During a Reddit Ask Me Anything on Friday, OpenAI CEO Sam Altman and key members of the GPT-5 team fielded a barrage of questions, everything from requests to bring back GPT-4o to poking fun at a now-infamous “chart crime” in their launch presentation.
One of GPT-5’s big upgrades is a real-time router that chooses the best model for each prompt, either answering instantly or taking extra time to “think” for more complex responses. Sounds clever… except it did not work.
Altman admitted the glitch made GPT-5 seem “way dumber” than intended when it launched Thursday. The router was down for part of the day, forcing the model to give sub-optimal answers. “GPT-5 will seem smarter starting today,” he told Redditors, promising tweaks so users can see which model is responding to their queries.
Still, nostalgia for GPT-4o ran strong. Several users lobbied hard to bring it back for Plus subscribers. Altman did not shut the door, saying, “We are looking into letting Plus users continue to use 4o. We are trying to gather more data on the tradeoffs.”
Related: OpenAI Drops GPT-5 With Pricing That Could Shake the AI Market
To sweeten the rollout, Altman pledged to double rate limits for Plus users, giving them more prompts to test GPT-5 in real-world use cases.
And then there was the “chart crime.” During the live demo, a bar chart showed a lower benchmark score but a much taller bar, a visual gaffe that went viral within hours. Altman called it a “mega chart screwup” on X, clarifying that the blog post version had the correct charts. But the internet never forgets, and jokes about using GPT to make corporate decks spread fast.
Even GPT-5 reviewer Simon Willison, who generally praised the model, flagged table creation as a weak spot, calling it “a good example of a GPT-5 failure.”
Altman ended the AMA with a promise: “We will continue to work to get things stable and will keep listening to feedback.” The question now: will GPT-5 earn back trust… or will GPT-4o’s fan club keep growing?
