Scientific LLM Benchmarks
GitHub
← All benchmarks
Biology· agentic-genomics-retrieval

VirBench

Anthropic (with NCBI, Broad Institute, Pachter Lab) · 2026

120 curated viral-sequence retrieval queries across ~40 pathogens testing whether LLM agents pull correct ground-truth sequence data from NCBI Virus.

Biology
Task type
agentic
Modality
text
Access
request
Size
120 items
License
CC-BY-4.0
Metrics
retrieval accuracy

Examples

Samples are not shown — this dataset is request.