OpenAI’s much-anticipated GPT-5 launch hasn’t been without its hiccups. During a Reddit AMA on Friday, CEO Sam Altman and members of the GPT-5 team fielded a barrage of questions — many of them nostalgic pleas for the return of the previous model, GPT-4o.
The Real-Time Router Glitch
One of GPT-5’s flagship features is a real-time router, designed to determine which model should handle a prompt — whether to respond instantly or take more time for deeper reasoning.
However, Altman revealed that a malfunction in this system was partly to blame for early frustrations.
“Yesterday, we had a sev and the autoswitcher was out of commission for a chunk of the day… GPT-5 seemed way dumber,” Altman admitted.
The fix is now in place, and OpenAI is also tweaking how the decision boundary works to ensure the right model is selected more often. Transparency about which model is answering a query is also on the roadmap.
Calls for GPT-4o’s Return
Reddit users didn’t stop at bug reports — they lobbied heavily for GPT-4o’s reinstatement. Many said it handled their use cases better. Altman responded by confirming the team is evaluating the possibility of making GPT-4o available again for Plus subscribers:
“We are looking into letting Plus users continue to use 4o. We are trying to gather more data on the tradeoffs.”
ALSO SEE : Apple Boosts AI Investments, Plans Major Features
Boosting Plus User Benefits
In a bid to smooth adoption, Altman announced that Plus subscribers will see double rate limits once the rollout stabilizes. This should give users more room to explore GPT-5’s capabilities without worrying about hitting monthly prompt caps.
The “Chart Crime” Incident
No tech launch is complete without an unintended meme. For GPT-5, it came in the form of a presentation slide with a glaring visual blunder — a bar chart where a lower benchmark score was represented with a much taller bar.
Social media quickly dubbed it a “chart crime.” Altman called it a “mega chart screwup” on X, clarifying that the official blog post contained correct charts. Still, the damage — and the jokes — were already circulating.
Early reviewer Simon Willison, despite generally praising GPT-5, noted that even converting data into a simple table was a weak spot for the model.
OpenAI’s Promise
Altman closed the AMA on a reassuring note:
“We will continue to work to get things stable and will keep listening to feedback.”
For now, OpenAI is focusing on stabilizing GPT-5, restoring user trust, and possibly giving nostalgic fans of GPT-4o another chance to use their favorite model.
Sources ( Techcrunch )
