OpenAI CEO Sam Altman has responded to criticism over GPT-5’s bumpy rollout, promising technical improvements, higher usage limits, and possibly bringing back GPT-4o for Plus subscribers.
In a Reddit AMA on Friday, Altman said a faulty “real-time router” — designed to pick the best model for each prompt — malfunctioned on launch day, making GPT-5 seem “dumber” than it should. OpenAI will tweak the system and make it clearer which model is responding.
Possible return of GPT-4o
Many users said GPT-4o worked better for them. Altman said OpenAI is considering letting Plus users keep using it while they gather more performance data. Rate limits for Plus subscribers will also be doubled during the rollout.
Presentation blunder goes viral
The AMA reignited jokes about the “chart crime” in GPT-5’s launch presentation, where a benchmark chart showed a lower score with a taller bar. Altman had earlier called it a “mega chart screwup,” and corrected versions were later posted online.
Mixed reviews so far
Some early testers praised GPT-5’s abilities, though others noted flaws — like struggles converting data into tables. Altman ended by assuring users: “We will continue to work to get things stable and will keep listening to feedback.”
