Patrick Flavin

Anthropic just introduced its new model, Claude Mythos 5.1 — and the protein design results are striking. In wet-lab validation across 12 targets, Mythos 5.1 achieved a hit rate of nearly 50%, the highest Anthropic says it has measured to date. On three targets, its designed binders achieved ~10× stronger binding affinity than the best designs submitted to Adaptyv Bio’s protein design competitions. For context, typical hit rates in protein design today are around 10–15%. A new Claude model, now designing experimentally validated proteins at a surprisingly high success rate. AI is moving from reading biology to designing it.

A PhD student in the Baker Lab is using Claude to take over the 4 a.m. lab shift. Zihao Song connected six instruments in under a week using Anthropic’s Model Hardware Standard (MHS), letting an AI agent monitor qPCR in real time, stop reactions at the right moment, coordinate robotic plate handoffs, and oversee experiments remotely. AI can already design proteins. Now it’s starting to operate the wet lab too. anthropic.com/news/model-har…