1Cademy - Measuring how AI agents handle everyday wet-lab science

Learn Before

AI-Assisted Discovery and Model Research References

Reference

Measuring how AI agents handle everyday wet-lab science

Source: https://openai.com/index/measuring-how-ai-agents-handle-wet-lab-science/

Published: 2026-11-06. Inspected version: Developer benchmark report. Retrieved: 2026-11-09T22:05:20.437463+00:00.

Source annotation (reported findings): OpenAI’s November 6, 2026 developer report describes BioBasicsBench, developed with Ligo: 100 routine laboratory tasks using raw experimental data. Testing 37 model-and-reasoning-effort settings with five attempts per task produced 18,500 answers. Scores award credit for individual grading criteria; GPT-6 Astra at the reasoning-effort setting called xhigh averaged 66.7% of available credit, not the share of tasks solved completely. Internal and external scientists reviewed task materials and a graded model attempt; this does not establish human review of all answers. No measured human-performance comparison or checking-time outcome is provided.

0

1

Updated 2026-11-10

Contributors are:

Who are from:

Learn Before

Related

Related