A translator who gets distracted by every rustle in the room turns an hour-long meeting into two. That is how aging clocks used to work: each tool understood only one "dialect" of biodata. One read chemical tags on DNA (methylation), another read proteins in the blood, a third read the doctor's notes. To add a new data source, you had to build and validate a separate model, and the explanations for its predictions were left to the researcher's judgment.
Longevity-LLM v0.1 is a translator that learned all the dialects at once. Scientists at Insilico Medicine took the base language model Qwen3-14B and fine-tuned it on 766,640 examples in which biodata were written as text with reasoning. The first stage was reading 4.5 billion tokens; the second was reinforcement learning, where the model was rewarded for the accuracy of its numerical predictions. Training ran on 24 NVIDIA B200 GPUs, then on 16 A100s.
The result proved stronger than specialized tools. In predicting 10-year mortality from NHANES data, the model reached an accuracy of 0.89, and in forecasting cancer survival from TCGA data, 0.77. The mean error in predicting age from DNA methylation was 4.34 years, which is statistically significantly better than the Horvath clock (4.61 years). The model also predicted age from proteomic data with an error of 7.9 years.
Such a translator is needed not only for accuracy. It removes the barrier between specialized omics models and conversational neural networks. In pharma, this is a step toward a monolithic pipeline "from hint to drug" that will unite target discovery, molecule design and clinical trial planning.
The translator that learned every dialect of aging now speaks the same language as the researcher. The next version, Longevity-LLM v0.2, will not only predict age but also explain the biological logic behind its predictions: linking molecular features to pathways and phenotypes. Then the meeting will finally move faster.
Before training, the base model Qwen3-14B could not produce a single valid prediction on any task, because it did not know how to format its answers. After fine-tuning, it outperformed commercial systems with hundreds of billions of parameters.
The idea of a single AI that reads biological data and explains aging echoes the medical AI from the series Altered Carbon, where consciousness is transferred between bodies and biological age is controlled by technology.
Not published on arXiv. Here are its page, the PDF and the formalization: you can check it yourself, not only read our account.