Aureka Biotechnologies closed a $100 million Series B on August 10. Granite Asia funded the first tranche, a strategic investor led a later tranche, HighLight Capital participated, and existing investors including MPCi and NRL Capital followed on. The company says it has raised nearly $200 million in total.
The money goes into feedback
Aureka says the proceeds will support next-generation biological foundation models, large-scale training, de novo molecular design, biological structure modeling and function prediction. It will also upgrade Lab-in-the-Loop, its experiment-centered feedback engine.
The useful translation is simple: the lab is becoming part of the training system. Models propose molecules and hypotheses; experiments test them; the data goes back into the foundation model and project-specific models.
“Biology does not yield to computation alone; it depends on feedback from the physical world.”
OpenDDE is the public proof point
Aureka's proprietary model is AuraIDE. Its open-source version, OpenDDE, was submitted to arXiv on July 4 as an all-atom biomolecular foundation model built around co-folding. The paper describes a system for structure prediction, de novo design, affinity estimation and structure-conditioned optimization.
Chinese interview material from 36Kr adds business context. Founder and CEO Weian Zhao said the company used a thousand-GPU-class cluster for AuraIDE and released OpenDDE as a public credibility test. The company says OpenDDE ranks among leading open-source biomolecular models in third-party evaluations. 36Kr reports that, on antibody-antigen structure prediction, OpenDDE improved overall prediction accuracy by about 1.5 times compared with AlphaFold3. Zhao also cited an internal autoimmune target: earlier workflows produced five or six active sequences out of 50 designs, while OpenDDE produced more than 30 active sequences, lifting hit rate to 60%.
The commercial test is molecules
The phrase biological world model is ambitious. Aureka's long-term roadmap points to models that simulate molecular interactions, reason about design outcomes and support AI agents that plan and iterate drug-design tasks.
The nearer test is more practical. Pharma partners can bring hard targets that animal immunization or library screening cannot handle, ask for candidate molecules under defined constraints, and then validate those molecules through Aureka's dry-wet loop, a pharma lab or a CRO. If candidates meet the bar, the relationship can move into licensing or co-development.
Aureka says it has strategic partnerships with multiple global pharmaceutical companies and generated tens of millions of dollars in revenue over the past two years. 36Kr reports more than ten candidate programs across cardiometabolic, autoimmune and central nervous system diseases, with the fastest cardiometabolic program expected to file an IND in the first quarter of next year.
Three checkpoints
The next checkpoints are concrete: whether a lighter OpenDDE inference branch ships within the planned half-year window, whether the lead program reaches IND, and whether pharma collaborations grow from pilot projects into larger licensing deals. The financing shows investor appetite. Drug candidates and paid partnerships will show whether the model-lab loop can hold up outside the pitch.
Sources: Aureka funding announcement, 36Kr, Synced, VBData, CocoLoop, arXiv:2607.03787; verification covers the $100 million Series B, nearly $200 million total funding, 2023 founding, OpenDDE and FoldBench v1 scope, about 1.5x prediction-accuracy claim, 50-sequence and 60% hit-rate example, tens of millions in revenue and IND plan.