Anthropic's AI models have demonstrated significant progress in protein design, achieving a 35.1% hit rate in a recent campaign. This surpasses the current field norm of 10–15%, according to the company's report published on August 18, 2026.
TL;DR
- Anthropic's AI models designed 354 binding proteins across 15 targets, with Mythos Preview hitting 35.1% in single-target mode.
- The campaign was notable for its high degree of automation, with human involvement limited to infrastructure maintenance and DNA orders.
- The results suggest that AI models can achieve expert-level skill in protein design, although further characterization is needed to confirm the hit rates and affinities.
What happened
Anthropic published results from a protein binder design campaign on August 18, 2026, stating that its models ran the campaign almost entirely autonomously. The models were given a design prompt, internet access, a library of protein-design papers, and unlimited token use and sub-agent spawning. Human involvement was minimal, limited to approving network requests, maintaining infrastructure, and placing DNA orders.
Two models were tested: Opus 4.8 and Mythos Preview, an unreleased frontier model. The campaign involved 1,320 designs across 15 targets, with 354 classified as binders. Mythos Preview achieved a hit rate of 35.1% in single-target mode, while Opus 4.8 hit 22.6% in multi-target mode. The overall hit rate for Mythos Preview in multi-target mode was 26.7%.
Adaptyv Bio, the automated contract research organization that carried out the binding assays, published a separate case study the following day. Adaptyv reported 95% expression across the designs and noted that the campaign's hit rates would have won five of the six protein design competitions it has run.
Why it matters
The campaign demonstrates the potential of AI models to achieve expert-level skill in protein design, matching or exceeding human experts on many tasks. This could significantly accelerate drug discovery and development processes, according to Adaptyv Bio's case study.
The high degree of automation in the campaign suggests that AI models can handle complex tasks such as target research, tool selection, candidate narrowing, and ranking under a protocol set by human experts. This could reduce the need for human intervention in protein design, making the process more efficient and cost-effective.
However, the campaign has limitations. The models did not generate structures using any protein model of Anthropic's own, and the campaign used thousands of H100 hours per target. Additionally, the 10–15% baseline is an aggregate Anthropic drew from publicly documented campaigns in the Proteinbase database, and the head-to-head comparisons are against competition entrants rather than a resourced industrial group working the same targets with comparable compute.
Key facts
- Anthropic's AI models achieved a 35.1% hit rate in single-target mode and a 26.7% hit rate in multi-target mode, according to the company's report.
- The campaign involved 1,320 designs across 15 targets, with 354 classified as binders.
- Adaptyv Bio reported 95% expression across the designs and noted that the campaign's hit rates would have won five of the six protein design competitions it has run.
- The campaign was notable for its high degree of automation, with human involvement limited to infrastructure maintenance and DNA orders.
- The models used publicly available structure design, sequence design, and co-folding tools, and did not generate structures using any protein model of Anthropic's own.
- The campaign used thousands of H100 hours per target, and most targets were established benchmarks, with two being novel.
- Anthropic plans to carry out further characterization to confirm both the hit rates and the affinities, and describes the current figures as first-pass.
- The campaign was open-loop, with the models designing, the designs being tested, and the results being reported without the models ever seeing them.
Context
This campaign is part of a broader trend in the AI industry towards automating complex tasks in drug discovery and development. AI models are increasingly being used to design proteins, discover new drugs, and optimize clinical trials, according to a report by OncoDaily Biotech.
Anthropic is not alone in this effort. Other companies, such as XenoSpectrum, are also exploring the use of AI models in protein design. However, the degree of automation achieved by Anthropic's models is notable, and the hit rates achieved in this campaign suggest that AI models can achieve expert-level skill in protein design.
The campaign also highlights the importance of independent scrutiny in AI research. While the results are promising, further characterization is needed to confirm the hit rates and affinities, and independent replication of the results would be valuable. Anthropic has released prompts, sequences, in silico design files, and raw data on Hugging Face and announced a Proteinbase deposition, which should facilitate independent analysis and replication.
