Just days after GPT-5’s launch, OpenAI CEO Sam Altman admitted the new AI model fell short of expectations — and is giving Plus subscribers the option to return to GPT-4o, along with double rate limits as compensation.
The admission came during a lively Reddit AMA, where Altman blamed part of GPT-5’s rocky debut on a malfunctioning real-time router — a feature meant to choose between fast or “thoughtful” responses. With the autoswitcher offline, GPT-5 appeared “way dumber” than intended. Fixes rolled out immediately, with improvements to the model’s decision-making and clearer indicators of which AI is responding.
User backlash was swift, with many claiming GPT-4o still worked better for their needs. Altman relented, confirming that Plus subscribers would soon be able to keep using it while OpenAI collects more data on tradeoffs.
The AMA also revived talk of the launch’s “chart crime” — a botched graph from the GPT-5 presentation that Altman had previously called a “mega chart screwup” on X. While the slides may have been messy, OpenAI insists the blog post numbers were correct.
In follow-up posts, the company promised stability improvements and an easier way to manually trigger GPT-5’s ‘thinking mode.’ Altman, speaking separately on a podcast, admitted GPT-5’s problem-solving abilities left him “rattled,” comparing its creation to the Manhattan Project and warning that AI development is outpacing regulation.
Despite the bumps, GPT-5 boasts enhanced multimodal reasoning, extended memory, better multi-step logic, and a 256,000-token context window. But tensions remain high — Microsoft, which has invested $13.5 billion, reportedly wants more say in OpenAI’s direction, while criminals are already exploiting generative AI for fraud.
For now, GPT-5 is getting quick fixes, Plus users have more prompts, and GPT-4o is making a nostalgic comeback. Whether the new model is truly “smarter starting today” is still an open question — but Altman’s transparency shows even cutting-edge AI launches can have… very human flaws.
