ChatGPT-5 is now rolling out. What happens to previous models like GPT-4o and o3?
- Graziano Stefanelli
- Aug 8
- 3 min read
OpenAI is replacing all previous ChatGPT models with GPT-5, beginning with Plus and Pro users and expanding worldwide in stages.
On August 7, 2025, OpenAI officially began rolling out ChatGPT-5, a new flagship model that replaces GPT-4o, OpenAI o3, o3-pro, and all other prior versions inside the ChatGPT interface.
According to the launch announcement, GPT-5 is designed to be the "most intelligent, fast, and helpful model ever released" by the company. It is not a continuation of the GPT-4 series but a new generation, and it is now being automatically assigned to all ChatGPT users over time.
GPT-5 replaces all previous versions inside ChatGPT.
GPT-5 is not offered as a selectable model next to older ones. Instead, it becomes the default model, and will automatically take over all new chats started in the ChatGPT app or web interface. Here’s how that affects different user groups:
Free users are gradually being moved to GPT-5 (in most countries, this is still ongoing as of August 8). When their message quota is exceeded, the system may fall back to a lighter variant, referred to internally as GPT-5 mini, designed to remain fast and available without significant hallucinations.
Plus users already have access to GPT-5 by default in most regions. There is no longer a model picker; GPT-5 is always on.
Pro, Team, Enterprise, and ChatGPT Edu users not only have GPT-5 by default but also gain access to a more capable internal variant—referred to in OpenAI infrastructure as GPT-5 Pro—which is more accurate and better at deep reasoning, especially for long documents, strategic analysis, and code.
API users can still explicitly request older models like gpt-4o or o3 for now, but it is expected that OpenAI will phase out those model endpoints gradually. No firm sunset date has been published as of today.
New behavior: GPT-5 adapts based on your instructions.
The new system uses what OpenAI calls a dynamic routing mechanism: if you ask the model to “think carefully,” “take your time,” or give “step-by-step” reasoning, it may internally switch to a slower, more accurate inference path, akin to a high-capacity version of itself. This effectively allows you to choose between fast and deep outputs without manually changing models.
You do not need to select a mode explicitly — but you can influence the behavior by writing prompts like:
“Think deeply before answering this.”“Take it slow and explain in multiple steps.”“Prioritize precision over speed.”
Is this the end of models like GPT-4o or o3?
In ChatGPT? Yes, almost entirely.
While technically those models still exist in the backend infrastructure, they are being fully replaced in the ChatGPT app by GPT-5. All your existing chats with GPT-4 or o3 are still accessible, but once you continue them, the new responses will be written by GPT-5—even if the old messages were generated by a previous model.
For developers using the OpenAI API, o3 and 4o endpoints remain temporarily available, but this is expected to change in the coming months.
No need to say goodbye — but everything is changing under the hood.
You don’t have to leave your favorite chat interface or change how you use ChatGPT. But yes, behind the scenes, the model you’re talking to is evolving into something more powerful, more precise, and more capable of understanding complex requests.
If you're still seeing OpenAI o3 or GPT-4o when you open ChatGPT, don’t worry — it simply means the rollout hasn't yet reached your region or account tier. It will arrive soon, and when it does, all new chats will automatically start with GPT-5.
________
FOLLOW US FOR MORE.
DATA STUDIOS
