Claude designed protein binders against 14 of 15 targets at a 26.7% hit rate.

Anthropic ran a de novo binder campaign where Claude researched each target's biology, picked docking sites, installed open-source tools from their public repos itself, and composed 24 workflows with no human making a design decision. Of 1,320 designs synthesized and measured by Adaptyv Bio and Twist Bioscience, 354 bound. Claude Mythos Preview hit 26.7% multi-target and 35.1% single-target against a stated 10-15% industry baseline, using up to 12,500 H100-hours. (Anthropic) Prompts, raw NMR/LC-MS files and complex models are on Hugging Face, so it doubles as a reproducible benchmark. Peer review is still pending, which is the caveat to hold. But note the contrast with the two findings above: this worked because binding affinity is a hard measurable, and the agent's judgment was checked by a wet lab rather than by itself.