Chai Discovery

Chai Discovery builds AI models to predict and reprogram molecular interactions, helping scientists design biomolecules with specified properties. Its product generations span Chai-1 molecular structure prediction, Chai-2 generative antibody design and Chai-3, included in its July 2026 Novartis collaboration. The company’s stated ambition is a “computer aided design suite” for molecules.[3][4][10][8]

Founded in 2024 by Joshua Meier, Jack Dent, Matthew McPartlon and Jacques Boitreaud, Chai Discovery is based in San Francisco. Meier is its cofounder and CEO. It is distinct from Chai AI, the company separately covered by Latent Space.[10][1]

On July 14, 2026, Chai announced $400 million in Series C funding at a $3.8 billion valuation, naming Index Ventures, Kleiner Perkins, Sequoia Capital and Dimension among its investors. It said Eli Lilly, Pfizer and Novartis had begun adopting its models across their organizations. The funding was intended to expand compute, data and research, accelerate product development and bring the technology to more scientists.[9]

Chai’s earlier work provides the structure-prediction context for its later generative design models. On September 9, 2024, it released Chai-1, a multimodal foundation model covering proteins, small molecules, DNA, RNA and covalent modifications. The launch offered a free web interface, including for commercial drug discovery.[3][4]

Chai-1 could run without multiple sequence alignments and accept additional information, including laboratory-derived restraints, to guide predictions. Chai reported that supplying even a handful of contacts or pocket residues doubled antibody–antigen structure-prediction accuracy. This made experimental information an input to the model rather than only a way to evaluate its output.[3]

Chai reported a 77% success rate on PoseBusters, compared with AlphaFold3’s 76%; a CASP15 protein-monomer Cα LDDT score of 0.849, compared with ESM3-98B’s 0.801; and a single-sequence multimer DockQ acceptable-prediction rate of 69.8%, compared with MSA-based AlphaFold-Multimer’s 67.7%. These were comparisons on distinct structure-prediction tasks, rather than measures of therapeutic success.[3]

Alongside Chai-1, the company announced approximately $30 million raised in seed financing, led by Thrive Capital, OpenAI and Dimension Capital. Participants included Conviction, Neo, Lachy Groom and Amplify.[6]

Released on June 30, 2025, Chai-2 integrated all-atom structure prediction with generative modeling to create novel, epitope-specific binders. Supported formats included scFv antibodies, VHH nanobodies and miniproteins. Chai opened access to selected academic and industry partners under its Responsible Deployment policy.[4]

For its launch experiments, Chai designed up to 20 antibodies or nanobodies per target across 52 protein targets lacking existing antibody binders in SAbDab. The company reported a 16% binding rate in one experimental round, with at least one successful binder for 50% of targets. Separately, it reported a 68% wet-lab hit rate for miniprotein binder design. These binding results did not establish clinical efficacy.[4]

On August 6, 2025, Chai announced $70 million in additional Series A funding led by Menlo Ventures, including its Anthology Fund, a partnership with Anthropic. The announcement named former Pfizer chief scientific officer Mikael Dolsten as joining the board. Chai planned to use the funding to develop its platform, address previously inaccessible targets and onboard selected partners.[7]

A November 19, 2025 research update extended Chai-2’s work to full-length monoclonal antibodies and properties beyond binding. Chai tested 88 designed IgGs across 28 antigens against four developability criteria: 86% had zero or one flagged issue, and 24 of the 28 antigens yielded at least one design with a clean profile. In a separate structural-validation panel, cryo-EM examination of five antibody–antigen complexes showed binding at the intended epitopes and, according to Chai, sub-angstrom accuracy in the difficult CDR loops.[5]

The same update reported binders to all six tested GPCR targets after testing 10–73 designs per target, including agonists for two GPCRs. A separate experiment tested 27–50 designs against each of three peptide-MHC complexes and produced two hits against a complex containing the KRAS G12V peptide. Those binders did not bind G12D or wild-type KRAS peptides, providing a concrete example of the reported designs’ specificity.[5]

On December 15, 2025, Chai announced a $130 million Series B raised from General Catalyst and Oak HC/FT, with participation from existing investors and new investors Emerson Collective and Glade Brook. Annie Lamont of Oak HC/FT and Hemant Taneja of General Catalyst were announced as prospective board additions. Chai designated frontier research and team growth as uses of the funding.[8]

The July 13, 2026 Novartis collaboration gave Novartis access to Chai’s latest models, including Chai-3, for antibody discovery across multiple therapeutic targets. It expanded more than a year of technical engagement, including early access to a next-generation folding model described in the announcement as occurring “last spring.” The companies extended their work toward broader deployment across therapeutic programs; financial terms were undisclosed.[10]

On July 15, 2026, Chai announced a collaboration giving argenx early access to its platform for de novo antibody discovery across therapeutic targets. Chai described Chai-3 as improving therapeutic binding, difficult-target design, multispecific design, developability and generalization, without supplying quantitative comparisons in that announcement.[11]

These partnerships also informed product and research priorities. In the August 11, 2026 Latent Space episode editorial, product lead Neil Patil explained that Chai could conduct “informed research based on what our partners have just been organically asking us for help with.” RJ Honicky described the resulting molecule editor as closer to CAD or graphics-design software than a chatbot.[1]