Chai Discovery

Chai Discovery is an American AI biotechnology company that builds foundation models for predicting and designing the three-dimensional structures of biomolecules to accelerate drug discovery. Founded in early 2024 and headquartered in San Francisco, it is best known for two systems: Chai-1, an open biomolecular structure-prediction model released in September 2024 and positioned as an openly distributed counterpart to Google DeepMind's AlphaFold 3, and Chai-2, a 2025 system that designs novel antibodies and proteins from scratch.^[1][2] In June 2026 it disclosed a proprietary third-generation design model, Chai-3, as part of a license agreement with Pfizer.^[20] Backed by OpenAI and Thrive Capital among others, Chai Discovery raised a $130 million Series B in December 2025 at a $1.3 billion valuation,^[1][3] followed in July 2026 by a $400 million Series C at a $3.8 billion valuation led by Index Ventures.^[24][25] During 2026 it announced agreements with Eli Lilly, Pfizer, Novartis, argenx, Bristol Myers Squibb and GSK, and Endpoints News counted GSK, announced in October 2026, as the company's sixth pharmaceutical partner in nine months.^{[18][20][23][28][29][30][31]} It is one of the higher-valued companies in the post-AlphaFold-3 wave of AI-native drug-discovery firms.

What is Chai Discovery?

Chai Discovery describes itself as an applied AI research lab working to "reprogram the interactions between biochemical molecules," with the stated long-term ambition of building a "computer-aided design suite" for molecules analogous to the CAD tools used in mechanical and electronic engineering.^[1][3] In practice this means training neural networks that take molecular sequences and chemical specifications as input and output atomic-resolution predictions of how proteins fold and how they bind to other proteins, small-molecule drugs, antibodies, and nucleic acids. The company's structure-prediction work sits in the lineage opened by AlphaFold and generalized by AlphaFold 3, while its newer design work moves from predicting existing structures to generating entirely new molecules computationally, in particular custom antibodies built from scratch ("de novo") against a chosen disease target.^[2][4]

The company pursues a hybrid commercial model. Its first model, Chai-1, was released openly (for non-commercial use at first, then under the Apache 2.0 license from November 2024), building a large academic and industry user base, while customers pay an annual access fee for its design software, and some large partners also commission custom models trained on their own proprietary data.^{[2][13][18][20]} Chief executive Joshua Meier has framed the underlying bet in language terms: "The question was, if these AI models can understand natural language processing, why can't they understand the real natural language, like DNA and proteins?"^[4] The following table summarizes the company at a glance.

Who founded Chai Discovery and when?

Chai Discovery was founded in March 2024, according to Contrary Research, by Joshua Meier, Jack Dent, Matthew McPartlon, and Jacques Boitreaud.^[2][23] The company emerged from stealth on 9 September 2024, the same day it announced its seed funding and published Chai-1.^[7][10]

Chief executive Joshua Meier is an AI researcher with a background spanning machine learning and biology. After studying chemistry and computer science at Harvard, he worked as a researcher at OpenAI during the GPT-1 and GPT-2 era, then joined Meta's generative-biology group, where he helped develop ESM1, an early transformer protein-language model and a precursor to the ESM family whose embeddings later powered Chai-1's alignment-free prediction mode.^[4][2][12] Before founding Chai, Meier spent three years at the biotechnology company Absci, where he became Chief AI Officer in 2023 and oversaw the company's AI efforts.^[2][4] President Jack Dent is a Harvard computer-science classmate of Meier who previously worked at Stripe, where he helped launch Stripe Link and Stripe Capital.^[2] Matthew McPartlon, the company's chief technology officer, worked with Meier at Absci as tech lead for de novo antibody-design modeling and later worked on protein-protein interaction models at Proxima (VantAI); Jacques Boitreaud studied RNA bioinformatics at McGill University before working as an AI scientist at the small-molecule company Aqemia.^[2][4]

According to a January 2026 TechCrunch account of the company's origins, the idea traced back about six years. Meier had worked at OpenAI in 2018; after he left, OpenAI chief executive Sam Altman messaged Dent to ask whether Meier would be open to working on a proteomics spinout, but Meier judged the technology premature at the time and went to work at Facebook.^[5] After Meier's years at Meta and Absci and Dent's time at Stripe, the two told Altman in 2024 that they were starting Chai together, and the four founders built the company while working out of OpenAI's offices in San Francisco's Mission neighborhood, with OpenAI becoming one of its first seed investors.^[5] Dent told TechCrunch that "every line of code in our codebase is homegrown," reflecting a decision to build custom architectures for molecular data rather than fine-tune off-the-shelf language models.^[5]

What are Chai Discovery's models?

What is Chai-1?

Chai-1 is a multimodal foundation model for biomolecular structure prediction, released on 9 September 2024.^[7][11] It predicts atomic-resolution 3D structures for proteins, small-molecule ligands, DNA, RNA, multimeric complexes, covalent modifications, and antibody-antigen interfaces.^[11][12] On the PoseBusters protein-ligand evaluation set, given only the protein sequence and the ligand's chemical composition, it reaches a 77% success rate, comparable to the 76% computed from AlphaFold 3's publicly released predictions, rising to 81% when the model is prompted with the apo (unbound) protein structure.^[12] On the CASP15 protein monomer set the company reported an average C-alpha LDDT of 0.849, against 0.801 reported for the largest 98-billion-parameter version of ESM3.^[11][12] A distinctive feature is an alignment-free ("MSA-free") inference mode in which residue-level embeddings from a large protein language model substitute for multiple sequence alignments, which the technical report says preserves most of the model's performance on single sequences alone.^[12] Chai-1 can also be prompted with experimental restraints, such as epitope residues or cross-linking contacts derived from lab data, which the report says boosts accuracy by double-digit percentage points.^[12]

The model was initially released with weights and inference code published on GitHub under a non-commercial community license, while a free web server could be used for commercial drug discovery.^[12][13] In late November 2024 the company replaced that license with the permissive Apache 2.0 license, covering both code and model weights and permitting broad commercial use.^[13] Chai-1 appeared about two months before the academic Boltz-1 model, which was announced in November 2024.^[2] As of April 2026, Contrary Research counted more than 400 academic citations of Chai-1 and about 1,900 GitHub stars, and reported that David Baker's laboratory at the University of Washington had used it in enzyme-engineering work.^[2]

What is Chai-2?

Chai-2, unveiled on 30 June 2025, extends the company's work from structure prediction into molecular design.^[8][2] It is a system for zero-shot de novo design of antibodies and nanobodies, meaning it generates entirely novel binder sequences against a specified protein target and epitope without requiring prior binders or task-specific fine-tuning.^[8] In a technical report accompanying the launch (also posted to bioRxiv as "Zero-shot antibody design in a 24-well plate"), the company reported that Chai-2 achieved roughly a 16% experimental ("wet-lab") hit rate when designing 20 or fewer antibodies or nanobodies per target across 52 diverse targets, none of which had an existing antibody or nanobody binder in the Protein Data Bank. It characterized this as a more than hundredfold improvement over previous computational methods, whose success rates were on the order of 0.1%.^[8][2][17] It said that in a single round of testing the system found at least one binder for about half of the targets (26 of 52), with the cycle from AI design to wet-lab validation completed in under two weeks.^[8] In a related miniprotein-binder design task, the company reported a 68% wet-lab hit rate, often with picomolar binding affinities.^[8] Chai-2 was first opened to selected academic and industry partners through an early-access program governed by what the company called its Responsible Deployment policy, and it is offered to pharmaceutical partners under paid access arrangements rather than as an open release.^[36][2]

In November 2025 the company published a follow-up study, "Drug-like antibody design against challenging targets with atomic precision," applying Chai-2 to full-length monoclonal antibodies rather than the antibody fragments used in most earlier design work.^[14][15] Of 88 designed IgG antibodies spanning 28 target antigens, 86% showed zero or one flagged issue across four developability criteria, and 24 of the 28 antigens yielded at least one design with a clean developability profile.^[14] Cryo-electron microscopy of five designed antibody-antigen complexes showed each antibody binding where the model intended, with sub-angstrom accuracy reported even in the CDR loops.^[14] The study also reported at least one binder to each of six G protein-coupled receptors (GPCRs) after testing 10 to 73 designs per target, including agonists for two of the receptors, and two binders to a peptide-MHC complex presenting a KRAS G12V peptide that did not bind G12D or wild-type KRAS peptides.^[14][15]

What is Chai-3?

Chai-3 is the company's third-generation model. It was first disclosed in June 2026 in the announcement of the Pfizer license agreement, which called it Chai's "previously undisclosed model."^[20][21] Chai said Chai-3 doubles the success rate of its predecessor in antibody design and produces antibodies that meet required therapeutic standards, with gains in therapeutic binding, multispecific molecules, hard-to-drug targets, and generalization.^[20] When it announced its Series C a month later, the company described Chai-3 as its latest model and said it was "unlocking targets that have long resisted traditional computational and laboratory methods."^[24] These performance figures are the company's own statements. Under their agreements, Pfizer gained early access to Chai-3, and Novartis gained access to Chai's latest models including Chai-3.^[20][23]

Who are Chai Discovery's pharmaceutical partners?

Chai Discovery earns revenue mainly by licensing access to its models and platform to drug developers, according to Contrary Research, rather than by developing its own drug candidates.^[2] In June 2025 the UK government named Chai among the companies invited to discussions, including a roundtable at 10 Downing Street, about co-investing in OpenBind, a consortium based at Diamond Light Source that aims to generate a large dataset of drug-protein interactions for training AI models.^[32] From January 2026 the company announced a run of agreements with large drugmakers. Endpoints News described the first, with Eli Lilly, as one of the pharmaceutical industry's largest AI software deals.^[19]

The Lilly collaboration followed a period in which Lilly evaluated a set of Chai's designs.^[18] Contrary Research reported in April 2026 that Lilly pays an annual access fee in the mid-eight figures under that agreement.^[2] The TuneLab arrangement, announced in June 2026, was Chai's second agreement with Lilly; Aliza Apple, global head of Lilly TuneLab, described it as bringing Chai's "frontier miniprotein design models" to the program's participants.^[22] The Novartis agreement built on more than a year of technical work between the two companies, including early access to Chai's next-generation folding model, and its financial terms were not disclosed; the Pfizer terms were also undisclosed.^[23][21] In the GSK evaluation, Chai's models generated designs to GSK targets zero-shot, without target-specific training, and GSK's internal wet-lab testing found binders to all tested targets.^[30] Christopher Austin, GSK's senior vice president of R&D technologies, said GSK "saw designs with strong binding affinities" and that Chai's technology complements GSK's own internal models.^[30]

How is Chai Discovery funded?

Chai Discovery has raised about $630 million across four disclosed rounds, according to SiliconANGLE; Fierce Biotech put the total above $600 million.^[26][25] The company's seed round of about $30 million was announced on 9 September 2024. Chai said it was led by Thrive Capital, OpenAI, and Dimension, with participation from Conviction, Neo, Lachy Groom, and Amplify, and included angel investors such as Greg Brockman, Blake Byers, and Fred Ehrsam.^[10][7] In August 2025 the company announced a $70 million Series A led by Menlo Ventures, including investment from its Anthology Fund (a partnership with Anthropic), with new investors including Yosemite, DST Global Partners, SV Angel, Avenir, and DCVC; Contrary Research reported that the round valued the company at $550 million.^[17][2][9] As part of that round, former Pfizer Chief Scientific Officer Mikael Dolsten joined the board.^[17][9] When Endpoints News named Chai to its 2025 "Endpoints 11" list in September 2025, it listed the company's total funding as $100 million.^[33]

The Series B, announced on 15 December 2025, was a $130 million round at a $1.3 billion valuation.^[1][3] Press coverage and the company's own announcement state the round was co-led by General Catalyst and Oak HC/FT, with participation from Menlo Ventures, OpenAI, Thrive Capital, Dimension, Neo, the Yosemite venture fund, Lachy Groom, SV Angel, and new investors Glade Brook and Emerson Collective.^[1][3][16] As part of the round, Oak HC/FT co-founder Annie Lamont and General Catalyst chief executive Hemant Taneja joined the company's board.^[1][3] The funding round raised the company's total capital to more than $225 million, and Chai said it would use the proceeds to accelerate research and product development and grow its commercialization efforts toward creating the equivalent of a computer-aided design suite for molecules.^[1][4]

The Series C, announced on 14 July 2026, raised $400 million at a $3.8 billion valuation, nearly triple the Series B valuation set seven months earlier.^[24][26] Index Ventures led the round alongside Kleiner Perkins, Sequoia Capital, and Dimension.^[25][27] New investors included Bain Capital Ventures, Battery Ventures, Baillie Gifford, BDT & MSD, Sapphire Ventures, and Avra Capital, while existing backers including Thrive Capital, OpenAI, Oak HC/FT, Menlo Ventures, and General Catalyst also took part.^[26][27] Dimension founder Zavain Dar called it "the largest check we've ever written at Dimension."^[34] Chai said the money would go toward the compute, data, and research behind its models and toward product development.^[24] The table below summarizes the disclosed financing history.

Why does Chai Discovery matter, and who are its competitors?

Chai Discovery is one of the most prominent companies in the wave of AI-native, foundation-model-driven drug-discovery firms that followed the May 2024 announcement of AlphaFold 3. Two features distinguish its trajectory. First, its decision to release Chai-1 openly, and later under a permissive license, contrasted with the initial release of AlphaFold 3 as a restricted web server, and helped Chai-1 become a widely used research tool.^[2] Second, with Chai-2 the company moved beyond predicting existing structures to designing new molecules, positioning itself in the emerging market for computational antibody and protein design.^[8][4] CEO Joshua Meier has summarized the company's framing this way: "We want to make biology look less like science and look more like engineering and then usher in as much progress as we can on drug discovery."^[4]

The competitive landscape is crowded. Isomorphic Labs, the DeepMind spinout led by Demis Hassabis, pursues a similar mission with substantially larger funding and pharmaceutical partnerships. EvolutionaryScale, founded by former Meta protein-AI researchers and the developer of the ESM3 model, worked on generative protein models until the Chan Zuckerberg Initiative acquired its team in November 2025.^[2] In the antibody- and protein-design segment, Chai competes with firms such as Nabla Bio, Xaira Therapeutics (co-founded by Nobel laureate David Baker), and Absci, Meier's former employer, as well as with open structure-prediction and design models such as Boltz and Boltz-2.^[2] Both the prediction and design problems Chai works on sit at the center of contemporary AI drug discovery, and the breadth of well-funded competitors reflects the perceived size of the opportunity to compress the timelines and costs of finding new medicines.^[1][2]

Observers have also pointed to risks. TechCrunch noted in January 2026 that some industry veterans doubt such tools will have a major impact on drug development, while General Catalyst managing director Elena Viboch, whose firm co-led the Series B, said the drugmakers that moved fastest to partner with companies like Chai would be the first to get molecules into the clinic, which in practice meant partnering in 2026 and seeing first-in-class medicines enter clinical trials by the end of 2027.^[5] Contrary Research's April 2026 report argued that drug-discovery platform companies have historically captured less value than companies developing their own drug assets, that Chai's $1.3 billion Series B valuation rested on its platform rather than on drug assets, and that rivals training on the same public structure databases could erode any technical lead. It also noted that Chai had chosen not to build significant internal wet-lab capacity, leaving it reliant on external data, and that Isomorphic Labs' proprietary IsoDDE engine, announced in February 2026, reported large accuracy gains over Chai-1 on protein-ligand structure prediction.^[2]

Debarghya Das's "The Ultimate Neolab List," a 2026 list of AI-lab startups, includes Chai Discovery, summarizing it as "Researcher-founded, models-first, compute-heavy, raised a lot early."^[35]

See also

• Chai-1
• Chai-2
• Protein design
• AI in drug discovery
• Isomorphic Labs
• List of neolabs
• Neolab

References

1. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10OpenAI-backed biotech firm Chai Discovery raises $130M Series B at $1.3B valuation, TechCrunch, 15 December 2025.
2. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15 ^16 ^17 ^18 ^19 ^20 ^21 ^22 ^23 ^24 ^25 ^26Chai Discovery Business Breakdown & Founding Story, Contrary Research (updated 23 April 2026).
3. ^1 ^2 ^3 ^4 ^5 ^6 ^7Chai Discovery Announces $130 Million Series B To Transform Molecular Discovery, Business Wire, 14 December 2025.
4. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10Chai's the Limit for AI Antibody Designer After $130M Series B Funding, Genetic Engineering & Biotechnology News, 16 December 2025.
5. ^1 ^2 ^3 ^4Lucas Ropek, From OpenAI's offices to a deal with Eli Lilly: how Chai Discovery became one of the flashiest names in AI drug development, TechCrunch, 16 January 2026.
6. ^Chai infuses AI drug discovery efforts with $130M series B, Fierce Biotech, 15 December 2025.
7. ^1 ^2 ^3Chai Discovery Gets OpenAI, Thrive Capital Backing to Bring AI to Biotech, Bloomberg, 9 September 2024.
8. ^1 ^2 ^3 ^4 ^5 ^6Chai-2: Zero-shot antibody design in a 24-well plate, Chai Discovery Team, bioRxiv, posted 6 July 2025; see also MarkTechPost coverage, 5 July 2025.
9. ^1 ^2 ^3Chai Discovery Announces $70 million Series A To Transform Molecular Design, Business Wire, 6 August 2025; valuation and board detail per Contrary Research.
10. ^1 ^2 ^3Seed: Announcing our first fundraise, Chai Discovery, 9 September 2024.
11. ^1 ^2 ^3 ^4Introducing Chai-1: Decoding the molecular interactions of life, Chai Discovery, 9 September 2024.
12. ^1 ^2 ^3 ^4 ^5 ^6 ^7Chai Discovery Team (J. Boitreaud, J. Dent, M. McPartlon, J. Meier, et al.), Chai-1: Decoding the molecular interactions of life, bioRxiv, posted 11 October 2024 (version 2, 15 October 2024).
13. ^1 ^2 ^3chaidiscovery/chai-lab: "update LICENSE.md (#186)", GitHub commit replacing the Chai Discovery Community License with Apache 2.0, 27 November 2024; current terms in the chai-lab repository.
14. ^1 ^2 ^3 ^4Drug-like antibody design against challenging targets with atomic precision, Chai Discovery, 19 November 2025.
15. ^1 ^2Chai Discovery Team, Drug-like antibody design against challenging targets with atomic precision, bioRxiv, posted 1 December 2025 (version 2, 3 December 2025).
16. ^Series B: Announcing our $130M fundraise, Chai Discovery, 15 December 2025.
17. ^1 ^2 ^3 ^4Series A: Announcing our $70M fundraise, Chai Discovery, 6 August 2025.
18. ^1 ^2 ^3 ^4 ^5Chai Discovery enters AI collaboration with Eli Lilly, The Pharma Letter, 12 January 2026.
19. ^1 ^2Andrew Dunn, Exclusive: With new AI deal, Eli Lilly and Chai Discovery kick off a 'year of deployment' in pharma, Endpoints News, 9 January 2026.
20. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9Chai Discovery Announces License Agreement with Pfizer to Accelerate Drug Discovery with AI, Business Wire, 4 June 2026.
21. ^1 ^2 ^3Nicholas Jacobus, Chai Discovery Enters License Agreement with Pfizer to Accelerate Drug Discovery with AI, Pharmaceutical Executive, 5 June 2026.
22. ^1 ^2Chai Discovery Collaborates with Lilly TuneLab to Offer AI Capabilities to Select Biotechs, Business Wire, June 2026.
23. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8Chai Discovery Announces Collaboration with Novartis to Advance AI-Driven Antibody Discovery, Chai Discovery, 13 July 2026.
24. ^1 ^2 ^3 ^4 ^5 ^6Series C: Announcing our $400M fundraise, Chai Discovery, 14 July 2026.
25. ^1 ^2 ^3 ^4 ^5Nick Paul Taylor, Chai brews up $400M series C to fuel AI molecule models used by Lilly, Novartis and Pfizer, Fierce Biotech, 14 July 2026.
26. ^1 ^2 ^3 ^4 ^5Kyt Dotson, Chai Discovery nabs $400M Series C as AI-designed antibodies reach Big Pharma, SiliconANGLE, 14 July 2026.
27. ^1 ^2 ^3Chai Discovery raises $400 million to advance AI-designed drug discovery, The Pharma Letter, 15 July 2026.
28. ^1 ^2 ^3Chai Discovery Enters Into Collaboration Agreement With argenx to Advance AI-Driven Immunology Breakthroughs, Chai Discovery (Business Wire release), 15 July 2026.
29. ^1 ^2 ^3Chai Discovery Announces Collaboration with Bristol Myers Squibb to Advance AI-Driven Antibody Discovery, Chai Discovery (Business Wire release), 20 August 2026.
30. ^1 ^2 ^3 ^4 ^5Chai Discovery Announces Collaboration with GSK Following Successful Wet-Lab Evaluation, Chai Discovery (Business Wire release), 9 October 2026.
31. ^1 ^2Andrew Dunn, Exclusive: AI startup Chai lands GSK deal, its sixth pharma partner in nine months, Endpoints News, 9 October 2026.
32. ^UK to become world leader in drug discovery as Technology Secretary heads for London Tech Week, UK Department for Science, Innovation and Technology, 9 June 2025.
33. ^Andrew Dunn, The Endpoints 11: OpenAI-backed Chai Discovery's brainiac quartet aims to design drugs on the computer, Endpoints News, 19 September 2025.
34. ^Chai Discovery Secures $400M Series C to Propel AI-Driven Molecular Innovation, SynBioBeta, 21 July 2026.
35. ^Debarghya Das, The Ultimate Neolab List, debarghyadas.com, accessed 27 October 2026.
36. ^1 ^2Introducing Chai-2: Zero-shot antibody design in a 24-well plate, Chai Discovery, 30 June 2025.