On May 29, 2026, OpenAI quietly announced in the ChatGPT release notes that it will gradually retire two major models: GPT-4.5 and OpenAI o3. This decision marks OpenAI's formal shift into a new phase centered on the GPT-5.5 series and has sparked discussion and reflection among some users and developers.

According to OpenAI's official notice, GPT-4.5 will be removed from the ChatGPT web and mobile interfaces on June 27, 2026 with a 30-day transition period; OpenAI o3 will be retired on August 26, 2026 with a 90-day transition period. During the transition, paying users can still manually switch to these models via the model selector, but they will no longer be accessible in the ChatGPT interface after removal.

Notably, these retirements apply only to the ChatGPT consumer interface; API services are not affected. Developers can continue to call GPT-4.5 and o3 through the API.

Reason for retirement: resource optimization and strategic focus

OpenAI said in the release notes that the move is intended to "phase out older, lower‑usage models in order to better serve newer, more capable ones." The GPT-5.5 series (including GPT-5.5 Instant, Thinking and Pro) has become the company's current development priority; concentrating resources will help improve overall service quality and efficiency.

The decision aligns with OpenAI's recent product strategy: simplifying model choices to reduce user decision costs while shifting compute and engineering resources toward the latest generation. GPT-4.5 and o3, representatives of the GPT-4 and o-series eras, contributed substantial users and revenue to OpenAI, but their usage has declined significantly as GPT-5.5 matures.

User and developer reactions

After the announcement, the news sparked wide discussion on communities such as Reddit and X. Many users expressed regret over the retirement of GPT-4.5, calling it "the last truly useful GPT-4 model," citing its strengths in writing, complex reasoning and stability. o3, as an early reasoning model, was also regarded by some as one of the strongest "thinking" options.

However, some developers understood the move, noting that keeping too many legacy models increases maintenance costs, and that the GPT-5.5 series already shows clear overall performance improve