Open AI returns old models to Chat GPT as Sam Altman admits ‘bumpy’ GPT-5 rollout

OpenAI's co-founder and CEO Sam Altman has openly acknowledged the significant setbacks experienced during the launch of GPT-5, the company’s latest and most advanced large language model (LLM). During a recent Reddit AMA session and a post on social media platform X, Altman discussed various issues hampering the rollout, including problems with model switching, subpar performance, and user confusion. In light of these challenges, OpenAI has decided to reverse some of its recent platform changes, allowing users to access previous models like GPT-4 once again. Altman admitted that the rollout was "a little more bumpy than we hoped for," responding to inquiries about the launch's difficulties. He explained that human error contributed to some of the issues, particularly in regard to inaccurate model performance charts presented during the livestream event. Despite these setbacks, Altman maintained that the information provided in the accompanying blog post and system card was accurate. In response to the feedback from users, OpenAI announced several adjustments to the GPT-5 rollout. They plan to double the rate limits for ChatGPT Plus users and allow them the option to continue using GPT-4. Altman also mentioned that the company is analyzing user data to determine how long to keep the older models available. One of the significant issues identified by Altman was the malfunction of an automatic "router" designed to assign user prompts to one of four variants of GPT-5—regular, mini, nano, and pro—along with a "thinking" mode for more complex tasks. He noted that this autoswitcher was out of order for part of the day, leading to a perception that GPT-5 was less capable than intended. To address these concerns, OpenAI is making changes to improve model transparency and will introduce a user interface update to allow manual activation of the thinking mode. Despite the company’s internal benchmarks suggesting that GPT-5 outperforms its competitors, real-world users have reported numerous errors in basic tasks such as math, logic, and coding. For instance, data scientist Colin Fraser showcased GPT-5 incorrectly solving a simple mathematical problem, highlighting inconsistencies that have arisen since launch. With 700 million weekly users on ChatGPT, OpenAI remains the dominant force in generative AI. However, the recent challenges have highlighted the pressures of maintaining stability as demand surges. According to Altman, API traffic doubled in the 24 hours following the GPT-5 launch, exacerbating platform instability. Moving forward, OpenAI is committed to refining its infrastructure and addressing user concerns, as it strives to demonstrate that GPT-5 represents a significant advancement rather than just a minor update.