Preprint: TigerAI benchmarks GPT-5 for aggregating genetic evidence in drug development

A preprint posted to bioRxiv on 1 September 2026 introduces TigerAI, a system designed to evaluate whether large language models (LLMs) — specifically GPT-5 — can reliably aggregate and summarise genetic evidence relevant to drug development decisions.

The authors developed a domain-grounded instruction framework and benchmarked LLM-derived genetic evidence against curated data drawn from 13,022 target–indication pairs sourced from a comprehensive drug development database. The paper focuses on the potential for AI-assisted curation to reduce the laborious manual effort currently required to compile genetic support for candidate drug targets before and during clinical trials.

The work positions LLMs as tools for scaling knowledge synthesis in drug discovery, rather than as replacements for expert judgement. The authors benchmark outputs against existing curated evidence to assess accuracy and reproducibility, though the preprint has not yet been peer-reviewed.

TigerAI is described as a research and evidence-generation platform to support clinical development decisions; it is not presented as a diagnostic or clinical decision-support product. The study will be of interest to translational researchers, computational geneticists, and those working at the interface of genomics and drug discovery. The preprint has not yet undergone peer review and findings should be interpreted accordingly.