Claude optimised inference for thirty open biology models, Anthropic says
The company reports a fourfold speed-up and is funding a protein design competition to test more than five thousand designs experimentally.
Anthropic said Claude optimised inference for more than thirty open-source biology models, reporting speed-ups of around four times. The models covered tasks such as predicting molecular structure, designing drug-like molecules and assessing the effects of genetic mutations.
The stated problem is cost. Specialist biology models are expensive to run, which limits who can use them, and the company argues that making them cheaper widens access more than releasing another model would.
Alongside it, Anthropic is funding a protein design competition with Adaptyv Bio in which more than five thousand designs will be validated experimentally. It is providing up to $1 million in model credits plus funding, with Modal contributing compute and Twist Bioscience supplying DNA.
Experimental validation is the part that distinguishes this from a benchmark. A design that scores well computationally and fails at the bench has not worked, and competitions of this shape produce results that can be checked.
The speed-up figures are Anthropic's own, though the company published its code and a technical report alongside them, which makes the claim checkable in a way most are not.