LatchBio releases OmicsBench v1, a 784-task meta-benchmark of AI agents on omics data analysis

LatchBio's OmicsBench v1 collects 784 tasks (741 short-horizon, 43 long-horizon) from eight of its benchmarks on omics analysis (SpatialBench, ScBench, EpiBench, VariantBench, BioSecBench-Surveillance, BioSecBench-Function, SpatialBench-Long, ScBench-Long). Each task is a Docker container with a data snapshot, a prompt and an automatic grader. Across 31 model-harness combinations, pass rates ranged from 32.3% to 49.0%. The source benchmarks have public GitHub repositories and a live leaderboard at benchmarks.bio; a write-up in paper form is at latch.bio/omicsbench. Blog post dated 19 October 2026; the results are the company's own.