Insilico Medicine launches a real-world benchmark designed to reveal whether AI can truly discover new medicines – not just ace exams.
Artificial intelligence has become one of the pharmaceutical industry’s favorite trends, promising to speed up everything from identifying disease targets to designing entirely new medicines. But beneath the excitement sits a question so uncomfortable: are today’s AI models genuinely capable of discovering drugs, or have they simply become very good at answering questions they’ve already seen?
Insilico Medicine believes it’s time to find out. The clinical-stage AI drug discovery company has launched what it describes as the industry’s first Drug Discovery and Development (DDD) Benchmark as a Service, a standardized framework designed to evaluate how well frontier AI models perform on the kinds of scientific decisions researchers face every day [1].
Rather than rewarding models for recognizing familiar information, the benchmark is built to measure something much harder: whether AI can make the complex, sequential decisions required to move a potential medicine from an early scientific idea toward becoming a real drug.
Moving beyond perfect test scores
One of the biggest challenges in evaluating AI is that many public benchmarks no longer reflect reality. As large language models have grown more powerful, many have effectively learned the answers to publicly available test sets because those questions were included somewhere in their training data. The result is impressive-looking scores that may not translate into better scientific decision-making.
Insilico’s new benchmark attempts to close that gap by using carefully filtered datasets alongside the company’s own validated drug discovery programs developed over more than 12 years. The goal is to create what is effectively a “blind test” that measures how AI performs when it encounters problems it has not already memorized.
The framework evaluates models across a wide range of drug discovery tasks, including understanding disease biology, predicting how molecules behave, planning chemical synthesis and supporting clinical development. It also examines whether AI can navigate the full journey from identifying promising compounds to selecting candidates suitable for preclinical development.
For organizations building AI systems for pharmaceutical research, the service offers an independent assessment of how their models compare against others, with the option of publishing results on a public leaderboard.
A benchmark built from experience
Unlike many academic evaluations, the DDD Benchmark draws on Insilico’s own track record in AI-enabled drug discovery.
Since the company was founded, it has nominated 31 preclinical candidates, secured more than 10 investigational new drug (IND) clearances and shortened the journey to preclinical candidate nomination to around 12 to 18 months – significantly faster than the traditional timeline of roughly 2.5 to more than four years.
Its lead candidate, Rentosertib (ISM001-055), an AI-discovered and AI-designed therapy for idiopathic pulmonary fibrosis (IPF), has progressed into Phase III clinical development. That experience now forms the reference point against which other AI models can be measured.
“The rapid progress of AI has made one question more urgent than ever: can these models actually discover drugs?” said Dr Alex Zhavoronkov, founder and CEO of Insilico Medicine. “For more than a decade, we have built and validated generative AI across the entire drug discovery value chain, nominating 31 preclinical candidates in six years and advancing an AI-discovered and AI-designed medicine into Phase III.”
Zhavoronkov explained that the DDD Benchmark transforms real-world experience into a rigorous, standardized metric for the entire field to utilize. He added that as the industry advances toward pharmaceutical superintelligence, evaluating authentic capabilities rather than memorized responses is critical for ensuring AI delivers high-quality, differentiated medicines that ultimately extend healthy, productive longevity for people worldwide.
Why this matters for longevity
The launch reflects a broader shift in how AI is being viewed across healthcare. The conversation is moving away from whether AI can generate convincing answers and toward whether it can consistently make reliable scientific decisions that stand up in the laboratory and, ultimately, in the clinic.
For longevity research, that distinction matters. Developing therapies that target the biology of aging depends on navigating enormous amounts of biological data, identifying promising targets and making countless high-stakes decisions before a treatment ever reaches patients.
If AI is to play a meaningful role in extending healthy lifespan, researchers need confidence that these systems can do more than produce persuasive outputs – they need evidence that they can help uncover genuinely new medicines.
By introducing a transparent, real-world benchmark, Insilico is attempting to raise the standard for how AI is measured across drug discovery. In an industry increasingly shaped by intelligent machines, proving that an algorithm can make sound scientific decisions may become just as important as building the next generation of models themselves.
Photograph courtesy of Insilico Medicine
[1] https://insilico.com/news/hxjtp34a41-insilico-medicine-launches-the-industrys
