Claude Speeds Up 30+ Open-Source Biomolecular Models 4x on Average, All Code Open-Sourced

FinanceYF5 · x · 2026-09-20

Anthropic announced that Claude (a general-purpose research model within Claude Science) optimized the inference paths of more than 30 open-source biomolecular models in under four weeks, achieving roughly 4x average speedups. It also added a low-memory mode enabling accurate prediction of biomolecular systems larger than 10,000 tokens on a single NVIDIA GPU node. Some optimizations come from custom GPU software Claude wrote itself; all code is open-sourced at anthropics/uplifting-biomolecular-modeling, with one drop-in optimization kit per upstream tool covering structure prediction, cofolding, binder/sequence design, and protein/genomic language models.

To showcase these gains, Anthropic and Adaptyv Bio launched a protein design competition: five weekly challenges from Sep 28 to Oct 31, with over 5,000 designs tested for free in Adaptyv's automated wet lab (backed by $1M in validation funding), plus up to $1M in Claude credits for selected teams (up to $50K academic, $25K industry). Modal provides up to $250K in compute and Twist Bioscience provides DNA. All results will be published openly on Proteinbase; entry is free.

Original post →

More from Research

Research channel →