Anthropic Open-Sources Claude’s Biology Model Optimizations
Anthropic says Claude optimized more than 30 open biology models, delivering roughly fourfold speedups and a lower-memory operating mode.
The result
Anthropic says an internal general-purpose research model optimized more than 30 open-source deep-learning models used in biomolecular research. Across structure prediction, protein design, genomics and protein-language tasks, the company reports roughly fourfold average speedups with minimal precision loss, and nearly twofold acceleration while preserving identical outputs.
The work also produced a low-memory mode for systems containing more than 10,000 tokens of biological input, including amino acids, nucleotides and atoms from small molecules and ions. Anthropic says these workloads can run on a single NVIDIA GPU node, potentially lowering the hardware barrier for research teams that cannot afford large clusters.
The announcement follows an earlier demonstration in which Claude orchestrated open-source tools to design de novo protein binders. Anthropic acknowledges that experiment required up to $10,000 of infrastructure per target, equivalent to about 2,500 H100 GPU hours. The new optimization work is presented as a way to make similar workflows materially more accessible.
From benchmark to wet lab
Anthropic is open-sourcing the optimized code and launching a protein-design competition with Adaptyv Bio. The program covers five challenging problems and includes up to $1 million in Claude credits, $250,000 in Modal compute credits and experimental validation for more than 5,000 designs.
The wet-lab component matters because faster inference alone does not establish that generated proteins work outside simulation. Experimental screening can expose failures in binding, stability or manufacturability that software metrics miss.
Why it matters
This is a notable example of a general AI model improving the efficiency of specialized scientific software rather than merely generating scientific prose. If the reported gains survive independent reproduction and laboratory testing, the effect could be felt in research budgets and iteration speed across drug discovery. For now, the strongest evidence is an engineering result from Anthropic; biological value remains to be demonstrated experimentally.