AI for Bio · Model

Chai-3

Chai Discovery's closed next-generation molecular design model for antibodies and multispecifics

ProteinsAI-native biotechUnited StatesProduct only, no public artifact中文

Overview

NameChai-3 · aliases: Chai-3
Full nameNo separate expanded name
Ambiguous names (not auto-linked)Chai
FamilyNot a family entry. Single-model entry, not a family
OrganizationChai Discovery
Organization typeAI-native biotech
CountryUnited States
Model classClosed molecular design model
DomainProteins (also: Antibodies)
First release† partly verified
Basis: Licence announcement on the company news page, the first public disclosure of Chai-3; no standalone technical report
Note: A Business Wire URL id suggests 2026-06-02, which is not a publication date source
Other milestones
  • Official announcement: † · Licence announcement on the company news page source
Latest update · product update
Basis: Series C announcement calling Chai-3 the latest model (company post)
A company funding announcement calls it the latest model; a company event, not a model version
Note: The 2026-07-13 collaboration and later pharma deals are not counted as model updates source
The most recent sourced event, not a claim that nothing changed since.
StatusProduct only, no public artifact · No public artifact; only partnerships and company announcements
OpennessCode: Closed
Weights: Closed
Closed; early access for partner pharma companies only
Benchmarks reported0 counted · none of the 44
Wet-lab validationNone recorded
Unverified fields† first_release
As of2026-11-16

What it does

Closed successor to Chai-2; the company positions it for antibody and multispecific design and harder targets. No technical report, no public benchmark, no public data.

Benchmarks reported

None of the 44 benchmarks. This model is on the zero-pairing list.

How it evaluates itself: Company statements; Pharma partnerships

Other evaluations (outside the 44)

  • Company statements (none disclosed) Self-reported: Press releases claim advantages over the predecessor on binding, hard targets, multispecifics and developability; no report or data
  • Pharma partnerships (user or partner report) partner: Several licences and collaborations announced; no partner results published

Developers report: Developers report advantages over the predecessor on antibody design and hard targets with roughly double the success rate; no quantified public evidence

Independent evidence: No independent evidence found. No independent evidence found

Training data

Group in Figure 7: Private or licensed data.

ClassValueDetail
a. Public databasesNot disclosed
b. Private or licensed dataYesOne pharma deal includes a custom model trained on the partner's proprietary data
c. Own wet-lab data used for trainingNot disclosed
d. Wet lab used only for validationNot disclosedThe company says wet-lab validation exists; details are not public

Wet-lab validation

None. The developers report no wet-lab validation.

Background

  • Organization: Chai Discovery was founded in San Francisco in 2024 as an AI-native molecular design platform
  • Funding: Closed a $400M Series C in 2026-07 at a $3.8B valuation, led by Index Ventures
  • Products and partnerships: Licences or collaborations announced with several large pharma companies; financial terms undisclosed
  • Related models: Chai-1, Chai-2

Reading notes

  1. Press-reported success-rate figures conflict and no primary data exists
  2. Funding and partnership news are not model updates; this record follows that rule

Caveats

  1. The first-release date rests on the company news page; partly verified
  2. No public benchmark results; on the zero-pairing list under this round's rules

Sources

  1. Official page businesswire.com (company blog)
  2. Official page businesswire.com (company blog)

Corrections

2 corrections
FieldWasNowBasisFound
first_release2026-06-022026-06-04 (company news page date; a URL id is not a date source)source2026-11-16
latest_updateFunding and collaboration described as a model updateMarked as a company event, not a model version; later pharma deals excludedsource2026-11-16

Part of the AI for Bio benchmarks survey · all 112 models · open in the explorer. Record as of 2026-11-16.