An open benchmark and language models for AI in aging biology.

Fine-tuned Longevity-LLMs with 0.6B-9B parameters outperform larger models on aging biology tasks in the LongevityBench benchmark.

• Why it matters: Understanding and interpreting diverse aging data types is crucial for advancing aging research, yet no existing AI systems have demonstrated comprehensive capability across these modalities.
• What they did: Researchers developed LongevityBench, an open set of 17 tasks across five biodata domains, and evaluated 18 AI models, then fine-tuned five multitask Longevity-LLMs on domain-specific aging data.
• The result: The compact Longevity-LLMs matched or surpassed larger models, enabling more accessible AI tools for aging biology and providing publicly available benchmarks, models, and research interfaces.