OpenAI introduces benchmark for AI lab science

OpenAI has introduced BioBasicsBench, a benchmark that measures how well AI agents can interpret raw experimental data from routine wet-lab scientific work. The benchmark includes 100 tasks across five areas of laboratory work, using real experimental evidence like gel images, microscopy data, and instrument readouts to test whether agents can identify errors, recognize when data is insufficient to draw conclusions, and correctly analyze results. This matters because misreading images or overlooking experimental problems can lead scientists to draw wrong conclusions and waste time on unproductive experiments. Testing on the benchmark showed that newer AI models perform better at these tasks, with GPT-6 Astra scoring 66.7 percent compared to GPT-5.6 Sol's 43.8 percent, though there is still room for improvement in image interpretation, control checking, and applying experimental context.