Claude designs 35,000 antibodies in a month, Anthropic shares pharma results

What this is about

Anthropic has published an update on its life sciences research. The headline: working with the biotech company Divergence Labs, the Claude language model generated about 35,000 antibody designs in roughly a month for a possible treatment of food allergies. An antibody is a protein molecule that docks onto specific other molecules. Anthropic also reports results from projects with Generate Biomedicines, Enveda and Bristol Myers Squibb. The timing stands out: following reports about a stock market registration document (Form S-1, filed on October 26, 2026 according to one of the sources), biology is among the growth areas Anthropic is showing to investors.

The allergy project with Divergence Labs

The goal is to remove IgE (immunoglobulin E), a molecule involved in allergic reactions. Standard drugs only neutralize IgE circulating in the blood. The new approach is meant to go further and also detach IgE from immune cells.

Results so far, according to Anthropic:

• Divergence Labs identified three designs that may bind IgE.
• 20 designs may detach IgE from immune cells.
• Three of those 20 appeared to outperform the standard drug.
• Next step: combine both functions in a single potential drug.

The figure of 35,000 only describes the number of drafts, not their quality. Think of it as the wide top of a funnel: every design still has to be produced, tested in the lab, and checked for binding, safety and manufacturability. How many designs end up being useful is only partly visible. The numbers of three and 20 promising candidates are a first hint.

Other projects

Generate Biomedicines: Claude helped with a shortened version of the genetic instructions for CAR-T cells. These are the body's own immune cells, genetically modified to recognize cancer cells. Here the research targets solid tumors. One design was about 70 percent shorter than the standard version. In lab tests, immune cells produced roughly ten times as many receptors and killed tumor cells more efficiently, Anthropic says.

Enveda: Claude read 1,061 pages of manufacturing documentation. It found all 11 problems that human reviewers had spotted and flagged seven more. Human review time fell by 89 percent, from more than 25 hours to under three.

Bristol Myers Squibb: This project looks at how Claude can help plan clinical trials and estimate their chances of success. The collaboration is due to expand in 2027.

Strategy and infrastructure

Anthropic says it wants to use AI across several phases of drug development, not just in designing molecules. By its own account, it has research partnerships with six organizations, plus its own drug programs, including ones on rare childhood forms of muscular dystrophy. According to one of the sources, the company is building a molecular biology lab, has acquired Coefficient Bio and has hired life sciences experts. The reports give no prices, locations or headcounts. Media outlets also linked the work to Novo Nordisk, Bristol Myers Squibb, AbbVie and Sanofi. The sources don't say whether those four are the same as the six partners, and they don't give contract terms. The S-1 cites the global pharma market, with more than $2 trillion in annual spending in 2026, as an opportunity. Anthropic quotes itself as saying:

We plan to significantly expand this work from here.

Our take

The results show two different strengths: Claude as a design tool (antibodies, genetic constructs) and as a reader of large volumes of documents (Enveda). The Enveda case is the easiest to follow because errors were compared against human reviewers. For the antibody project, solid proof is still missing: hit rates, binding data and larger-scale lab confirmation haven't been published. All figures come from Anthropic itself, and there has been no independent verification so far. Anthropic stresses that the research is early and mostly preclinical, meaning the potential drugs haven't been tested in humans. An IPO raises the incentive to highlight successes, so it's worth waiting for published data and follow-up results. The sources offer no comparison with other AI providers or specialized biotech AI companies.