During the GPT-5 launch livestream, OpenAI presented several charts meant to highlight the model's impressive capabilities. However, closer inspection revealed inconsistencies in the graphs—such as mismatched bar sizes that didn't correspond to the actual data points. For example, GPT-5's 'coding deception' rate was shown as 50%, but the bar was smaller than a lower 47.4% score from an earlier model. In other cases, scores that were different were represented with bars of equal size, causing confusion. The errors drew attention on social media, and even OpenAI CEO Sam Altman acknowledged it as a 'mega chart screwup,' with an apology from the marketing team. While it's unclear if GPT-5 generated these charts, the incident cast a shadow over the launch, especially given the company's emphasis on reducing hallucinations with the new model.
