OpenAI may ship two models this year, but its flagship faces safety delays

OpenAI could release one or two models before year-end, but its next flagship may miss that window as the company works on alignment and testing. The model’s looped-transformer design can boost math and coding performance while making its reasoning harder to monitor. Even after internal safety reviews, U.S. government evaluation could delay access for outside partners by up to 30 days. Source

Read the full article at the source.