OpenAI Swapped the Model Inside Codex Without Telling Anyone. Developers Found Out by Accident.

Editor's take

OpenAI deployed an undisclosed model, apparently a precursor to GPT-5, to paying Codex users, a fact discovered by developers through observed behavior and a significantly expanded context window. This move raises concerns because it impacts the predictability of AI tools crucial for development workflows, potentially affecting how applications built on Codex function without prior notification.

The lack of transparency here is particularly noteworthy given the industry's increasing reliance on these models for critical infrastructure. Developers often tune their applications based on specific model characteristics, and unexpected shifts can lead to performance degradation or outright failures. This incident underscores the tension between rapid iteration and open communication in the LLM space.

Future developments will reveal whether OpenAI implements clearer communication protocols for model updates, especially for paid services. The industry will also be watching how quickly other foundational model providers, like Anthropic with its Claude models, communicate changes to their deployed systems to maintain trust with their developer communities.