Measuring How AI Agents Handle Everyday Wet-Lab Science
OpenAI, Friday, November 6th, 2026
OpenAI introduces BioBasicsBench, 100 routine wet-lab tasks testing whether AI agents can interpret raw experimental data.
OpenAI and biotech company Ligo built BioBasicsBench, a benchmark of 100 routine wet-lab tasks across protein methods, imaging and cytometry, assays, nucleic acid methods and molecular cloning.
Tasks use raw evidence such as gel images, colony plates, microscopy, sequencing files and instrument exports, and are scored against expert rubrics that reward recognizing errors and unsupported conclusions.
At the xhigh reasoning setting, GPT-6 Astra scored 66.7% of available credit, compared with 43.8% for GPT-5.6 Sol, which OpenAI credits to better multimodal handling and use of experimental context.
It also shows remaining failures, such as letting assumptions override controls, discounting contradictory evidence and miscounting colonies.