Phylo

BiomniBench-DataAnalysis (Biomni-DA-v0)

Preliminary 15-task preview of Phylo's trace-based evaluation for long-horizon biological data analysis. It scores analytical process—including data handling, method selection, statistical rigor, source reliability, and reasoning—rather than only final-answer accuracy. Results are preliminary and the benchmark remains a work in progress.

biology-researchbiomedical-data-analysistrace-evaluationtabularcodetext
Status
curated
Saturation
unsaturated
Reproduction
not yet
Published results
6 models

Editor's note

Preliminary 15-task preview of Phylo's trace-based evaluation for long-horizon biological data analysis. It scores analytical process—including data handling, method selection, statistical rigor, source reliability, and reasoning—rather than only final-answer accuracy. Results are preliminary and the benchmark remains a work in progress.

Published results

ModelScoreVariantProvenance
Senior pharmaceutical scientists68.5% trace scorehuman comparison groupexternal
Biomni Lab (2026-02-03)65.0% trace scoreBiomni-DA-v0 · 15 tasksexternal
Edison Analysis52.7% trace scoreBiomni-DA-v0 · 15 tasksexternal
OpenAI Agents SDK (GPT-5.2)51.4% trace scoreBiomni-DA-v0 · 15 tasksexternal
Junior scientists48.5% trace scorehuman comparison groupexternal
Claude Code (Opus 4.6)47.8% trace scoreBiomni-DA-v0 · 15 tasksexternal

Sources

  • external Phylo — BiomniBench preliminary results