LongevityBench
Benchmark evaluating whether state-of-the-art LLMs grasp aging-biology principles and can process low-level biodata. 30,193 structured prompts aggregated from seven sources spanning transcriptomes, DNA-methylation, proteomes, genomes, clinical blood tests, biometrics and natural-language annotations, across 17 tasks (e.g. human time-to-death, gene-mutation effects on lifespan, age-dependent expression).
Composite
54.4
Experimental validation
N/A — retrospective LLM-evaluation benchmark over public aging datasets
Stages
Disease modelingTarget ID
Modalities
small molecule
Task types
classificationregressiongeneration
Size
prompts: 30,193
tasks: 17
sources: 7
splits: {'train': 0, 'val': 0, 'test': 30193}
note: Multimodal longevity prompts (methylation, proteomics, transcriptomics, genome, clinical, biometrics, NL); used to evaluate o3/GPT-5/Gemini and the Longevity-LLM foundation model
tasks: 17
sources: 7
splits: {'train': 0, 'val': 0, 'test': 30193}
note: Multimodal longevity prompts (methylation, proteomics, transcriptomics, genome, clinical, biometrics, NL); used to evaluate o3/GPT-5/Gemini and the Longevity-LLM foundation model
License
Other — see bioRxiv preprint
First release
2026-01-12
Last updated
2026-06-01
Official site
Leaderboard
→ leaderboard
Dataset
Code / GitHub
→ repository
HuggingFace
→ HF
Paper
LongevityBench: Are SotA LLMs ready for aging research? · LongevityBench authors — see bioRxiv 10.64898/2026.01.12.698650 · 2026 · paper · doi:N/A — bioRxiv 10.64898/2026.01.12.698650 · 0 citations
Flags
none
Experts
—
Groups
—
Hosted by
—
Related benchmarks
Rubric (7-criterion)
rigor
3
coverage
4
maintenance
3
adoption
1
quality
3
accessibility
3
industry_relevance
2
Notes
Broad multimodal aging benchmark (30,193 prompts, 17 tasks across seven data types) is its main strength (coverage 4). Rigor/quality 3: aggregates heterogeneous public sources and is paired with the authors' own Longevity-LLM model — watch for co-development bias when interpreting rankings that favor that model. Adoption 1 as a preprint refreshed through mid-2026; industry relevance 2 (research LLM-evaluation focus). Distinct from the Insilico Longevity Benchmark already tracked, which is a portal-hosted task suite rather than an LLM prompt set.