OpenAI co-founder and CEO Sam Altman has acknowledged several significant issues that arose during the launch of GPT-5, the company’s latest large language model (LLM). In a recent Reddit AMA and a post on X, he detailed problems such as faulty model switching, poor performance, and user confusion, which have prompted OpenAI to revert some recent platform changes and restore access to earlier LLMs like GPT-4o.
Altman described the launch as “bumpy” and attributed some of the missteps to a new automatic router system designed to assign user prompts to different variants of GPT-5. He noted that this system experienced malfunctions that caused the model to perform below expectations. Moving forward, OpenAI plans to enhance the model decision boundary and enhance transparency in indicating which variant is responding to user queries.
Additionally, in direct response to user concerns regarding GPT-5’s inconsistent performance, OpenAI will allow ChatGPT Plus users to continue using GPT-4o while it gathers data on the advantages and disadvantages of the updated model.
Despite claims of superior performance in internal benchmarks, users have reported a number of basic errors in math, logic, and coding tasks since the launch. For instance, instances of GPT-5 incorrectly resolving simple mathematical problems have been documented, alongside developer feedback indicating its struggles with certain programming tasks compared to competitors.
The recent rollout has also led to increased API traffic, which Altman indicated contributed to instability. In an effort to address these challenges, OpenAI is adjusting rate limits and infrastructure. Nevertheless, the early issues, combined with unexpected user experience changes, have raised questions about the platform’s capabilities and opened opportunities for competitors in the generative AI space.
Source: https://venturebeat.com/ai/openai-returns-old-models-to-chatgpt-as-sam-altman-admits-bumpy-gpt-5-rollout/