Pfizer licenses Chai Discovery’s AI platform to speed up biologics and antibody design

Pfizer has entered a licensing agreement with Chai Discovery to use the startup’s generative artificial intelligence platform in drug discovery, aiming to compress early research cycles for biologics and antibody programs. Chai Discovery said the deal gives Pfizer early access to Chai-3, the company’s most advanced AI model for antibody design, which it described as previously undisclosed and a step-change improvement over its earlier system.

Under the agreement, Pfizer will deploy Chai’s AI platform as part of its drug discovery “engine,” Chai said. The companies also described a custom model that uses Pfizer’s proprietary data and is tailored to Pfizer’s internal workflows, reflecting a common enterprise approach in pharma: integrate AI tools into existing discovery processes rather than treat them as standalone software.

Chai said Chai-3 targets what it called hard-to-reach steps in antibody discovery, improving the quality of generated molecules and broadening the types of therapeutics candidates the model can support. The startup’s statement said Chai-3 doubles the success rate of its predecessor and produces antibodies that meet required therapeutic standards.

Chai-3, according to the announcement, is intended to advance multiple practical design dimensions, including therapeutic binding capabilities, the creation of multi-specific antibodies, and the ability to design molecules for difficult-to-drug targets. The model was also described as improving generalization, a reference to how well an AI system performs beyond the specific cases it may have been tuned on.

The licensing agreement “demonstrates the increasing speed of adoption” of frontier AI models by major pharmaceutical companies as tools move from research experiments into operational discovery work, Chai said. Chai’s co-founder Joshua Meier described the partnership as putting Chai’s software directly into the hands of a large drug-discovery organization and said the goal is to combine the AI platform with Pfizer’s scientific depth, data, and discovery capabilities to expand what biologics teams can pursue.

Chai said its prior model, Chai-2, released in 2025, was the first zero-shot antibody design platform to achieve double-digit experimental hit rates and to produce molecules with drug-like properties, which it said represented a roughly 100-fold improvement over earlier computational approaches. The company also claimed that the AI approach can reduce discovery timelines from months or years into short sprints, with Chai-2 enabling discovery processes to be completed in weeks rather than months, but it did not provide details on how those timelines translate to Pfizer’s specific programs.

Pfizer did not disclose financial terms, the scope of which research areas will be covered first, or any independent validation results tied to Pfizer’s own experimental pipeline. In addition, because Chai-3 was described as “previously undisclosed,” there is limited information available publicly about the training data, evaluation benchmarks, and how “success rate” will be measured once the model is deployed inside Pfizer.

What to watch next is whether Pfizer provides follow-up disclosures on integration progress, including whether the system generates measurable increases in experimental throughput or hit rates in specific biologics programs. Chai’s claims focus on antibody design performance, but the practical business question is whether that translates into faster advancement of candidates toward preclinical development and, eventually, clinical trial starts.