Liquid AI and Insilico's Longevity LLMs Hit Cell Cover, Beat Frontier Models

Liquid AI and Insilico Medicine's joint work on longevity foundation models was published on the cover of Cell. Per posts by @helloiamleonie, @MaxUnfried, and @JosephJacks, general Liquid Foundation Models (LFM) were fine-tuned into compact Longevity-LLMs that achieved SOTA on LongevityBench, surpassing Gemini-3.1-Pro, GPT-5, Claude Opus, and Grok.

Confirmed

• The paper appeared on the cover of Cell, a collaboration between Liquid AI and Insilico Medicine, with model sizes down to 1.2B parameters.
• Per @JosephJacks relaying biogerontology, a roughly 9B-parameter model beat GPT-5, Gemini 3.1, Grok, and Claude Opus on aging biology tasks, with confidence intervals not overlapping any other model in the study.
• A single model family spans the full biological data stack: clinical data, genetics, DNA methylation, transcriptomics, and (plasma) proteomics.
• They also released LongevityBench, a benchmark of 17 biological longevity tasks covering clinical records, methylation, transcriptomics, plasma proteomics, and genetics.
• Fine-tuned compact models met or exceeded prior SOTA, outperforming multiple frontier LLMs.
• @JosephJacks relayed that the Longevity Claw system built on these models nominated drug candidates (the original post was truncated at the count).

Why it matters

• A notable case of compact specialized models (1.2B–9B) beating general-purpose frontier flagships on vertical biology tasks, with non-overlapping confidence intervals indicating statistical significance and showing the cost-effectiveness of domain fine-tuning.
• A Cell cover publication is a landmark for the AI–longevity medicine intersection, and LongevityBench provides a standardized evaluation benchmark for the field.

2026-09-17 ~ 2026-09-18 · 7 related posts

• [source] Liquid AI's small Longevity models beat GPT-5 and Claude Opus on Cell-published aging benchmarks — helloiamleonie · 2026-09-17
• Insilico Medicine and Liquid AI's Longevity-LLMs Make Cell Cover, Nominate 328 Aging Gene Targets — JosephJacks_ · 2026-09-18
• [source] Liquid AI and Insilico publish aging-biology foundation model on Cell cover — MaxUnfried · 2026-09-18
• [source] Liquid AI's aging-biology LFMs published in Cell; 1.2B model rivals frontier LLMs — JosephJacks_ · 2026-09-18
• A 9B-Parameter Model Beat GPT-5, Gemini 3.1 and Claude Opus at Aging Biology — JosephJacks_ · 2026-09-18

2 near-duplicate retellings: helloiamleonie · JosephJacks_