Claude Models Clear Wet-Lab Benchmark in Autonomous Protein Design
<cite index="2-2">Anthropic published experimental findings on Tuesday revealing that its Claude artificial intelligence (AI) models autonomously designed functional protein binders and executed complex analytical chemistry workflows in minutes.</cite> The results, sourced directly from Anthropic's research post and a case study published by Adaptyv Bio, offer the clearest wet-lab performance benchmark yet for a general large language model (LLM) applied to early-stage drug discovery.
Campaign Design and Models Used
<cite index="2-4,2-5">In a multi-target campaign, Claude Opus 4.8 and Mythos Preview were tasked with creating de novo protein minibinders — small molecular structures designed to latch onto target proteins. Across 15 biological targets, the models produced 1,320 candidate designs.</cite> <cite index="3-12">Anthropic described the exercise as an attempt to assess how much of the early-stage protein-design workflow could be handled by AI with limited human intervention.</cite>
<cite index="11-1,11-2">The campaign ran in two operational modes: a multi-target mode, where Claude designed against all targets simultaneously in a single Claude Science session, and a single-target mode, in which each session addressed one target with all sessions run in parallel.</cite> <cite index="15-1">Claude carried out the work using publicly available specialist protein design models after receiving an initial prompt of about 30,000 tokens written by a human expert.</cite>
Hit Rates and Comparative Performance
<cite index="2-6">Independent wet-lab testing conducted by Adaptyv Bio and Twist Bioscience verified that 354 designs successfully bound to their targets, hitting 14 of the 15 targets tested.</cite> <cite index="2-7">The models achieved overall hit rates ranging from 22.6% to 35.1% depending on the operational configuration, well above the 10% to 15% success rate typical of standard industry campaigns.</cite>
<cite index="15-4">In the multi-arm design campaign, Claude's Mythos Preview and Opus 4.8 models achieved overall hit rates of 26.7% and 22.6%, respectively, when designing against all targets simultaneously in a 48-hour session.</cite> <cite index="15-7">When Mythos Preview focused on individual targets in separate 24-hour sessions, its hit rate rose to 35.1%.</cite>
Performance was uneven across targets. <cite index="2-8">Against RBX1, a regulatory protein target, Mythos Preview achieved a 40% hit rate in single-target mode, outperforming human entrants in a prior Adaptyv Bio competition who registered a 3.7% success rate.</cite> <cite index="4-14">Claude reached an 80% hit rate against TREM2, up from 38.3% in a prior Adaptyv Bio competition.</cite> Conversely, <cite index="7-6">against maltose-binding protein, a notoriously smooth target, none of 90 designs was confirmed to bind.</cite>
<cite index="2-9,2-10">Claude also generated cross-reactive binders against TNFα, an inflammatory signaling protein targeted by blockbuster treatments like Humira. Opus 4.8 produced 12 valid designs capable of binding human, monkey, and mouse versions of TNFα.</cite>
Analytical Chemistry Parallel
<cite index="2-11">In a separate trial evaluating analytical chemistry tasks, Anthropic tested its generally available Claude Opus 5 model on raw nuclear magnetic resonance (NMR) and liquid chromatography–mass spectrometry (LC-MS) data.</cite> <cite index="10-13,10-14">Anthropic gave Opus 5 only the raw files from a contract lab and a two-sentence prompt, with no vendor software and no operator. Working inside Claude Science, the model returned processed results in 23 and 19 minutes, running the two analyses in parallel.</cite> <cite index="14-15">Claude measured the purity of a sample at 96.4%, compared with 96.33% in the lab's results.</cite>
Data Release and Access Constraints
<cite index="8-6">Together with the Anthropic team, Adaptyv Bio is releasing the actual protein sequences Claude designed as well as the experimental data on Proteinbase, the open protein data platform.</cite>
Access to the highest-performing models remains restricted. <cite index="9-5">Protein design and other dual-use biology capabilities are currently blocked for general access in those models because Anthropic has determined it cannot yet reliably distinguish legitimate drug discovery queries from potential bioweapon development applications.</cite> <cite index="11-8">One of Anthropic's highest priorities is to launch an access program for scientists, and the company expects to share more on this soon.</cite>
<cite index="10-10,10-11">Anthropic casts this work as foundational and says it wants Claude to eventually run drug development end to end, while acknowledging that a designed binder is only the first step toward an actual drug.</cite>