Radical Numerics raises $50M seed to build AI for biology and biodefense
Eric Nguyen and the Evo research team are turning generative genomics into a company with Omnii, a model aimed at DNA, RNA and proteins.
By Ryan Merket ยท Published
Why it matters
Radical Numerics is turning a major generative-genomics research lineage into a company, with a rare dual-use pitch: the same models that may improve diagnostics and drug design must also detect misuse.
Eric Nguyen (@exnx), Michael Poli (@MichaelPoli6), Stefano Massaroli (@Massastrello) and Armin W. Thomas (@athmsx) launched Radical Numerics from stealth on June 15 with a $50 million seed round to build what they call general biological intelligence, an AI system meant to read, write, design and defend across biology rather than optimize one drug-discovery workflow.
The round was led by Emergence Capital, with Obvious Ventures, Triatomic Capital, Factory and First Spark Ventures participating, according to Radical Numerics' funding announcement and a Business Wire release. Fortune reported that Patrick Collison, Stripe's CEO and an Arc Institute cofounder, backed Radical Numerics at pre-seed. Radical Numerics did not disclose a valuation.
The money is large for a seed round because the founding team is not starting from a pitch deck. Nguyen, Poli, Massaroli and Thomas were part of the research lineage behind Evo and Evo 2, models that pushed long-context sequence modeling into genomics. Nguyen's own site says he co-founded Radical Numerics in 2025 after work at Stanford, Google DeepMind, Google Research, Adobe Research and Facebook AI; his academic path moved from civil engineering at UC Berkeley and Stanford into computer science at Cornell and then a Stanford Bioengineering and AI PhD. In Fortune's telling, Nguyen went back for the PhD to find a problem worth spending his life on. He found DNA.
From Evo to Omnii
Radical Numerics is positioning Omnii, its first model preview, as the next step after Evo-class genome language models. The central claim is breadth: Omnii is designed to operate across DNA, RNA, proteins and biological annotations, not just one molecular type. Radical Numerics says the model uses a multi-hybrid architecture with a 2 million-token context window and is trained with multimodality and alignment methods so researchers can prompt it for prediction, design and interpretation tasks.
Those claims are still largely Radical Numerics' own. In its technical preview, Radical Numerics says Omnii outperforms Evo 2 and other baselines across several clinical-variant, complex-trait and RNA-fitness benchmarks. The benchmark tables are detailed, but the model is still a preview, and Radical Numerics has not yet supplied the kind of independent external benchmarking that would turn "general biological intelligence" from a company thesis into a standardized category.
That distinction matters. AI biology has already produced a crowded map of companies aiming at proteins, RNA, lab automation, target discovery and synthetic biology. Fortune placed Radical Numerics against peers including Ginkgo Bioworks, Isomorphic Labs and Inceptive, and reported that Inceptive has a deal with Alnylam potentially worth $2 billion. Radical Numerics is making a different bet: that the next valuable model in biology will not merely design a molecule, but reason about the system the molecule enters.
Nguyen put the commercial challenge plainly to Fortune: "No one has figured out the right business model for how AI companies commercialize in life sciences." Radical Numerics is starting with two early partnerships, according to Fortune and Radical Numerics' launch materials: one with an unnamed diagnostics company for pancreatic and multi-cancer detection, and one with an unnamed U.S. national lab for pathogen detection and characterization. Revenue, pricing, ARR and paying-customer count were not disclosed.
The safety bet is also the product bet
Radical Numerics' most important strategic choice is not that it is building for health. It is that Radical Numerics is building health design and biodefense inside the same lab.
That decision comes directly from the team's earlier success. Radical Numerics says its founding team helped train Evo and Evo 2, models that learned from genomes across life and showed that AI could generate biological sequence at scale. Fortune reported that researchers using Evo's open-source weights produced the first fully AI-designed functional virus, a bacteriophage that infects bacteria and is not harmful to humans. Radical Numerics' own launch post argues that the design and defense labs should be the same entity.
The biodefense preview for Omnii for defense frames the same architecture as a detection layer for natural, synthetic and AI-generated pathogens. Radical Numerics says existing detection approaches are not enough as generative biology improves, and the company is building toward systems that can identify suspicious sequences, detect engineered function and support biosurveillance. The company has added advisors including Eric Horvitz, Microsoft chief scientific officer, Chris Re of Stanford, George Church of Harvard and Andrew Weber, the former U.S. assistant secretary of defense for nuclear, chemical and biological defense programs.
The round buys time, compute and credibility
Radical Numerics says the seed funding will go toward scaling the next generation of models and hiring frontier AI researchers. That is a compute-heavy plan for a young company, but biology models require more than rented inference capacity: Radical Numerics is trying to train on physical biological data, combine sequence and annotation modalities, and develop interpretability tools for model behavior in living systems.
The founding mix explains why investors moved early. Poli's public profile describes a Stanford computer science PhD focused on machine learning, systems and signal processing, and says he was a founding scientist at Liquid AI, where he worked on hardware-aware AI systems and long-context architectures. Thomas was a Stanford Data Science fellow from 2022 to 2024 after a PhD in machine learning from Technische Universitat Berlin, with work spanning neuroscience, psychology and whole-brain fMRI analysis. Massaroli is identified in Radical Numerics' release as president and a former Liquid AI founding-team member.
The company is hiring through its join-us page, where Radical Numerics describes its work as bringing distributed systems, model architecture and numerics research to scaling learning on biological data. That phrasing is not accidental. Radical Numerics is selling itself as an AI systems lab first, and a biology company through that lens.
The open questions are the ones that matter for whether Radical Numerics becomes a platform or an expensive research lab. The valuation is undisclosed. The early partners are unnamed. Omnii's strongest performance claims come from Radical Numerics' own previews. "General biological intelligence" has no agreed benchmark. And Nguyen's own comment to Fortune makes clear that life-sciences commercialization remains unsettled.
Still, Radical Numerics begins with a rare starting point: a team whose prior research already changed what genome models could generate, and enough capital to test whether that capability can become infrastructure for both medicine and biosecurity. The founder bet is that biology will not be won by a narrower model for one molecule at a time. Radical Numerics is betting that the model has to understand the system, and that the builders of such a model have to defend against what it enables.