← Capa | Mercado

Anthropic Says Claude Sped Up 30+ Open-Source Biomolecular Models 4x in Under Four Weeks

2026-11-04 · Fonte:deep-s://www.unite.ai/feed/?paged=6

Anthropic Says Claude Sped Up 30+ Open-Source Biomolecular Models 4x in Under Four Weeks

Anthropic says Claude optimized 30+ open-source biomolecular models in under four weeks at ~4x average speedup, open-sourcing the code and launching a protein design contest with Adaptyv Bio.

Anthropic reported on September 17, 2026 that Claude optimized more than 30 open-source biomolecular models in just under four weeks, speeding the models up by roughly 4x on average. The company laid out the results in a research post accompanied by a technical report, open-sourced the optimized code, and announced a protein design competition co-sponsored with Adaptyv Bio.

According to the research post and its accompanying technical report, the optimization campaign was executed by an internal general-purpose research system. The deliverables include 36 optimized packages together with FlashPairformer kernels. Anthropic characterized the roughly 4x figure as an average across the portfolio of more than 30 models rather than a single best-case result, and compressed the entire effort into a window of just under four weeks.

The open-source release puts the accelerated implementations directly into the hands of computational biology labs and developers, while the protein design competition co-sponsored with Adaptyv Bio extends the initiative from code optimization into applied protein engineering. Together, the code drop and the contest signal that Anthropic is treating biomolecular tooling as a showcase domain for Claude's applied capabilities.

For AI compute buyers, the result is a data point on how frontier models are being deployed as scientific software engineers: Claude's task was to make existing biomolecular models run faster, not to run the science itself. A roughly 4x average speedup across more than 30 open-source models, delivered in under four weeks, points to potential compute-cost reductions for biomolecular and protein design workloads. Because the optimized packages are public, those gains can now be independently benchmarked by the labs that depend on them.

Tags: Anthropic · Claude · Adaptyv Bio · Biomolecular Models · Protein Design · Open Source


Leituras relacionadas


萬安算交所 AIXX · AIXX · Exchange verificado · Ir ao mercado