Claude in Pharma: Anthropic Reports Results on 35,000 Antibodies

Anthropic has detailed the results of several months of work by its Claude AI model across six drug development organizations: BigHat Biosciences, Bristol Myers Squibb, Divergence Labs, Enveda, Generate Biomedicines, and The Column Group. Concurrently, the company has launched its own research programs targeting rare pediatric forms of muscular dystrophy.

• Anthropic summarized months of Claude's application across six biotech organizations.
• Key outcomes include 35,000 antibody variants in a month and an 89% reduction in document review time.
• Current developments remain strictly at the preclinical stage without human trials.

The main takeaway from the report is that bottlenecks in drug discovery are distributed throughout the entire pipeline, meaning efficiency gains come from deploying the model across multiple stages simultaneously. Based on these collaborative projects, Anthropic has already prioritized specific technical enhancements for its upcoming model iterations, focusing on data analysis and automation.

Partner results demonstrate concrete technical achievements. Divergence Labs tackled a food allergy target by using Claude to generate approximately 35,000 antibody variants in about a month, yielding candidates that successfully separated the target molecule from immune cells. Meanwhile, Generate Biomedicines utilized Claude Code to design a genetic instruction sequence 70% shorter than standard variants, which prompted immune cells to produce roughly ten times more receptors.

BigHat Biosciences coupled the AI model with an automated laboratory named Reccy to test 1,010 variants over ten weeks, producing molecules with significantly stronger target binding properties. Enveda applied Claude to manufacturing documentation, successfully identifying all necessary corrections across 1,061 pages and cutting document review time by 89% from over 25 hours to under three hours.

Despite these successes, the report explicitly outlines the boundaries of the methodology. The model excels where large datasets and formalized metrics are available, but it cannot replace human domain experts in ambiguous situations. Furthermore, initial regulatory drafts generated by AI still required human error correction, emphasizing that stakes in pharmaceutical development leave no room for oversight.

For researchers and industry professionals, these findings do not mean fully developed drugs will hit pharmacies anytime soon. Most processes remain strictly preclinical, with Anthropic's own lead internal project targeting human clinical trials by the end of 2028. The core value lies in creating a robust workflow coupling large language models directly with laboratory loops and corporate documentation systems.