OpenAI: GPT-5 demonstrates the ability to assist scientific research in real laboratory environments

According to Jinse Finance, AXIOS reported that OpenAI has revealed for the first time that GPT-5 has demonstrated the ability to assist scientific research in real laboratory environments, indicating that artificial intelligence is expected to play a more central role in scientific experiments. Although AI has made rapid progress in fields such as mathematics and physics, major breakthroughs in biology have been relatively slow, as biology heavily relies on real-world laboratory work rather than just computational simulations. OpenAI collaborated with the biosecurity startup Red Queen Bio to build a testing framework to evaluate the performance of AI models in laboratories. The laboratory involved "wet" operations such as liquids, chemicals, and biological samples, distinguishing it from data analysis-focused "dry" labs. In the experiment, GPT-5 proposed improvements to the molecular cloning experimental process, which were executed by human scientists, and the results were fed back to GPT-5, which then iteratively optimized the plan based on the feedback. The experimental results showed that GPT-5 improved the efficiency of a standard molecular cloning process by 79 times.