Researchers build the first scorecard for whether AI actually works at drug discovery

Scientists created a benchmark test to measure how well large language models perform at drug discovery tasks — generating descriptions of drug properties, predicting interactions, and explaining their reasoning. Right now, nobody knows if AI is actually better than existing drug discovery tools, because there's no standard way to measure it.

Drug discovery is slow and expensive, and AI companies are claiming they can speed it up. This benchmark is the first attempt to actually test those claims against specific, measurable tasks instead of just marketing promises. If the benchmark becomes widely adopted, it stops AI vendors from making vague claims about acceleration and forces them to show actual performance on concrete problems — which either proves AI works for drug discovery, or reveals it doesn't.