Qwen3-30B-A3B achieves 8.3% on FukuyamaBench for mechanistic reasoning

Researchers have developed a novel large-scale reasoning dataset and the FukuyamaBench benchmark to evaluate hierarchical mechanism reasoning in large language models, addressing physical inconsistencies common in current chemical LLMs.

• The study introduces FukuyamaBench, a difficult benchmark derived from Fukuyama's Advanced Organic Reaction Mechanism book.
• A fine-tuned Qwen3-30B-A3B model was trained on the new dataset to enhance chemical intelligence.
• The model achieved an 8.3% exact pathway match on FukuyamaBench Set~A, surpassing the specialized FlowER model's 5.1% score.

This demonstrates that mechanism-aware training substantially enhances chemical reasoning capabilities in language models.