OpenAI Launches BioBasicsBench to Test AI Agents on Real Wet Lab Analyses
A new way to measure how AI agents handle the routine analyses of wet-lab science—BioBasicsBench. Developed with Ligo, a biotech company focused on enzyme design, the benchmark draws on 100 tasks from real experiments to test whether agents can interpret raw lab data, identify errors, and recognize when the evidence doesn’t support a conclusion. Accelerating biological research starts with getting the everyday work right. BioBasicsBench shows where models need to improve, and helps us track what those gains mean for scientific progress, biosafety, and biosecurity.
Context
It measures whether agents can do the everyday lab work that real biology depends on, not just answer textbook questions.
OpenAI built the benchmark with Ligo, a biotech company focused on enzyme design. Its 100 tasks come from real experiments and test whether agents can read raw lab data, spot errors and recognize unsupported conclusions. OpenAI says it will help track progress for science and for biosafety.
Deep dive
OpenAI introduced BioBasicsBench on Friday, a benchmark for how AI agents handle the routine analyses of wet lab science. It was developed with Ligo, a biotech company focused on enzyme design, and draws on 100 tasks from real experiments.
The tasks test whether an agent can interpret raw lab data, identify errors and recognize when the evidence does not support a conclusion. Posts relaying the release describe inputs such as gel images and instrument readouts rather than clean text questions.
OpenAI framed the benchmark as a way to see where models still need to improve and to track what gains mean for scientific progress, biosafety and biosecurity. That dual framing matters because better lab skills can help both research and misuse.
The launch fits a week of science and security releases from OpenAI, including GPT 6 Cyber on Nov 5. Scores for specific models were not in the post, and OpenAI did not say how its own models performed.
What to expect next
Watch for published scores across frontier models and for outside labs to adopt the benchmark.
Also watch how OpenAI uses the results in its biosafety evaluations, and whether Ligo or others release more tasks. A key question is whether models that score well here also cut real lab mistakes, and how OpenAI balances wider access to strong lab skills against biosecurity risk. Results from other labs on the same tasks would show whether OpenAI's models lead.