Aug 2026· Proceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2· pp. 12291-12302· 0 citations· 14 references
TL;DR
The proposed Chem-R, a general Chemical Reasoning model designed to emulate the deliberative processes of chemists, achieves state-of-the-art performance on comprehensive benchmarks, surpassing leading LLMs, including Gemini-3-Pro and Kimi-k2.5.
Abstract
Despite the potential of Large Language Models (LLMs) in chemical discovery, current LLMs still lack fundamental chemical domain knowledge, produce incoherent reasoning trajectories, and exhibit suboptimal performance across diverse chemical tasks. To address these challenges, we propose Chem-R, a general Chemical Reasoning model designed to emulate the deliberative processes of chemists. To build advanced reasoning capabilities of Chem-R, we design a three-phase training framework, including: 1) Chemical Foundation Training (CFT), which establishes core chemical knowledge. 2) Chemical Reasoning Protocol (CRP) Distillation, incorporating structured, expert-like reasoning traces to guide systematic and reliable problem solving. 3) Chemical Multi-Task Optimization (CMO) that optimizes the model for generalizable capabilities 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 LLMs, including Gemini-3-Pro and Kimi-k2.5, by up to 19% on molecular tasks and 40% on reaction tasks. Meanwhile, Chem-R also consistently outperforms existing chemical foundation models across both molecular and reaction level tasks. These results demonstrate Chem-R's superior generalization, interpretability, and potential as a foundation for next-generation AI-driven chemical discovery. The code and model are available at https://github.com/davidweidawang/Chem-R.
Single-step retrosynthesis is a central component of computer-aided synthesis planning, yet its intrinsically one-to-many nature is poorly captured by single-answer evaluation and benchmarking protocols. To address this, we introduce Top-K prompting as a robust training and inference paradigm to better capture diverse, plausible reaction predictions. We compile CREED-CCV-2+USPTO-XL, an ultra-large-scale dataset of ~45.6 million verified reactions to train the C3LM (Chemistry Constraint-Consistent Language Model). By integrating fine-tuning with ChemCensor-based and novelty-oriented rewards, our model achieves state-of-the-art performance on the OOD URSA-expert-2026 benchmark. Further analysis of reaction uniqueness shows that LLMs and conventional models explore complementary reaction spaces, motivating ensemble-based retrosynthesis systems. Overall, our results establish Top-K, plausibility-aware training as a practical new direction for robust future LLM-based synthesis planning.
B. Zagribelnyy, Ivan D. Ilin, N. Bondarev et al.· 0 citations
The rapid advancement of large language models (LLMs) has led to increasing interest in their application to scientific domains such as chemistry. However, existing chemistry benchmarks often provide only a narrow view of model capability, focusing on limited task sets while overlooking robustness to variations in problem formulation and chemical representation. As a result, reported performance may overestimate a model's true ability to reason consistently across realistic settings. To address this challenge, we introduce ChemDIRT (Diversified Instruction, Representation, and Task Benchmark), a comprehensive evaluation framework designed to assess the robustness of chemical reasoning in LLMs. ChemDIRT systematically measures model performance across variations in instructions and molecular representations while spanning eight categories of chemistry tasks. By evaluating both accuracy and consistency under these controlled perturbations, ChemDIRT provides a more reliable assessment of model reasoning capabilities than conventional single-format benchmarks. We benchmark a diverse set of open- and closed-source LLMs, revealing substantial prompt sensitivity, representation dependence, and uneven performance across task families.
Eric Inae, Tim Gunn, Chris Bond et al.· 0 citations
A novel, large-scale reasoning dataset of reaction mechanisms, and the FukuyamaBench, a difficult benchmark derived from Fukuyama's Advanced Organic Reaction Mechanism book, to rigorously evaluate model performance on hierarchical mechanism reasoning, demonstrate that mechanism-aware training substantially enhances chemical reasoning in language models.
Xingyu Dang, Haocheng Tang, Junmei Wang et al.· arXiv.org· 0 citations
A task-adaptive large reasoning model that integrates chemical knowledge through a synergistic multispecialist architecture, chain-of-thought supervision, and molecule-informed reinforcement learning is presented, demonstrating a versatile multitask framework for knowledge-guided molecular reasoning and design.
Pengfei Liu, Shuang Ge, Xiaobo Wang et al.· Journal of Physical Chemistr...· 0 citations
Language models are playing an increasingly important role in laboratory science, performing tasks such as experiment planning, execution, and post-hoc analysis. However, precisely measuring their abilities is difficult, as scientific capabilities require a mixture of both problem-solving skills and domain-specific intuition. Existing evaluations rarely measure the capabilities required to make reliable decisions in a physical laboratory and often rely on public data that may have appeared in model training corpora. We introduce onepot-Bench 0, a proprietary benchmark suite for evaluating language models on synthetic chemistry capabilities relevant to wet-lab execution. onepot-Bench 0 comprises three complementary evaluations: ChemAbacus measures tool-free cheminformatics literacy and numerical reasoning; SynthRefusal characterizes safety and refusal behavior across a variety of benign, controlled, and designer-drug targets; and SynthBench evaluates reaction-outcome prediction and catalyst selection using private experimental data generated in our laboratory. Together, these evaluations probe basic competency, reliability, and deeper knowledge, all skills which are required for reliable performance in the lab.
Brandon Wang, Andrei S. Tyrin, Daniil A. Boiko· 0 citations
Chemistry navigates an immense molecular space using compact, transferable knowledge structures that are neither strict physical laws nor merely qualitative intuition. We argue that this approach is best understood as compression: deliberate, lossy, and interpretable reductions of chemical details that generalize surprisingly well across molecules, reactions, and domains. Machine learning (ML) performs a different kind of compression, optimizing against training objectives rather than chemical judgment. The result is models that excel within their training domains but can fail outside them in ways that are difficult to diagnose, often without providing the mechanistic insight that makes chemical knowledge valuable and transferable. This difference defines an epistemic gap that is the central challenge for artificial intelligence (AI) in chemistry. Physics-based AI systems offer a partial path forward, but we believe this gap will persist for the foreseeable future with important consequences for how chemistry is practiced and taught.
Oliver Steinbock, Wen Zhu, Wei Yang et al.· Chemistry· 0 citations