OpenAI Addresses GPT-5 Rollout Issues, Promises Fixes and Model Reversion

During a Reddit AMA, OpenAI CEO Sam Altman and the GPT-5 team addressed concerns over the model's rollout, including its underperformance and a widely mocked "chart crime." Users reported that GPT-5 seemed "dumber" than its predecessor, GPT-4o, due to a malfunctioning real-time router that failed to properly allocate responses between models. Altman acknowledged the issue, stating that the router was out of commission for part of the day and that interventions were being made to improve model selection.
The "chart crime" incident, where a misleading graph was presented during the GPT-5 launch, also sparked widespread criticism. Altman referred to it as a "mega chart screwup" on X, though corrected charts were later published.
In response to user demands, Altman hinted at reintroducing GPT-4o for Plus subscribers, emphasizing that tradeoffs were being evaluated. Additionally, he promised to double rate limits for Plus users, providing them more flexibility to experiment with the new model.
Altman concluded the AMA by pledging continued efforts to stabilize GPT-5 and listen to feedback, ensuring the model meets user expectations.

Published: 8/10/2025