In their first episode of the New Year, Katherine Forrest and Scott Caravello unpack OpenAI’s release of GPT-5.2, covering its performance on benchmarks like Humanity’s Last Exam and on OpenAI’s own GDPval—an evaluation designed to test the model's ability to match or surpass professionals' performance on real world tasks. Our hosts also examine the model’s sharp drop in hallucinations and break down OpenAI’s discussion of the model’s resistance to prompt injections and how it stacks up under the company’s safety framework.

##

Learn More About Paul, Weiss’s Artificial Intelligence practice: