OpenAI and biotech firm Ligo released BioBasicsBench on November 6, 2026, a 100-task benchmark that tests AI agents on routine wet-lab work using raw biological data including gel photos, colony plates, microscopy images, sequencing reads, and bench videos.
Agents are scored on interpreting data, identifying errors, and recognizing when evidence does not support a conclusion, with rubrics awarding partial credit. GPT-6 Astra scored 66.7% against 43.8% for GPT-5.6 Sol, with Astra using roughly half the output tokens per task. OpenAI framed the benchmark as a way to track what model gains mean for scientific progress, biosafety, and biosecurity.