Profluent
Last reviewed
Jun 8, 2026
Sources
14 citations
Review status
Source-backed
Revision
v1 · 1,671 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 8, 2026
Sources
14 citations
Review status
Source-backed
Revision
v1 · 1,671 words
Add missing citations, update stale details, or suggest a clearer explanation.
Profluent (legally Profluent Bio Inc.) is an American artificial intelligence company that uses large language models and generative AI to design novel proteins and genome editors. Headquartered in Berkeley, California, the company trains frontier-scale models on massive corpora of protein-sequence and biological data, aiming to move biology "from reading to writing" by generating molecules with specified functions on demand. [1][2] Profluent describes itself as an "AI-first" protein design company building a foundational layer for programmable biology, with applications spanning therapeutics, gene editing, biomanufacturing, and agriculture. [3]
The company is best known for OpenCRISPR-1, which it released in 2024 as the first CRISPR gene editor designed entirely by AI and made openly available for ethical research and commercial use. [4] Profluent's models draw on its proprietary Protein Atlas, a curated dataset the company reports contains more than 115 billion protein sequences, described as one of the largest such datasets ever assembled. [2][3]
Profluent has raised roughly $150 million in venture funding as of late 2025, including a $106 million Series B round co-led by Altimeter Capital and Jeff Bezos's Bezos Expeditions, announced in November 2025. [1][2] It operates in the emerging field of AI-driven biological design alongside companies such as Generate Biomedicines, EvolutionaryScale, and Cradle. [5]
Profluent was founded in 2022 by Ali Madani and Alexander Meeske. [6][7] Madani serves as chief executive officer. Before starting the company, he was a research scientist at Salesforce Research, where he led ProGen, an early protein language model that demonstrated large language models could generate functional protein sequences from scratch. [6][8] The ProGen work, which trained a transformer-based model on roughly 280 million protein sequences spanning more than 19,000 protein families and conditioned generation on control tags describing protein properties, was published in Nature Biotechnology in January 2023 and showed that AI-generated artificial proteins could match or exceed the activity of natural variants. [8] This result is widely cited as the first demonstration that language models can produce functional proteins, and it forms the scientific lineage behind Profluent's platform. [2][6]
Co-founder Alexander Meeske is an assistant professor of microbiology at the University of Washington, bringing CRISPR and microbial biology expertise to the company. [7] The leadership team also includes Hilary Eaton as chief business officer. [4] Profluent emerged from the broader effort to apply protein design and generative AI to drug discovery, with the goal of designing biological molecules deliberately rather than discovering them by chance. [5]
In April 2024, Profluent announced OpenCRISPR-1, which it described as the world's first AI-created and openly released gene editor. [4] OpenCRISPR-1 consists of a Cas9-like protein and a guide RNA, both generated using Profluent's protein language models rather than copied from any natural organism. The editor maintains the prototypical architecture of a Type II Cas9 nuclease but differs from SpCas9 and other known natural CRISPR-associated proteins by hundreds of mutations. [9] To create it, Profluent assembled the CRISPR-Cas Atlas, a curated dataset of more than one million CRISPR operons drawn from 26 terabases of genomes and metagenomes, and used it to train models that generated millions of diverse, novel CRISPR-like proteins. [9]
Profluent released OpenCRISPR-1 openly, making both the editor's sequence and the CRISPR-Cas Atlas available for ethical research and commercial licensing. [4] The company reports that tens of thousands of academic and industry researchers across drug discovery, agriculture, and crop development have since accessed it. [9] Independent coverage noted the open release as a notable departure from the heavily patented CRISPR landscape. [4]
On July 30, 2025, Profluent published its genome-editing research in the journal Nature, in a paper titled "Design of highly functional genome editors by modelling CRISPR-Cas sequences." [9][10] The work reported that the AI-designed editors achieved improved activity, specificity, and immunogenicity profiles relative to naturally occurring and previously engineered systems, and that the models generated roughly 4.8 times the number of protein clusters found across natural CRISPR families. [9] Profluent has also released related tools, including Protein2PAM, a model that makes gene editors programmable for recognition of specific DNA motifs, and proseLM, a structure-conditioned protein design model, and has stated plans to extend its approach to base editors, prime editors, and large-insertion gene-editing techniques. [9]
In April 2025, Profluent introduced ProGen3, a family of generative protein language models presented as evidence that scaling laws apply to biological design in the same way they govern large language models. [11][12] ProGen3 spans configurations from 339 million to 46 billion parameters, with the largest variant, ProGen3-46B, trained on 1.5 trillion tokens as a compute-optimal model. [11][13] Training used the Profluent Protein Atlas v1, which the company describes as comprising about 3.4 billion full-length proteins and roughly 1.1 trillion amino-acid tokens. [13] For comparison, coverage noted this is an order of magnitude more sequence data than the structural dataset behind Google DeepMind's AlphaFold 3. [12]
Profluent reported that larger models generated higher-fitness proteins across a wider diversity of protein families, produced fewer invalid or repetitive sequences, and responded better to alignment against laboratory data. [11][13] The company stated that ProGen3-46B produced 59 percent more sequence diversity than its 3-billion-parameter model and 198 percent more than its 339-million-parameter model. [13] The findings were released as a scientific preprint in April 2025 and the work was subsequently recognized as a spotlight at the NeurIPS machine learning conference in 2025. [2][12]
Alongside ProGen3, Profluent announced OpenAntibodies, an initiative to generate novel antibodies that the company says are as effective as commercial versions while being structurally distinct enough to avoid existing patents, with plans to release a set of antibody DNA "recipes" royalty-free or under single upfront licensing fees. [12] The company has also described E1, a retrieval-augmented model for protein engineering. [5]
Profluent has raised approximately $150 million in disclosed venture funding across three rounds. [1][2] The figures below are drawn from company announcements and press coverage; some early-round details vary by source and are attributed accordingly.
| Round | Date | Amount | Lead investor(s) | Selected participants |
|---|---|---|---|---|
| Seed | 2023 | ~$9 million [7] | Insight Partners | Air Street Capital, AIX Ventures, Convergent Ventures |
| Series A | March 2024 | $35 million [6][14] | Spark Capital | Insight Partners, Air Street Capital, plus angels including Google DeepMind chief scientist Jeff Dean |
| Series B | November 2025 | $106 million [1][2] | Altimeter Capital, Bezos Expeditions | Spark Capital, Insight Partners, Air Street Capital |
The Series A, announced on March 25, 2024, was led by Spark Capital and notably included a syndicate of angel investors from companies such as OpenAI, Salesforce, and Google, among them Jeff Dean. [6][14] The $106 million Series B, announced on November 19, 2025, was co-led by Altimeter Capital and Bezos Expeditions, the personal investment firm of Amazon founder Jeff Bezos, and brought the company's total funding to about $150 million. [1][2] Profluent said it would use the capital to scale its frontier AI models and expand its platform beyond gene editing into antibodies, antigens, and enzymes. [1][3] Altimeter's Jamin Ball described programmable biology as opening "the largest design space in nature" and reaching "multitrillion-dollar markets." [3] Profluent's valuation has not been publicly disclosed. [12]
Profluent operates within the rapidly growing field of generative AI for biology, where companies and research groups apply foundation models to design proteins, antibodies, and gene editors rather than relying on natural discovery. Reported competitors and peers include Generate Biomedicines, EvolutionaryScale (founded by former Meta researchers behind the ESM protein language models), Cradle, Isomorphic Labs (an Alphabet subsidiary), Ginkgo Bioworks, and Evozyne. [5][12] The Arc Institute's Evo genomic language models and similar biological foundation models occupy adjacent territory in AI drug discovery. [5]
Profluent's significance lies in being among the first organizations to show that protein language models can produce functional molecules, and then to extend that capability to a working, openly released CRISPR gene editor designed by AI. [2][4] OpenCRISPR-1 is frequently cited as a milestone because gene editors had previously been discovered in nature and adapted, whereas Profluent generated a novel editor computationally and released it without the restrictive licensing common to CRISPR technology. [4][9] The company's ProGen3 scaling work further argued that biology, like natural language, exhibits predictable returns to model and data scale, suggesting that larger biological foundation models may unlock broader and more controllable design of proteins for medicine, agriculture, and industry. [11][13] To support these applications, Profluent has announced commercial partnerships with Corteva Agriscience in crop innovation, Revvity in base-editing systems, and Integrated DNA Technologies in enzyme co-design. [5]