Google Research and Move37 Labs introduced NucleoBench, the first large-scale benchmark for nucleic acid design algorithms, and AdaBeam, a novel AI-driven algorithm that outperformed existing methods on 11 of 16 tasks. The open-source tool enables efficient DNA/RNA sequence optimization for therapeutics, scaling better on long sequences and large models like Enformer. The benchmark evaluated 400K+ experiments across gene expression control, transcription factor binding, and chromatin accessibility tasks. AdaBeam combines adaptive selection and gradient-guided mutations to enhance drug discovery workflows.