Insilico Medicine Launches Open-Source AI Toolkit for Aging Research
LongevityBench benchmark and compact LLMs aim to accelerate AI-driven longevity discoveries.
1Comprehensive aging biology benchmark
Insilico Medicine introduced LongevityBench, a suite of 17 tasks across five biodata domains: clinical data, genetics, epigenetics, transcriptomics, and proteomics. The benchmark uniquely evaluates AI systems' ability to interpret heterogeneous aging-related data types, moving beyond simple recall to test pattern recognition and problem-solving. It includes datasets from NHANES clinical measurements, GEO DNA methylation, and GTEx bulk RNA-seq among others. The open framework aims to establish standardized metrics for AI progress in longevity research.
2Compact models outperform frontier AI
Researchers developed five specialized Longevity-LLMs ranging from 0.6B to 9B parameters, fine-tuned on aging-specific data. Surprisingly, these compact models matched or exceeded performance of much larger frontier systems from OpenAI, Google, and Anthropic on benchmark tasks. The 9B-parameter L-Qwen3.5-9B achieved the best aggregate rank score (4.4), outperforming Gemini 3.1 Pro (8.2) and Claude Opus-4.6 (9.2). In specific tasks like GEO DNA-methylation age prediction, it reached 0.868 concordance versus 0.685 for the best frontier model.
3Agentic research platform capabilities
The open-source Longevity Claw platform integrates the top-performing L-Qwen3.5-9B model with specialized tools for multi-step research workflows. It performs gene-set enrichment analysis, biological aging-clock calculation, and candidate target evaluation across 14 hallmarks of aging. The platform nominated 328 potential intervention targets, showing up to 5.6-fold enrichment against validated aging-related genes. One nominated target, KDM1A, was independently validated to extend lifespan in C. elegans in a separate study.
4Rigorous evaluation methodology
The study assessed 18 frontier AI systems from six developer teams across all benchmark tasks. Results showed no single model dominated all domains, with performance varying by question phrasing. Omics-based age prediction emerged as the most challenging task regardless of model scale. The public leaderboard tracks 26 models across 17 tasks, using datasets from OpenGenes, SynergyAge, and Olink plasma proteomics. Evaluation focused on biological reasoning rather than training data recall.
5Technical architecture details
The Longevity-LLMs were trained using Insilico's proprietary MMAI Gym for Science platform, combining the company's biological datasets with reasoning tasks. They leverage Liquid AI's LFM2 architecture and Alibaba's Qwen3/3.5 model families. Longevity Claw incorporates 233 biological age clocks spanning six modalities with 429,165 coefficients. The target discovery module scores candidates on novelty, druggability, and safety across all aging hallmarks. All components are MIT-licensed and available on Hugging Face.
6Clinical validation context
This release follows Insilico's September 7 Nature Biotechnology study showing their AI-discovered drug rentosertib reduced biological age across six proteomic clocks in Phase IIa trials for pulmonary fibrosis. The new toolkit extends this work by enabling systematic evaluation of AI systems' biological reasoning capabilities. The company emphasizes translating these tools into personalized longevity assistants that monitor and improve healthspan, marking a shift from drug discovery to AI-powered health maintenance.
7Collaborative research ecosystem
The study involved researchers from Liquid AI, Buck Institute for Research on Aging, Harvard Medical School, and Brigham and Women's Hospital. By open-sourcing the benchmark, models, and platform, Insilico aims to create a common foundation for measuring AI progress in aging biology. The approach contrasts with proprietary systems by enabling independent validation and distinguishing genuine biological reasoning from training data reproduction. The toolkit could accelerate target discovery across academic and industry research.
8Strategic implications for longevity research
This release represents a paradigm shift in aging research methodology, combining rigorous benchmarking with specialized, efficient AI models. The open framework lowers barriers to high-quality longevity research while maintaining scientific rigor through transparent evaluation. By focusing on biological reasoning rather than model scale, it challenges assumptions about AI performance in life sciences. The platform's ability to identify and prioritize novel targets could significantly compress drug discovery timelines for age-related diseases.
Original article: Unite.AI — Insilico Medicine Launches Open-Source AI Toolkit for Aging Research