OpenAI tackles GPT-5 rollout issues

OpenAI faces a bumpy GPT-5 rollout as users push to restore GPT-4o and a chart error fuels public chatter.

OpenAI tackles GPT-5 rollout issues and mulls GPT-4o return

During a Reddit AMA on Friday, OpenAI CEO Sam Altman and the GPT-5 team fielded questions about the new model. A real-time router now decides which model should respond to a prompt, aiming to balance speed with depth. Some users reported GPT-5 performing worse than GPT-4o, and Altman said the dip was caused by an autoswitcher outage that affected the rollout. He promised improvements and promised more transparency about which model answers a given query.

Altman also signaled that GPT-4o could be made available again for Plus subscribers while the team gathers more data on the tradeoffs. He said rate limits for Plus users would be doubled as rollout continues, giving people room to test the new system. The AMA followed a wave of jokes about a live presentation chart that became known as chart crime, a misstep Altman later called a mega chart screwup. He stressed that the published blog charts remained correct and that the team would push for better accuracy and clarity as the rollout proceeds.

The router that chooses which model answers a prompt is a bold move with big risks. It changes who sees which capability and when, making reliability the measuring stick for success. The public will judge the project by consistency and transparency, not just speed or flashy features.

OpenAI’s willingness to consider restoring GPT-4o for Plus users and to expand rate limits signals a pragmatic response to feedback. The chart misstep reveals the pressure of presenting numbers publicly before the data is fully stable. If OpenAI wants lasting trust, it must explain how model choice works and how users can anticipate which version they will get.

Highlights

The road to reliability is long, and the next updates will reveal how far the system has traveled.