Anthropic has shown that Claude can orchestrate a complete protein design workflow by automatically installing and coordinating a dozen open-source specialized tools, designing binders for 16 protein targets with a 26.8 percent hit rate—meaning 354 of 1,320 lab-tested designs successfully bound their targets.
The key innovation is not the protein models themselves, which already exist, but the language model layer above them that chooses targets, installs tools, combines workflows, and ranks results without human intervention on design decisions.
Since all the tools are open-source, the company argues this capability is within reach for any laboratory.
What happened
Anthropic demonstrated that Claude language models (Mythos Preview and Opus 4.8) can run a complete protein design workflow by installing and coordinating open-source software tools, designing binders against 16 target proteins with a 26.8 percent hit rate (354 of 1,320 designs tested in the lab actually bound to their targets).
Why it matters
De novo protein design normally requires days of expert decisions and manual orchestration of specialized software; Anthropic's result shows that a general language model can automate the entire pipeline without humans touching individual design decisions, potentially making the workflow accessible to any lab with access to Claude and cloud compute.
What to watch
The compute cost was $50,000 per multi-target campaign and $10,000 per single target run through Modal; Anthropic has published the prompts, design data, and measurement datasets on Hugging Face, enabling other labs to reproduce and benchmark the approach—though the authors acknowledge no parallel expert campaign was run as a control.
Ask the AI about this article →
Protein design has undergone a fundamental shift since the introduction of AlphaFold and tools like RFdiffusion, which can predict and generate new protein structures. What has not been automated until now is the orchestration layer—the decisions about which targets to pursue, which tools to deploy, how to combine them across dozens of workflow variations, and how to rank the results. Anthropic's contribution is to show that a general-purpose language model can occupy that role, automating not just one design step but the entire campaign from target selection through final ranking. The report notes that the actual protein design work is still being done by the specialized open-source tools (PXDesign, RFdiffusion, SolubleMPNN, ESMFold2, and others); Claude is coordinating them using a 16,000-word protocol prompt that embeds both scientific knowledge and operational discipline.
The significance lies in accessibility and efficiency. Traditionally, de novo protein design campaigns require expert researchers to spend days orchestrating specialized software and managing compute. By removing the human from the decision loop—except for target selection, prompt writing, and result interpretation—the workflow becomes replicable by any lab with access to Claude and cloud compute. The 26.8 percent hit rate substantially exceeds the publicly documented 10–15 percent range, though the authors are explicit that no parallel expert campaign was run as a control, so they do not claim Claude's designs are better than what specialists would achieve with the same tools and budget. The report also notes that for four of six contest targets, contest results were included in Claude's reading list, which may have aided its performance.
For example, today's edition would include:
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. Q&As are published on this page for other readers too.
Chinese large-model developer Z.ai says it can now support large-scale inference using roughly 100,000 domesti…
Analyst Ming-Chi Kuo says Nvidia has revived the Rubin CPX AI accelerator with a substantially redesigned arch…
A UK study by UK AI Security Institute and Limbic AI surveyed 6,474 British adults
Broadcom's Clayton Donley says companies are doing mission-critical work with AI agents quickly, but without t…
OpenAI released a new evaluation framework on July 17, 2026, urging companies to measure AI ROI by 'useful out…
As AI agents perform real business tasks, 'Agentic Identity' (giving each AI a unique employee-like ID) and 'D…
