OpenAI’s GPT-5 launch was not as smooth as expected. Friday, during a Reddit AMA, CEO Sam Altman and the GPT-5 team faced a barrage of questions, from the performance of the model to a very troublesome chart that has since gone viral, and has become known as a “chart crime.”
One of GPT-5’s key characteristics is its router, which, in a real-time mode, chooses the most appropriate model for a given prompt. But when launched on Thursday, the router was not functioning properly. Therefore, GPT-5’s performance considerably reduced, and this was recognized by Altman, who reported, “We had a sev and the autoswitcher was out of commission for a chunk of the day.” The fixes, which are currently underway, include shifting the decision boundary and adding transparency about which model is answering.
People have shown a preference for bringing back GPT-4o, since a lot of Plus subscribers said GPT-4o worked better for them. As a result, Altman said, “We're considering allowing Plus members to keep using 4o. Trying to get more data on the tradeoffs.” He also promised to double the rate limits for Plus members to allow them to experiment on GPT-5 without hitting the monthly caps.
The "Crime Heard 'Round the Web" chart was shown off by OpenAI during its live demo. The chart featured a lower benchmark score and a higher bar, causing mirth on platforms like X and Reddit. Altman called it a “mega chart screwup,” and while the next blog post did show the right one, the damage was already done. Simon Willison, one of the reviewers of GPT-5, concurred that this table layout was a “good example of a GPT-5 failure.”
Despite the challenges he faced, Altman ended the AMA on a promise:
"We will continue our efforts to make it stable and watch for any anticipated reaction."