AI-Unicorn Chai Discovery and Eli Lilly to Train Custom AI Models for Biologics Design
San Francisco-based Chai Discovery has partnered with Eli Lilly to integrate Chai’s molecular design platform into Lilly’s discovery workflows and develop a bespoke AI model trained on Lilly’s proprietary data, exclusively customized to Lilly’s targets and internal datasets, intended to streamline early-stage biologics programs.
This comes as Eli Lilly extends TuneLab, its federated learning platform, via a Benchling integration, giving 1,300+ biotechs access to small-molecule and antibody prediction models trained on $1B+ in proprietary molecular and safety data, with a parallel Lilly-Revvity setup using Signals Xynthetica to run TuneLab at scale with secure data exchange and anonymized result-sharing.
The agreement was announced shortly after Chai’s $130 million Series B funding round in December 2025, which valued the company at $1.3 billion. The company is backed by OpenAI, Oak HC/FT, General Catalyst, Menlo Ventures, Thrive Capital, and Dimension. Chai’s founding team includes researchers from OpenAI, Meta FAIR, Google X, and Stripe.
Chai will provide Lilly with access to its generative antibody design models, including Chai-2, which reportedly achieves double-digit experimental hit rates in zero-shot design settings. The collaboration follows prior internal evaluation of Chai’s designs by Lilly. No financial terms were disclosed.
Chai-2 model can design full-length monoclonal antibodies with high structural accuracy, going beyond earlier tools that only handled antibody fragments or predicted binding. In testing, most of its designs met common preclinical quality standards, and nearly all target proteins produced at least one promising antibody candidate. Cryo-EM studies confirmed that the antibodies bind exactly where predicted, even in complex regions. Chai-2 also generated antibodies for difficult targets like GPCRs and KRAS G12V with relatively few design attempts. The company aims to eventually design drug-ready antibodies in a single computational step and limits access to its platform under a policy focused on responsible use.
Topic: Industry Movers
