OpenAI is retiring GPT-4.5 and o3 from ChatGPT — here's what changes and when
GPT-4.5 leaves ChatGPT on June 27, 2026 and o3 on August 26. No API changes. Here's what it means for you.
OpenAI has confirmed two upcoming model retirements from ChatGPT: GPT-4.5 will be removed on June 27, 2026, and o3 will follow on August 26, 2026. These are ChatGPT-only changes — the API is unaffected, for now.
If you are on a free ChatGPT plan, you will not notice anything different. These models are only available to paid subscribers via the model picker, and most free users are already running on GPT-5.5 anyway. For paid users who have been deliberately choosing GPT-4.5 or o3 for specific tasks, the clock is now ticking.
What’s actually being retired, and why
GPT-4.5 launched in February 2026 as OpenAI’s most knowledgeable model at the time. It was widely praised for its broad knowledge base and natural, smooth conversational quality. Four months later, it is being retired. That is a short lifespan by any standard.
The reason is straightforward: GPT-5 and GPT-5.3-Codex have overtaken GPT-4.5 on every meaningful benchmark, and GPT-5 is also faster and cheaper per token. Keeping GPT-4.5 in the model picker adds maintenance overhead without adding much value for most users. OpenAI is making the same call it made with GPT-4o earlier this year.
o3 is a slightly different story. It was OpenAI’s dedicated reasoning model before the GPT-5 series brought reasoning capabilities into the core product. Its successor, o4, is faster, cheaper, and more accurate on reasoning tasks. o3 will disappear from ChatGPT in August but will remain available through the API for the foreseeable future.
The end of the GPT-4 era in ChatGPT
This retirement is worth marking. GPT-4.5 is the last GPT-4 family model available to ChatGPT end users. With its removal, the GPT-4 generation is fully gone from the ChatGPT interface.
GPT-4 and its variants, especially GPT-4o, were the models that genuinely changed how a lot of people thought about AI. The transition from GPT-4o to GPT-5 in ChatGPT earlier this year drew real pushback from users, some of whom preferred the older model’s tone and personality. OpenAI actually brought GPT-4o back temporarily in response to that feedback before removing it again once GPT-5 had matured. That feedback reportedly shaped some of the personality and creative improvements in GPT-5.1 and 5.2.
The pattern is consistent: OpenAI is shipping new models faster than ever and retiring older ones to reduce complexity and keep the ecosystem focused on its current generation. GPT-4o and several related models were deprecated from ChatGPT back in February 2026. GPT-5.1 variants followed in March. GPT-4.5 and o3 are next.
What this means if you use ChatGPT
For most ChatGPT users, nothing changes. If you are on a paid plan but have never manually changed the model in settings, you are likely already using GPT-5 by default and will not be affected.
If you are a paid user who has been deliberately selecting GPT-4.5 or o3, those options will simply disappear from the model picker on their respective retirement dates. ChatGPT will route you to the current default model automatically. No action is required on your part.
Some users on X have expressed frustration about this, with one commenting that “4.5 is still the best writing model” and that the GPT-5 series does not match what GPT-4.5 and o3 offered for certain tasks. That sentiment is real. Power users do notice differences in tone, writing style, and reasoning approach between models, and not everyone agrees that newer automatically means better for their specific use case. If that is you, the June 27 and August 26 dates are your deadlines to make the most of those models while they are still accessible.
What this means if you build with the API
This is the part that actually requires attention. These retirements only apply to the ChatGPT interface, but there is a practical warning worth absorbing here.
o3 will remain available in the API after August 26. GPT-4.5 was already removed from the API previously, so that is not a new issue for developers. However, the broader trend is clear: OpenAI’s model lifecycles are compressing. A model released today might be deprecated within twelve months. If you are building on any OpenAI model and have hardcoded the model name as a string in your application, you are setting yourself up for a broken integration down the line.
The practical fix is simple: store the model name in a configuration value or environment variable rather than in the code itself. That way, when a model is retired, you can update it in one place without a full deployment. It takes ten minutes to do correctly and saves a headache later.
The bigger picture
OpenAI is consolidating around the GPT-5.x generation. GPT-5.4 is rolling out now with integrated reasoning, coding, and agentic capabilities, and GPT-5.4 mini is already available to Free and Go users via the Thinking feature. The company is clearly building toward a smaller number of more capable models rather than maintaining a long menu of specialised options.
That direction makes sense operationally. A single frontier model that handles reasoning, coding, and conversation well is easier to maintain and easier for users to understand than a growing list of named variants each suited to different tasks.
The trade-off is that users lose fine-grained control over model selection. For most people, that is a reasonable exchange. For the subset of power users who have built workflows around specific model personalities or capabilities, it is a genuine inconvenience, and it is worth being aware of these dates so you are not caught off guard.