Insilico Medicine Releases Open AI Toolkit for Aging Research

Insilico Medicine Releases Open AI Toolkit for Aging Research

Insilico Medicine has published a landmark study in Cell introducing three open-source AI resources designed to accelerate aging research and longevity therapeutic development. The September 17, 2026 cover study, conducted with Liquid AI, the Buck Institute for Research on Aging, and Harvard Medical School, makes available a benchmark for evaluating AI in aging biology, a family of specialized language models, and an autonomous research platform.

The release follows closely after Insilico's September 7 Nature Biotechnology study showing that rentosertib, its AI-discovered drug candidate for idiopathic pulmonary fibrosis, reduced biological age markers in a Phase IIa clinical trial.

Why it matters

The toolkit addresses a critical gap in AI-enabled drug discovery: general-purpose large language models often struggle with specialized biological reasoning despite their broad capabilities. By demonstrating that compact, domain-specific models can outperform frontier AI systems on aging-related tasks while requiring far less computational power, Insilico provides research institutions with practical tools that don't demand massive infrastructure investments. The open-source approach also establishes standardized methods for measuring genuine biological reasoning versus simple memorization.

LongevityBench Exposes Gaps in Frontier AI Performance

The research team developed LongevityBench, the first open benchmark specifically designed to evaluate AI reasoning across five biological domains relevant to aging: clinical data, genetics, epigenetics, transcriptomics, and proteomics. The benchmark tests whether AI systems can analyze biological data and recognize meaningful patterns rather than simply recall training information.

When the researchers evaluated 18 leading AI systems from OpenAI, Google, Anthropic, xAI, DeepSeek, and Moonshot AI, they found substantial performance gaps. No single frontier model achieved the strongest results across all five data types, and performance varied significantly based on question phrasing. The most challenging task—predicting biological age directly from omics measurements—proved difficult even for the largest models.

Compact Specialized Models Outperform General-Purpose AI

Insilico developed five Longevity Large Language Models (L-LLMs) ranging from 0.6 billion to 9 billion parameters, fine-tuned on aging-specific clinical and multi-omics data. Built on open architectures from Liquid AI and Alibaba, these specialized models matched or exceeded all 16 frontier systems evaluated on LongevityBench.

The best-performing model, L-Qwen3.5-9B with 9 billion parameters, outperformed every tested frontier model including Google's Gemini 3.1-Pro. Even the smallest model at 0.6 billion parameters surpassed most frontier systems, demonstrating that curated scientific training data and domain-specific optimization can matter more than raw model size.

Autonomous Platform Identifies Therapeutic Targets

The team embedded L-Qwen3.5-9B into Longevity Claw, an open-source agentic platform that combines the specialized language model with scientific tools for gene-set enrichment analysis, aging-clock calculation, population profiling, and target evaluation. Unlike conventional chatbots, the platform formulates and executes multi-step research workflows autonomously.

Deployed across 14 recognized hallmarks of aging, Longevity Claw nominated 328 genes as potential intervention targets. These candidates showed up to 5.6-fold enrichment when compared with an independently published reference set of experimentally supported aging-related targets. One nominated gene, KDM1A, was independently validated in a separate published study as extending lifespan in C. elegans.

Open Access to Research Community

Insilico is releasing the benchmark, specialized models, training resources, evaluation code, and Longevity Claw platform for independent testing and development. The resources are available through Hugging Face, GitHub, and a public leaderboard at longevitybenchmarks.org.

The company reported approximately $106 million in revenue for the first half of 2026, a 287% year-over-year increase, and achieved its first profitable half-year with adjusted net profit exceeding $51 million. Total contract value of transactions announced in 2026 reached approximately $7.3 billion.

Details of the research were first reported by Insilico Medicine in a press release accompanying the Cell publication.