The news, 365 days behind — on purpose Delayed live · replaying 2025

One Year Ago.AI

Remember how fast this is.

08AUG2025replayed
one year on
productOpenAI · Sam Altman

Altman addresses bumpy GPT-5 rollout, promises return of GPT-4o for Plus users

One day after launching GPT-5, Sam Altman uses a Reddit AMA to address user requests to bring GPT-4o back to Plus, acknowledge a busted model router, and double the Plus rate limit.

OpenAI CEO Sam Altman and members of the GPT-5 team held a Reddit AMA today, one day after the launch of GPT-5, to address a wave of user backlash. Many users complained GPT-5 wasn’t working as well for them as 4o did, and asked for more model choice.

Altman admitted the rollout was bumpy, explaining that an automatic model router, or ‘autoswitcher,’ suffered a severity outage yesterday, making GPT-5 seem much dumber than it is. He promised the model will seem smarter starting today, and said OpenAI would make it more transparent about which model is answering a given query. To placate users, Altman said OpenAI is looking into letting Plus users continue to use GPT-4o and will roughly double rate limits for Plus users.

He closed the AMA promising to keep working on stability and to keep listening to feedback.

R
Reddit user

User wrote 'BRING BACK 4o GPT-5 is wearing the skin of my dead friend'

S
Sam Altman

Altman replied that OpenAI hears users and is working on bringing back 4o

S
Simon Willison

Pointed out that turning data into a table was a good example of a GPT-5 failure

One year later — open only if you can handle spoilers

OpenAI made the reversal official two days later, restoring GPT-4o for all paying users and promising "plenty of notice" before any future deprecation; within a week it also added Auto, Fast, and Thinking modes and kept several legacy models on tap. Altman would later concede the company "totally screwed up" the rollout. The episode became the canonical proof that users bond with specific models—capability alone doesn't govern a launch.

Replay thisPost on XRedditHNLinkedIn

The Weekly Replay · free by email

This week, one year ago — every Sunday.

One email each Sunday: the week's replayed AI news, with the one-year-later annotations included. Written like it's breaking — dated like it isn't.

Free · double opt-in · unsubscribe anytime · privacy