Benchmark Radar
AI BENCHMARK PROFILE

ViroBench

Health & Life SciencesKnowledge & ReasoningQIANJINYDX / ViroBench team

ViroBench evaluates nucleotide foundation models on viral genomics tasks across two axes: biological understanding (taxonomy classification, host prediction) and latent biosecurity risk (genome modeling, CDS completion). It includes 18 tasks, 58,314 viral samples, multiple split strategies (genus-disjoint, temporal), and 11 metrics, with a unified evaluation protocol and leaderboard.

Released
2026-05-25
Readiness
Runnable
Primary field
Health & Life Sciences

Why it matters

Viral genomics lacks a unified evaluation standard for foundation models. ViroBench provides a diagnostic benchmark that measures both task performance and biosecurity risk, enabling model comparison and guiding development of safer, more robust viral nucleotide models.

Motivation

Nucleotide sequences constitute the fundamental genetic basis of biological systems, rendering viral genomic analysis critical for biomedical advancement.

Primary resources

Benchmark Radar records only publicly supported details and links back to primary sources for verification.