Sam Altman addresses ‘bumpy’ GPT-5 rollout, bringing 4o back, and the ‘chart crime’

During a Reddit Ask me. Session on Friday, OpenAi CEO Sam Altman and important members of the GPT-5 team were peppered with questions about the new model and requests to reduce his previous model, GPT-4O.

They also asked Altman about the most embarrassing – and perhaps funniest – Snafu in the presentation, the ‘card crime’.

One of the new functions that GPT-5 has rolled out is a real-time router that decides which model to use for a certain prompt, respond quickly or take extra time to ‘think’ through answers.

But several people in the AMA on the R/Chatgpt Reddit complained GPT-5 did not work as well for them as 4o. Altman said that the reason that GPT-5 seemed that “dumber” was that the router did not work well when it was rolled out on Thursday.

“GPT-5 will seem smarter from today. Yesterday we had a SEV and the car switcher was out of use for part of the day, and the result was GPT-5 seemed much more stupid. We also make a number of interventions how the decision-making limit works that you should get more often.

Yet people on the AMA lobby so hard to bring 4O back for Plus subscribers that Altman promised to look at it at least. “We look at leaving users to keep using 4O. We try to collect more information about the considerations,” he wrote.

And Altman also promised: “We are going to finish the limits for Plus users double rate when we finish.” This should give people the opportunity to play and learn the new model, to take over it from their use cases without worrying about noticing monthly instructions.

He was also asked for the wild inaccurate graph that the team presented during the live presentation that quickly became the butt of many “Chart Crime” jokes. The graph presented a lower benchmark score with a much larger bar.

Altman did not answer any questions about the graph during the AMA, but on Thursday he called the graph a “mega assessment” on X. Others noted the graphs in the Published blog post was correct.

But the damage has been done. Jokes followed about the use of GPT for graphs in a company presentation. GPT-5 reviewer Simon Willison, who had early access and generally liked the performance of the model, Also on former That converting data into a table “was a good example of a GPT-5 failure.”

In any case, Altman promised solutions to the items that make people the most. He ended the AMA with a promise: “We will continue to work to get things stable and will continue to listen to feedback.”