OpenAI publishes BioBasicsBench to test AI agents on wet-lab evidence

OpenAI said it published BioBasicsBench on November 6, 2026 to test whether AI agents can interpret raw wet-lab evidence instead of polished figures from papers. The benchmark has 100 tasks across five areas—protein methods, imaging and cytometry, quantification and assays, nucleic acid methods, and molecular cloning—and was built with Ligo from scientists’ own data, then reviewed multiple times. OpenAI said it ran 18,500 evaluations across 37 settings; at xhigh, GPT-6 Astra scored 66.7% versus 43.8% for GPT-5.6 Sol, though models still struggled with image interpretation and experimental context.