Aureka Releases OpenDDE for Open Drug Discovery

Aureka released OpenDDE, an open-source biomolecular foundation model for drug discovery, on July 6-7, 2026. The model, with 655 million parameters and trained on 414,000 GPU-hours, provides a reproducible benchmark for co-folding and structural reasoning across biomolecules. It aims to advance drug discovery from structure prediction to design and closed-loop lab workflows.

For AI and data-science practitioners, OpenDDE matters less as a finished drug-discovery product than as a reproducible benchmark and model stack for biomolecular reasoning. Aureka announced OpenDDE on July 6/7, 2026, as an Apache-2.0 open-source, all-atom biomolecular foundation model for co-folding and structural reasoning across proteins, nucleic acids, ligands, and complexes. The release includes training code, inference pipelines, checkpoints, benchmarks, and a technical report. Aureka reports 655 million trainable parameters, roughly 414,000 GPU-hours of training, and antibody-antigen co-folding results across PXMeter-AB, FoldBench-AB, and 2026ARK-AB. The immediate value is not clinical validation; it is a public artifact for testing whether drug-discovery systems can move from isolated structure prediction toward design, affinity estimation, and closed-loop lab workflows.