OpenAI CEO Sam Altman has provided an update on the rollout of the company's latest large language model, GPT-5, acknowledging unexpected user preferences for its predecessor, GPT-4o, and anticipating a significant capacity challenge in the coming week. The announcement comes shortly after GPT-5's official unveiling on August 7, which touted the model as "smarter, faster, and more useful."
GPT-5 was launched with promises of PhD-level expertise and enhanced capabilities in coding, writing, and complex actions, with Altman describing it as a "significant step along the path to AGI." However, initial user experiences have varied, with some expressing a preference for the "vibe" and specific qualities of GPT-4o. Reports indicated that an "autoswitcher" issue during the early rollout contributed to perceptions of GPT-5 performing below expectations for some users.
Altman admitted that OpenAI had "underestimated how much some of the things that people like in GPT-4o matter to them, even if GPT-5 performs better in most ways." He further noted the "very different opinions on the relative strength of GPT-4o vs GPT-5." This feedback has reinforced the company's long-term commitment to offering extensive user customization, recognizing that:
there isn't one model that works for everyone.
The immediate focus for OpenAI is to finalize the GPT-5 rollout, which now reaching 100% of Pro users and nearing completion for all users. Following stabilization, the team plans to implement changes aimed at making GPT-5 "warmer" in its interactions. While comprehensive per-user customization will take longer to develop, the company has been investing in steerability research and has previewed different personality options.
Despite the ongoing efforts, Altman warned of an impending "severe capacity challenge for next week," indicating potential service disruptions. He assured users that OpenAI would be transparent about the principles guiding decisions on resource allocation, stating:
Not everyone will like whatever tradeoffs we end up with, obviously, but at least we will explain how we are making decisions.
This operational hurdle follows previous expressions of concern by Altman regarding the rapid advancement of AI, which he has likened to the Manhattan Project.