An open benchmark and language models for AI in aging biology.
Key claim: LongevityBench shows no single frontier AI dominates aging-biology tasks—omics age prediction is hardest—while compact fine-tuned Longevity-LLMs (0.6B–9B) match or exceed far larger systems and are released with an agentic research interface.
Abstract
Over the past two decades, human aging has been characterized across DNA methylation, transcriptomic, proteomic, and clinical modalities, yet no benchmark evaluates whether AI systems can interpret these heterogeneous data types in the context of aging biology. We introduce LongevityBench, an open suite of 17 tasks spanning five biodata domains, and use it to assess 18 frontier AI systems from six developer teams. Despite recent advances in AI, no single model dominates all tasks, with omics-based age prediction being the hardest task regardless of scale. To test whether these gaps can be closed without frontier-scale resources, we fine-tuned a family of five multitask Longevity-LLMs on domain-specific aging data. The compact (0.6B-9B parameters) Longevity-LLMs matched or exceeded far larger frontier systems on LongevityBench, showing that general-purpose language models can be adapted to structured-omics tasks. We publicly release the benchmark, models, and Longevity Claw, an agentic research interface for aging researchers.
Why this matters
LongevityBench is the first attempt to actually score whether frontier AI can interpret the heterogeneous data streams the AI-in-longevity narrative rests on — and the answer so far is ‘no single model dominates, and omics age prediction is the hardest task.’ That the small domain-adapted Longevity-LLMs match much larger systems reframes the trend from a ‘bigger frontier model = better geroscience’ story toward domain-specific tooling and benchmarks. It gives the field a shared yardstick, which is what most AI-longevity press claims currently lack.