PreprintCC BYFull textGeneral chemistryChemRxiv
Recovering Small Molecule Ligand—RNA Binding Pocket Binding Modes and Exploring Potential Binding Pathways Using Metadynamics
Molecules (Basel, Switzerland)26 Sep 2026Peer-reviewedCC BYFull textGeneral chemistryEurope PMC
Emerging applications of Zintl clusters in small molecule activation and catalysis
Bono van IJzendoorn, John S. Wenger, Meera Mehta
Chemical science27 Oct 2026Peer-reviewedCC BYFull textOrganic chemistryEurope PMC

Cite

BibTeX RIS

BibTeX

@misc{han2026smddbench,
  title = {{SMDD-Bench: Can LLMs Solve Real-World Small Molecule Drug Design Tasks?}},
  author = {Kevin Han and Rui Zhang and Kathy Y. Wei and Hamed Mahdavi and Niloofar Mireshghallah and Amir Barati Farimani},
  howpublished = {arXiv (Cornell University)},
  year = {2026},
  month = may,
  doi = {10.48550/arxiv.2605.21740},
  url = {https://arxiv.org/abs/2605.21740},
  note = {CC BY},
}

RIS

TY  - UNPB
TI  - SMDD-Bench: Can LLMs Solve Real-World Small Molecule Drug Design Tasks?
AU  - Kevin Han
AU  - Rui Zhang
AU  - Kathy Y. Wei
AU  - Hamed Mahdavi
AU  - Niloofar Mireshghallah
AU  - Amir Barati Farimani
PY  - 2026
DA  - 2026/05/20
PB  - arXiv (Cornell University)
DO  - 10.48550/arxiv.2605.21740
UR  - https://arxiv.org/abs/2605.21740
AB  - LLM agents have incredible potential for scientific discovery applications. However, the performance of LLM agents on real-world, small molecule drug design (SMDD) tasks across diverse chemistries and targets is unclear. Current evaluation methods are either ad hoc, too simple for real-world discovery, limited in scale, or restricted to single-turn question answering. In effort to standardize the evaluation of LLM agents on small molecule design, we introduce SMDD-Bench, a challenging, multi-turn, long-horizon agentic benchmark consisting of 502 guaranteed-solvable task instances spanning 5 task types: 2D Pharmacophore Identification, Interaction Point Discovery, Scaffold Hopping, Lead Optimization, and Fragment Assembly. SMDD-Bench tasks span a wide region of chemical space and involve 102 unique protein targets. Completely solving the benchmark would require having strong chemical and biological reasoning and 3D intuition, understanding specialized tool use, and displaying planning expertise over a limited number of oracle calls. We benchmark 7 frontier open and closed source LLMs and find even the most performant LLM, GPT5.4, solves only 40.2\% of tasks. We hope SMDD-Bench provides a standardized testbed to invigorate the field towards training and evaluating LLM agents for fully autonomous computational drug design. We host a public leaderboard at smddbench.com .
SN  - 2331-8422
ER  -

Abstract

Related

Cite