By Chai DiscoveryLearning Objectives
- Understand what Chai Discovery's models do
- Understand why de-novo antibody design is a hard, valuable problem
- Evaluate generative molecular design as validated hypothesis generation
What Is Chai Discovery?
Chai Discovery is an AI structural-biology company building foundation models for molecular design. Its Chai-1 model predicts the three-dimensional structures of biomolecules — the shape a protein folds into, which determines what it does — in the lineage of breakthroughs like AlphaFold. Its 2025 Chai-2 model goes further, performing fully de-novo antibody design: generating brand-new antibodies against a chosen target from scratch; Chai describes it as the first zero-shot antibody design platform to reach double-digit experimental hit rates. Its newest model, Chai-3, disclosed in June 2026 as part of a Pfizer license, doubles Chai-2's success rate by Chai's own measure and targets multi-specific antibodies and hard-to-drug targets. Designing antibodies that bind a target well, without starting from an existing one, is one of the hardest and most commercially valuable problems in drug discovery.
Backed by OpenAI and Thrive Capital, and licensed in 2026 by Eli Lilly, Pfizer, Novartis, argenx and Bristol Myers Squibb, Chai offers its models as accessible tools to researchers, positioning itself as infrastructure for antibody and protein discovery rather than a drug-development company that runs its own pipeline. That posture — models others can use — makes it a natural catalog entry alongside Cradle and Latent Labs. As with all generative design in biology, the honest framing is that AI-designed candidates are starting points requiring experimental validation: a designed antibody must be made and tested before it means anything. The promise is scale — dramatically expanding the space of viable molecules a team can explore — not a shortcut around the laboratory.
💡Key Concept
De-novo antibody design: Instead of tweaking a known antibody, Chai-2 and Chai-3 generate entirely new ones against a target from scratch. Doing this well is one of drug discovery's hardest, most valuable problems.
📝Note
Infrastructure, not a pipeline: Chai offers its structure-prediction and antibody-design models as tools for researchers, rather than running its own drug programs — its "product" is the models.
🎯Tip
Visit Chai Discovery: chaidiscovery.com — models offered as accessible tools for protein and antibody discovery.
Pricing
Chai offers access to its models to researchers and organizations, with free and academic access to some capabilities and commercial arrangements for broader use.
- Structure prediction (Chai-1)
- Model access for research
- Community use
- De-novo antibody design (Chai-2)
- Broader usage and support
- Enterprise arrangements
- Early access to Chai-3
- Custom models trained on the partner's own data
- Pfizer, Lilly, Novartis, argenx, BMS
Core Features
Biomolecular Structure Prediction (Chai-1)
Predicts three-dimensional structures of biomolecules, a foundational capability for understanding and designing them.
De-Novo Antibody Design (Chai-2)
Generates novel antibodies against a target from scratch. Chai reports double-digit experimental hit rates, a hard, valuable capability.
Chai-3 (June 2026)
Chai's most advanced model, first disclosed through its Pfizer license. Chai says it doubles Chai-2's success rate and produces antibodies that meet required therapeutic standards, including multi-specific antibodies and molecules against hard-to-drug targets. Early access goes to pharma licensees.
Accessible Models
Offers its models as tools researchers can use, positioning Chai as infrastructure rather than a drug developer.
Frontier Backing and Pharma Partners
Backed by OpenAI and Thrive Capital, with a 400 million dollar Series C in July 2026. In 2026 it signed platform deals with Eli Lilly (January, expanded through TuneLab in June), Pfizer (June, including a custom model trained on Pfizer's data), Novartis and argenx (July) and Bristol Myers Squibb (August).
Strengths
- Two strong capabilities — structure prediction and antibody design
- Tackles a hard problem — de-novo antibody generation
- Accessible infrastructure — models researchers can use
- Frontier backing — OpenAI-supported, well-capitalized
- Big-pharma adoption — 2026 deals with Lilly, Pfizer, Novartis, argenx and BMS
- Peer to accessible design tools — alongside Cradle and Latent Labs
Limitations and Considerations
- Designs need validation — generated candidates must be made and tested
- Early field — de-novo design is advancing but unproven at scale
- Hit rates vary — reported performance is task- and target-dependent
- Infrastructure, not outcomes — it enables discovery, not approvals
- Expertise required — best used by teams who can validate results
Best Use Cases
| Use Case | Why Chai Discovery Fits | Caveat |
|---|---|---|
| Biomolecular structure prediction | Chai-1 predicts 3D structures | A starting point for design |
| Antibody discovery | Chai-2 and Chai-3 design novel antibodies | Candidates require lab testing |
| Expanding the molecule space | Explore far more viable candidates | Hit rates vary by target |
| Research infrastructure | Accessible models to build on | Validation still required |
Key Takeaways
- Chai Discovery builds AI foundation models for molecular design — Chai-1 for structure prediction, and Chai-2 and its successor Chai-3 for de-novo antibody design
- De-novo antibody design — generating new antibodies against a target from scratch — is one of drug discovery's hardest, most valuable problems
- OpenAI-backed and licensed in 2026 by Eli Lilly, Pfizer, Novartis, argenx and Bristol Myers Squibb, Chai offers its models as tools, positioning as infrastructure
- AI-designed candidates are starting points that require experimental validation; the promise is scale, not a shortcut past the lab
- It is best for teams doing structure prediction and antibody discovery who can validate the model's designs
