Beyond Memorization: Insilico's DDD Benchmark Tests Whether AI Can Actually Discover Drugs
Insilico Medicine launches DDD Benchmark as a Service — decontaminated real-world drug-discovery evaluation for frontier AI models, with 300+ foundation tests and proprietary candidate baselines.
Drug-discovery AI benchmarks have a contamination problem: models memorize test questions from training data and score well on paper, then fail when a chemist asks for a real preclinical candidate decision. On July 30, 2026, Insilico Medicine (HKEX:3696) launched Drug Discovery and Development (DDD) Benchmark as a Service — a standardized evaluation framework built on decontaminated real-world data and proprietary validated programs.
Can these models actually discover drugs?
Insilico's framing is blunt: "Can these models actually discover drugs, or are they just good at taking tests?" Most public benchmarks, the company argues, suffer from severe data contamination — inflated scores that collapse under real candidate-selection pressure.
The DDD Benchmark anchors evaluation in carefully decontaminated real-world data and more than 30 validated preclinical candidate programs drawn from Insilico's own pipeline — including Rentosertib (ISM001-055), a first-in-class TNIK inhibitor now in Phase III for IPF.
Two complementary suites
Drug Discovery Foundations — more than 300 evaluations from proprietary out-of-distribution test sets and rigorously decontaminated public data. Core competencies span disease biology, molecular property prediction, retrosynthesis, structure-based design, and clinical development.
Drug Candidate Essentials — end-to-end decision-making tests using proprietary reference baselines from Insilico's validated programs.
Organizations submit models via a standard chat-completions API; Insilico scores outputs against expert baselines and delivers verified scorecards. Public leaderboard placement is optional.
Why BaaS matters now
As frontier general-purpose models infiltrate pharma R&D, buyers need independent verification — not leaderboard vanity metrics. Insilico reports compressing preclinical-candidate nomination timelines to roughly 12–18 months versus 2.5–4+ years traditionally, with 10+ IND clearances across its AI-designed portfolio.
The service is live at dddbench.insilico.com, alongside Insilico's broader benchmark explorer at ddb.insilico.com covering 217 benchmarks across biology, chemistry, materials, and longevity.
For pharma AI, July 30 marks a shift from "trust our demo" to pay-for-proof — with contamination controls that public academic benchmarks rarely enforce.
### Sources
- PRNewswire — Insilico DDD Benchmark as a Service launch (July 30, 2026)
- Insilico Medicine — DDD Benchmark portal (July 2026)
- Insilico Medicine — Drug Discovery Benchmark explorer (2026)