Widely regarded as the field's most rigorous test — every entry validated by independent wet-lab experiments under identical conditions — this blinded benchmark saw Aureka Biotechnologies take first place in Challenge 1 (in-silico affinity maturation) with AuraIDE, its in-house antibody design model, entered and cited in the paper as AuraBind.

The winning antibody designed by AuraIDE reached an affinity of 94.7pM as measured by KinExA, an approximately 2,000-fold improvement over the parental antibody. Two further antibodies submitted by AuraIDE ranked second and fifth, giving the model three of the top five places in Challenge 1. In addition, AuraIDE designed a total of six antibodies with affinities below 10 nM that also met the competition's developability criteria.

The AIntibody Challenges: A Blinded, Independently Wet-Lab-Validated Benchmark for AI Antibody Design

Generative AI is already widely used in antibody discovery and protein design, yet the field still has no objective way to measure what any given model can actually do. In protein structure prediction, the blinded-benchmark paradigm established by CASP has become the accepted standard for judging a model's true capability. Through prospective blind testing and unified evaluation criteria, it pulls computational methods out of their own datasets and internal benchmarks and holds them all to one common standard — the way AlphaFold2's breakthrough performance was validated.

AIntibody applies that same approach to antibodies, putting two questions the field has yet to answer to the test: how good are AI-designed antibodies, really — and does any model's design capability hold up under standardized, independent wet-lab testing?

The first AIntibody challenge targeted the receptor-binding domain (RBD) of the SARS-CoV-2 Spike protein. The RBD is among the most thoroughly studied proteins in the world, backed by extensive public structural and sequence data, and the organizers supplied participants with a rich set of experimental screening data on top of that. That abundance is exactly what makes it a demanding test: with so much of the RBD already in public training data, a model can only stand out by having genuinely learned the rules of antibody binding rather than memorized them.

The competition comprised three tasks. Challenge 1 focused on in-silico affinity maturation: from organizer-supplied NGS data covering only the first stage of the parental antibody's affinity maturation, participants had to design new antibodies outright — higher in affinity, and developable. Challenge 2 asked participants to rank existing candidate antibodies by affinity; Challenge 3, to design novel CDR combinations beyond the screening data.

For Aureka, Challenge 1 was the most important of the three. It was the only task built on data from a sequencing pipeline like the ones used in real antibody development, and the one that maps most directly onto lead optimization as it is actually practiced. In its Discussion, Nature Biotechnology further notes that Challenge 1 is, among this year's tasks, the scenario in which the applied value of AI is currently clearest.

The first competition attracted 29 participating organizations and tested 511 AI-designed or AI-predicted antibodies, with participants spanning academia, non-profit organizations, AI startups, biotechnology companies, and large pharmaceutical and technology companies.

Teams had 14 days from the release of each task to submit, and in Challenge 1 no more than 10 sequences each. All submitted sequences were expressed as full-length IgG by the organizers under uniform conditions, first measured for affinity by SPR, with high-affinity candidates then further validated by single-point and standard KinExA.

Affinity alone was not enough. Each antibody also had to clear five developability assessments — HIC, BVP, AC-SINS, Tm and Tagg — covering hydrophobicity, polyreactivity, self-interaction, thermal stability and aggregation propensity. Only candidates whose composite scores met the competition's thresholds counted as developable. And throughout, the process stayed blinded: no participant saw any experimental result until the competition closed.

In a field awash in technical reports and preprints, a blind test like AIntibody earns its weight. It does not take a model's word for its own performance, scored retrospectively on its own dataset. It puts every team's model into head-to-head competition under identical wet-lab conditions, stripping out the distortion of experimental variation and selective reporting, so that capability is judged on real experimental data and nothing else.

AuraIDE Wins: Compute in Place of Wet-Lab Cycles, Efficiently Powering Antibody Affinity Maturation

Challenge 1, which Aureka won, focuses on a critical step in antibody development: once a hit capable of binding the target has been obtained, how to further improve affinity without altering the molecular framework and within a defined variable-region scope, while keeping the molecule developable throughout.

The raw data the competition provided came from NGS sequencing of the parental antibody's first affinity maturation stage. Yeast display libraries were constructed and screened separately for HCDR1, HCDR2, LCDR1, LCDR2 and LCDR3, while HCDR3 and the framework were held constant. That first-stage sequencing output was the only data the competing models ever saw.

In a conventional experimental workflow, however, this is only the first step.

From here, researchers would normally recombine the best-performing mutations from the separate CDR libraries into a combinatorial library and run another round of wet-lab screening, searching the resulting combinations for the clones with the highest affinity.

The data from this round of combinatorial screening, however, was withheld entirely from all competing models.

So the models saw only first-round NGS data, yet had to answer outright a question that a conventional workflow would settle only through further library construction, screening and repeated experimentation:

Which combinations of mutations would actually yield antibodies that are both higher in affinity and developable?

Twenty-five organizations took on that question, submitting 165 antibody designs to Challenge 1. Every one of them was then put through real wet-lab validation under uniform conditions.

This is what makes Challenge 1 fundamentally different from retrospective prediction in the usual sense: the models were not reproducing an experimental answer that already existed, but proposing directly — with the final experimental results unknown — the antibody sequences most worth validating in the next round.

94.7pM: AI Design Surpasses the Best Experimental Result

The final results show that the winning antibody designed by AuraIDE, Aureka's in-house foundation model, had an affinity (KD) of 94.7pM as measured by KinExA — an approximately 2,000-fold improvement in affinity over the parental antibody.

By comparison, the best experimental antibody, obtained through a second round of combinatorial library construction and wet-lab screening, had a KinExA KD of 113 pM.

In other words, a sequence AuraIDE designed in under a week outperformed the best clone from three months of phage maturation experiments. No other team in the competition matched it.

AuraIDE's designed candidate sequences placed 1st, 2nd and 5th on the final leaderboard — all three among the top five. Across its full submission, Aureka produced six antibodies below 10 nM that also met the competition's developability standard. Under blinded conditions and a hard cap on submissions, clustering several candidates near the top says more about a model's reliability than any single top hit does.

What that reliability means in practice is this: AI is no longer just triaging which candidates are worth taking into the lab; it is beginning to do part of the optimization work that once took repeated rounds of library construction, screening and trial and error. Aureka also placed third in Challenge 2 and ninth in Challenge 3 — the only team to finish in the top 10 of all three.

Starting from Existing Data, Exploring a Broader Sequence Space

The winning sequence was something new — not a recombination of the high-frequency mutations already present in the experimental data.

Calculated by Levenshtein distance over the concatenated full-length CDR sequences, the winning antibody differs from the original parental antibody at 19 amino acid positions; even compared with the most similar sequence in the experimental dataset, it still differs by at least 12 amino acids. By contrast, the other top-ranked designs were on the whole closer to existing experimental sequences.

AuraIDE, then, is not mining the existing NGS data for its most frequent mutation combinations and stitching them together. Instead it extracts from the first round of experimental data the signal that guides affinity optimization, then explores sequence space the screening data never reached, generating high-affinity antibodies that hold up under independent wet-lab validation.

From first-round NGS data to a 94.7pM winning antibody; from a single top hit to multiple high-affinity, developable candidate sequences — AIntibody's prospective, blinded wet-lab benchmark has provided a direct, independent validation of what AuraIDE can do in antibody optimization.

For Aureka, though, the win matters for more than first place on a benchmark.

As compute begins to replace part of a wet-lab process that has depended on repeated library construction, screening and trial and error, the value of AI in drug discovery is also beginning to shift from "predicting more accurately" to "developing more quickly."

Paper: https://www.nature.com/articles/s41587-026-03238-6
DOI: https://doi.org/10.1038/s41587-026-03238-6 

About Aureka Biotechnologies

Aureka Biotechnologies is an AI-native TechBio company dedicated to building a new generation of biological foundation models and closed-loop, AI-native infrastructure to reengineer the entire drug discovery process. The company has raised nearly US$200 million to date and has established strategic collaborations with several leading global pharmaceutical companies to jointly advance the development of differentiated antibody therapeutics.

AuraIDE, the company's in-house foundation model, is trained on proprietary protein co-evolution data and has established a leading position in protein folding and de novo design. Its open-source version, OpenDDE, has passed independent third-party evaluation and ranks among the world's leading open-source models; AuraIDE has now also won the global AI antibody design competition published in Nature Biotechnology. Combining a proprietary single-cell functional screening platform with project-specific post-training techniques, the company has produced high-value, differentiated antibodies at scale across a number of programs that are difficult to address with conventional methods, such as GPCRs and dual-target monoclonal antibodies. Through end-to-end, agentic R&D infrastructure, it accelerates the translation of innovative concepts into candidate molecules, providing sustained support for the scaled advancement of both its internal pipeline and external collaboration projects.

The company will direct proceeds primarily toward research and large-scale training of its next generation of biological foundation models, further strengthening performance on core tasks such as de novo molecular design, biological structure modeling and function prediction. Aureka will also upgrade Lab-in-the-Loop, its experiment-centered feedback engine, strengthening the closed-loop between those models and its proprietary single-cell functional screening, high-throughput experimental validation and drug development platforms.

With its closed-loop, AI-native infrastructure already built, Aureka is now strengthening the intelligence core of that system: its foundation models. Aureka combines large-scale pre-training, project-specific post-training, AI agents and experiments that run at scale into AI-for-Science infrastructure for the life sciences. In it, models do not just solve individual drug discovery tasks; they learn the rules of biology, to understand, generate, predict and intervene in complex biological systems.

As foundation models and automated R&D converge, Aureka is shifting from using AI to make drug discovery more efficient to using AI to model living systems, pushing both the technical frontier and the commercial ceiling of AI-driven drug discovery.

Closed-Loop AI-Native Infrastructure Builds a Stronger Intelligence Core

Founded in 2023, Aureka Biotechnologies is an AI-native TechBio company developing a new generation of biological foundation models and closed-loop infrastructure that surrounds them, combining AI models, agents, digital biology and experimental platforms to redesign the drug discovery process end to end.

Biology does not yield to computation alone; it depends on feedback from the physical world. Sustained improvement in large biological models requires more than advances in compute, algorithms and model architecture. It also takes high-quality experimental data that faithfully reflects molecular function, and an experimental system able to continuously test model hypotheses, correcting model bias and feeding results into the next iteration.

Aureka therefore treats Lab-in-the-Loop as core infrastructure for model development, integrating AI agents, high-throughput digital biology, proprietary single-cell functional screening and its in-house experimental platform. The resulting loop runs from molecular generation through experimental design, functional validation and model post-training to candidate development.

In this system, the laboratory is no longer a validation step that follows model output; it is a core part of how the model learns and improves. Models propose experimentally testable molecular designs and scientific hypotheses; the experimental platform generates high-quality functional data; and that data flows back into both the foundation model and project-specific models, driving continuous iteration into the next round of design and validation.

This Lab-in-the-Loop mechanism lets Aureka generate its own large-scale, information-dense functional experimental data for use in foundation model pre-training, reinforcement learning and project-specific post-training. Compared with development paths that rely mainly on public, static datasets, Aureka's models receive experimental feedback from live drug discovery programs and evolve through a continuous design–validation–learning cycle — a flywheel in which data, models, experiments and drug assets reinforce one another.

Foundation Model Capability Confirmed by Third-Party Evaluation

That infrastructure produced AuraIDE, Aureka's own biological foundation model. Trained at scale on proprietary protein co-evolution data, it learns how protein sequence, structure, evolution and function relate to one another. On biomolecular structure prediction and de novo molecular design, it now ranks among the leaders.

Rather than a single-purpose algorithm, AuraIDE is built to transfer across multiple drug discovery programs through task adaptation and project-specific post-training. Its capabilities extend from protein structure modeling and molecular generation into biomolecular interaction modeling, function prediction and multi-objective optimization under complex design constraints.

OpenDDE, the open-source version of AuraIDE, ranks among the world's leading open-source biomolecular models in independent third-party evaluations.

Together, the third-party evaluations and the wet-lab results indicate that Aureka's models lead on protein structure prediction and de novo design, and can translate that capability into measurable molecular function. Through continuous Lab-in-the-Loop feedback, they are moving from predicting biological structure toward generating biomolecules with intended function.

Diversified Commercialization Turns Model Capability into High-Value Drug Assets

Building on its biological foundation models, proprietary single-cell functional screening platform and project-specific post-training, Aureka has produced high-value, differentiated antibodies at scale for problems that conventional approaches struggle with — from difficult target classes such as GPCRs to dual-target antibodies that require a single molecule to engage two targets.

Within a given program, Aureka post-trains its foundation model around target mechanism, functional phenotype and developability objectives, converting general biological intelligence into a dedicated model for a specific drug discovery problem. AI agents then work together across target understanding, molecular generation, computational assessment, experimental design and results analysis, with the resulting experimental data feeding back into that program's model.

The company's end-to-end, agentic R&D infrastructure connects molecular generation, developability assessment, experimental validation, results feedback and candidate development, allowing scientific hypotheses, model capability and experimental capacity to be converted rapidly into developable drug assets that support both internal pipeline programs and external collaborations.

Aureka has established strategic partnerships with multiple leading global pharmaceutical companies to advance the development of differentiated antibody therapeutics, and has generated tens of millions in revenue over the past two years — evidence of the platform's delivery capability, scalability and commercial potential in live drug discovery programs.

From Step-Level Efficiency to Simulating Biological Systems: Toward a Biological World Model

"When leading biological foundation models are genuinely combined with R&D infrastructure that can run at scale, we are no longer simply making one step of drug discovery more efficient — we are building the next-generation drug discovery engine, one that can understand, generate and predict biological systems," said Dr. Weian Zhao, Founder and Chief Executive Officer of Aureka Biotechnologies. "This is a critical step in Aureka's progress toward a biological world model."

On Aureka's long-term roadmap, a biological world model does more than predict static molecular structures. It will simulate interactions between molecules, reason about the likely outcomes of molecular design and engineering, and support AI agents that plan, execute and iterate on drug design tasks autonomously.

With this financing, Aureka will further advance the co-evolution of its biological foundation models and closed-loop AI-native infrastructure, accelerate the validation and translation of model capability in live drug discovery programs, and continue to expand what generative AI can do in antibody drug development.

Investor Perspectives

Granite Asia, Yinghui Kuang:

"AI-driven drug discovery is moving beyond competition on isolated model capability and into a new phase defined by the co-evolution of data, models and experiments in a closed loop. Aureka has built complete AI-native R&D infrastructure with leading biological foundation models as its intelligence core, giving it the systemic ability to generate high-quality data continuously, iterate its models and convert them into drug assets. We are optimistic about the company's long-term technical ceiling in generative antibody design and its potential to grow globally."

HighLight Capital (HLC):

"We are delighted to have completed our investment in Aureka. The company is working to deeply integrate generative AI algorithms with high-throughput wet-lab platforms, with the potential to reshape the paradigm for biologics R&D. We think highly of the team's deep technical foundation in AI-enabled drug development and the efficiency of its closed loop. Going forward, we will continue to commit resources to help the company accelerate pipeline progress and technology iteration, and to bring AI to drug innovation worldwide."

MPCi, Yuye Wang:

"Aureka is a team we backed early and have continued to believe in. OpenDDE, the open-source all-atom model it released, demonstrates this young team's foresight in AI drug discovery and the strength of its technical foundations. Together with the differentiated pipeline the company has now built, we believe a team that combines innovative edge with strategic discipline will keep breaking through bottlenecks in drug development and bring genuine paradigm change to the industry."

NRL Capital:

"Aureka is defining the infrastructure standard for AI drug discovery. The open-source release of OpenDDE marks a key transition from innovation to systematic platform building, closing the loop between computational models and ultra-high-throughput automated wet-lab work and establishing Aureka's global voice in native infrastructure for broadly accessible drug discovery. We recognize the strategic vision of Dr. Weian Zhao and his team in driving industry change through an open-source ecosystem, and we are firm believers in the exponential gains that a dry-lab/wet-lab closed loop delivers in antibody design efficiency. NRL Capital has completed a follow-on investment in this round and will continue to support the company's global expansion and the realization of the global value of its AI infrastructure."

About Aureka Biotechnologies

Aureka Biotechnologies is an AI-native TechBio company building next-generation biological foundation models and closed-loop AI-native infrastructure to transform the therapeutic discovery process. The company has raised nearly $200 million to date and established strategic partnerships with multiple leading global pharmaceutical companies to advance the development of novel antibody therapeutics. Aureka's proprietary foundation model, AuraIDE, is trained on its internal protein co-evolution data and has demonstrated leading capabilities in protein folding and de novo design. Its open-source version, OpenDDE, ranks among the world's leading open-source biomolecular models in independent third-party evaluations.

About Granite Asia

Granite Asia is Asia's most trusted private capital platform, partnering with visionary founders and leaders to build industry champions. With USD 10 billion in assets under management and co-managed capital, the firm has invested in 127 companies valued at over USD 1 billion and supported 67 IPOs worldwide.

About HighLight Capital (HLC)

HighLight Capital (HLC) is a private investment firm dedicated to creating long-term values through promoting technology innovations. Leveraging deep expertise in chemical, biological and materials sciences and proprietary industry research, we invest in companies that enhance manufacturing efficiency and improve human wellness. HLC currently manages over US$4.2 billion.

About MPCi

Founded in 2008, MPCi is one of the leading venture capital firms focused on early stage and early growth deals in China, now managing over 70 billion RMB. MPCi mainly invests in new economy, deep technology, industrial digitalization, healthcare, frontier technology and new consumer brands. MPCi has over 40 investment professionals with deep sector knowledge. The firm also established one of the largest portfolio management teams in the market. Over 80 professionals formed 10 different functions including strategy and operation consulting, recruiting, and healthcare services, etc., to provide value added services to entrepreneurs.

About NRL Capital

NRL Capital is a long-term capital platform that unites top-tier domestic and international industry resources. The firm has built an integrated investment and fund management platform with three interconnected capabilities — direct investment, private equity secondaries, and fund-of-funds — powered by dual-currency operations in both onshore RMB and offshore USD. NRL Capital is deeply focused on pharmaceuticals, medical devices, life sciences, and advanced manufacturing.

LAGUNA HILLS, Calif. and SHANGHAI, July 7, 2026 /PRNewswire/ -- Aureka, an AI TechBio company building infrastructure for AI-driven biologics discovery, today announced the release of Open Drug Discovery Engine (OpenDDE), an open-source, all-atom biomolecular foundation model designed to serve as the structural reasoning core of next-generation drug discovery systems.

OpenDDE uses biomolecular co-folding as the entry point to model interactions across proteins, nucleic acids, small-molecule ligands, and other biomolecular components. Rather than treating structure prediction as an isolated endpoint, OpenDDE is designed as a shared structural reasoning layer for sequence–structure–function modeling, enabling complex structure prediction today while laying the foundation for de novo design, affinity estimation, structure-conditioned optimization, and closed-loop discovery workflows. Across in silico benchmarks, OpenDDE shows competitive co-folding performance and narrows the gap with reported IsoDDE-level results, while offering an open and reproducible framework for the broader drug discovery community.

Three Technical Contributions Behind OpenDDE

• Atomic latent reasoning over biomolecular tokens. OpenDDE introduces latent reasoning for biomolecular modeling by refining representations of local geometry, chemical context, and cross-molecular interfaces before all-atom structure generation.
• A folding-centered foundation for an extensible drug-discovery engine. OpenDDE currently focuses on complex structure prediction, but its unified architecture is designed to support de novo molecular design, affinity prediction, and other structure-conditioned modules.
• Scaling laws and data distillation. Aureka studies scaling directions along model-parameter, data, inference, and training axes, pointing to practical routes for continued improvement in biomolecular foundation models.

Open Benchmark Performance in Antibody-Antigen Co-Folding

In Aureka's technical report, OpenDDE demonstrates strong antibody-antigen co-folding performance across three benchmarks. Under top-ranked selection, OpenDDE reaches 51.0% success on PXMeter-AB, 70.0% on FoldBench-AB, and 66.4% on the newly curated 2026ARK-AB benchmark. Under oracle selection, the corresponding success rates rise to 65.9%, 81.9%, and 80.1%, indicating strong latent sampling capacity and a clear opportunity for further gains through confidence calibration and candidate ranking.

The results are particularly relevant for therapeutic discovery because antibody-antigen interfaces are difficult, flexible, and chemically diverse. Aureka reports that OpenDDE improves not only low-threshold recovery but also medium- and high-quality DockQ regimes, suggesting stronger modeling of binding geometry rather than merely producing marginally acceptable complexes.

Biomolecular Foundation Models Are Entering the Scaling Era

OpenDDE has approximately 655 million trainable parameters. Aureka reports that the model required approximately 414,000 GPU-hours for training, equivalent to roughly 54 years on a single computing unit. This scale reflects a broader shift in AI for Biology: the frontier is no longer only an algorithm problem, but an infrastructure problem requiring compute, data pipelines, engineering, evaluation, and long-running training windows.

Aureka's analysis identifies clear scaling trends for biomolecular foundation models, suggesting that larger effective training corpora, larger models, more capable inference-time sampling, and post-training improvements can systematically translate into stronger biological reasoning and structure prediction. For Aureka, this is a signal that biomolecular AI is beginning to enter a scaling regime analogous to the one that transformed large language models.

An Open Release for the Global Scientific Community

OpenDDE is released to make frontier biomolecular modeling more accessible to researchers, startups, academic laboratories, and multinational corporations. Aureka is releasing training code, inference pipelines, checkpoints, and benchmarks under the Apache-2.0 license, with the goal of enabling independent validation, community-driven extension, and global collaboration.

The release is available at:

• GitHub: https://github.com/aurekaresearch/OpenDDE
• Hugging Face: https://huggingface.co/aurekaresearch/OpenDDE
• Website: https://aurekaresearch.github.io/OpenDDE-Website/

Building TechBio Infrastructures

OpenDDE is an initial foundation for that future. Today, it serves structure prediction, antibody-antigen modeling, and drug discovery research. Over time, Aureka will extend the system toward de novo design, affinity estimation, conformational ensemble modeling, structure-conditioned optimization, experimental feedback, and broader scientific world modeling.

Aureka is also pairing this computational foundation with a high-throughput automated wet-lab platform to build a dry-wet closed-loop discovery system. By integrating autonomous antibody-design agents with high-throughput single-cell functional screening and automated yeast evolution, Aureka aims to create a high-throughput, high-content experimental data flywheel for functional antibody discovery. This platform allows AI agents to propose candidates, test them through automated experimental workflows, absorb functional and phenotypic feedback, and iteratively improve their design strategies.

Together, Aureka's TechBio infrastructures are designed to support the next generation of antibody discovery across complex modalities such as epitope-specific antibodies, multispecific antibodies, internalizing antibodies, and pH-switch antibodies, with the long-term goal of developing differentiated First-in-Class and Best-in-Class therapeutic pipelines.

About Aureka

Aureka is an AI TechBio company developing infrastructure for AI-driven biology and therapeutic discovery. The company integrates large-scale compute, biomolecular foundation models, high-throughput wet-lab systems, and data flywheel capabilities to accelerate discovery in complex biological systems. Aureka's mission is to improve human health by digitalizing and democratizing therapeutic discovery.

Forward-Looking Statements

This press release contains forward-looking statements regarding Aureka, OpenDDE, future research directions, product plans, and potential applications. These statements are based on current expectations and assumptions and are subject to risks and uncertainties, including technical, experimental, regulatory, commercial, and market factors. OpenDDE's current release focuses primarily on biomolecular structure modeling and co-folding. Downstream capabilities such as molecular design, affinity prediction, conformational ensemble modeling, active learning, experimental feedback, and clinical translation require further research, validation, and development. Aureka does not guarantee the discovery, approval, or commercialization of any specific therapeutic candidate.