Chai Discovery Partners With BMS to Generate Antibodies From Scratch

Chai Discovery announced on August 20, 2026 a collaboration with Bristol Myers Squibb to apply its generative AI platform to therapeutic antibody discovery across BMS's portfolio. The deal adds another major pharma name to Chai's rapidly expanding partner list and reflects a broader industry shift toward foundation model driven molecular design.

Financial terms and specific therapeutic targets were not disclosed. Under the agreement, BMS will integrate Chai's molecular folding and design models, including the Chai 3 system, into its research workflow as part of what the drugmaker describes as an AI powered, continuously learning discovery system.

What Chai brings

San Francisco based Chai Discovery, founded in 2024 by alumni from OpenAI, Meta FAIR, Stripe, and Absci, positions its platform as a computer aided design suite for biology. Rather than screening vast compound libraries experimentally, Chai's models predict and reprogram molecular interactions to generate targeted biomolecules from defined criteria.

The company's Chai 3 architecture uses multimodal diffusion to design complete antibody sequences and structures from scratch, conditioned on a target protein and binding epitope. Chai reported that its predecessor Chai 2 achieved experimental hit rates of roughly 16 to 20 percent in fully de novo antibody design, compared with sub 1 percent rates for prior computational methods.

BMS in context

For Bristol Myers Squibb, the Chai partnership is one piece of a wider AI strategy. BMS deployed Schrödinger's agentic platform Bunsen in August 2026 and signed an enterprise agreement with Anthropic for Claude across global operations in May 2026.

The Chai deal is the fifth major pharma platform agreement Chai has disclosed in under eight months, following collaborations with Eli Lilly, Pfizer, Novartis, and argenx. It arrives roughly five weeks after Chai closed a $400 million Series C at a $3.8 billion valuation.

Why antibody design matters now

Antibodies remain one of the most commercially validated therapeutic modalities, but early stage discovery is slow and expensive. AI platforms that can generate candidates computationally and compress discovery cycles from months to weeks address a bottleneck that traditional screening has struggled with for decades.

The open question is whether hit rates in silico translate into clinical success. As of mid 2026, the broader AI drug pipeline has seen 175 programs enter trials but zero FDA approvals attributed to AI designed molecules. Partnerships like the Chai BMS deal are bets that the computational gains will eventually show up in validated clinical endpoints.

What to watch

BMS's stated goal of a continuously learning discovery system implies feedback loops between experimental results and model retraining. Whether Chai's platform integrates deeply enough to capture proprietary assay data will determine if the collaboration produces differentiated candidates or remains a modular tool in a larger stack.

For the AI biotech sector, the deal reinforces that pharma buyers want specialized foundation models, not generic chat interfaces wrapped around chemistry workflows.

Sources

BioSpace August 20 2026; Hit Consultant August 20 2026; AllSci August 20 2026