Chai-2 designs full-length antibodies with 16–20% hit rates across 52 targets. About 100x better than prior methods, compressing antibody discovery from months to days.
DeployStage 4 of 5
16–20% hit rates across 52 clinical targets. Series C July 2026; commercial partnerships with major pharma.
Updated Jul 2026·Checked 10 Oct·0 updates this week
Milestones
Next · First clinical trial data expected
Chai Discovery founded2024Complete.
Seed funding and Chai-1 releaseSep 2024Complete.
Series A fundingAug 2025Complete.
Chai-2 results across 52 targets publishedJun 2025Complete.
Translation to in-vivo efficacyAntibodies that bind a target in vitro may not work in living organisms due to off-target effects, immunogenicity or poor pharmacokinetics.
Manufacturability screeningSome AI-designed sequences may be difficult to express in mammalian or microbial cell lines, or may aggregate during purification.
Physics limits
Hit rate is not potency16–20% of designed antibodies bind the target, but binding strength and specificity vary widely. The best candidates must still be engineered.
Immunogenicity is unpredictableEven strong, specific antibodies may trigger immune responses in patients, limiting therapeutic utility. Predicting this computationally remains open.
Difficult targets resist designSome proteins lack accessible epitopes or have high structural flexibility. These remain hard targets even for AI methods.
How it works
4 parts
Training data
Learning from nature's antibodies
Chai-2 trains on sequences of antibodies that occur naturally in humans and other animals, learning the rules of how protein structure enables binding.
Target encoding
Understanding what to bind
Given a protein target (like a tumor antigen or virus spike), the model encodes its 3D structure and surface chemistry in a way the design module understands.
Generation
Designing novel sequences
The model generates antibody sequences optimized for binding while maintaining the structural features needed for stability and expression in cells.
Scoring
Ranking designs by confidence
Designs are ranked by the model's confidence in binding affinity. Top candidates are synthesized and tested experimentally.
AI antibody design, protein engineering, zero-shot design
Chai Discovery uses AI to design antibodies from scratch. Chai-2 achieves 16–20% hit rates across diverse targets, roughly 100x better than prior methods. Founded 2024; $400M Series C at $3.8B valuation in July 2026.
Chai-2 designs full-length antibodies with 16–20% hit rates across 52 clinical targets, ~100x improvement over prior methods.
July 2026 Series C of $400M at $3.8B valuation accelerates commercial deployment with major pharma partners.
Bypasses traditional phage display, replacing months of lab work with days of computation.
