Claude made 30+ biology AI models ~4× faster in under a month

Anthropic says Claude optimized more than 30 open-source biomolecular models in just under four weeks, making them about 4× faster on average with only a small loss in precision. With outputs kept exactly identical, the speedup is still close to 2×. The work produced 36 optimized packages across six model families: co-folding and structure prediction, structure generation, inverse folding, hallucination, genomics and protein language models. The biggest gain in fast mode was 6.4× for Chai-1.

Claude also built a low-memory "big" mode that accurately predicts systems larger than 10,000 tokens (amino acids, nucleotides, and atoms from small molecules and ions) on a single NVIDIA GPU node, and it has handled systems of up to about 70,000 tokens. Anthropic tested it on large structures such as mitochondrial complex I and bacterial ribosomes. For protein design, Anthropic reports about 100× lower GPU cost than its earlier runs: comparable results came from one H200 in 24 hours, for roughly $150 in compute and tokens.

The code is open source on GitHub (anthropics/uplifting-biomolecular-modeling) with a technical report. Anthropic is also running a protein design competition with Adaptyv Bio, with up to $1M in Claude credits and lab testing of more than 5,000 designs.

Why it matters: here the AI rewrote the scientific software itself, and every lab using these models now gets faster, cheaper runs on larger molecules.