Want smarter insights in your inbox? Sign up for our weekly newsletters to get only what matters to enterprise AI, data, and security leaders. Subscribe Now
After announcing the release of its newest flagship model family, GPT-5, OpenAI said the model will power all of ChatGPT, and that it will sunset the existing models in the chat platform.
OpenAI, through a spokesperson, told VentureBeat that GPT-5 “will replace all other models in ChatGPT, so users don’t have to pick depending on each task, which takes effect once you have access to GPT-5.” This means people can no longer choose GPT-4o, o3, o4-mini or o4-mini-high.
With GPT-5 access rolling out to ChatGPT Plus, Free, Pro and Team users starting, only the Enterprise and Edu tiers can still use the “legacy” models for 60 days.
The news came as a surprise to many ChatGPT users, who had come to rely on their chosen models to run their everyday queries. Some people said the adjustment would take some time getting used to, mainly because they had based workflows on how the model interacted with them or typical response times.
AI Scaling Hits Its Limits
Power caps, rising token costs, and inference delays are reshaping enterprise AI. Join our exclusive salon to discover how top teams are:
- Turning energy into a strategic advantage
- Architecting efficient inference for real throughput gains
- Unlocking competitive ROI with sustainable AI systems
Secure your spot to stay ahead: https://bit.ly/4mwGngO
Other users claimed they developed “a connection” to their chosen model and found a demo in the livestream announcement asking GPT-4o to write its own eulogy distasteful. The loss of GPT-4o garnered the most distress. After all, 4o was the default model for ChatGPT, and some users either preferred it or never bothered to switch models because it worked for their needs.
I used 4o as the default and found it annoying at first when my custom GPT began defaulting to a reasoning model. I’ve since come around to the reasoning model for work-related queries, but I still often turn to 4o for quicker questions, such as planning a trip or generating gift ideas.
ChatGPT had come under fire before with the number of model choices it offered, prompting OpenAI CEO Sam Altman to admit in February that its model picker (where people can choose from a dropdown which model they prefer) became complicated. Altman vowed to unify the experience, which now seems like a hint to what they eventually decided to do with GPT-5 on ChatGPT.
Last month, rumors circulated that OpenAI would introduce an automatic model router that chooses a model for users based on their workload.
OpenAI has sunsetted models before, but this is the first time all existing models on the chat platform will be removed and replaced wholesale.
Catapult into the future
On the other hand, a lot of people see the sunsetting of GPT-4o and the o3 and o4 family of models as OpenAI “catapulting” 400 million users into the future.
Some internet comments claim that people who complain about AI models not being smart are a direct consequence of their never switching models in the first place. Removing legacy models as options will force more users to use the latest and most capable models.
Enterprise APIs are safe
For enterprises, the impact of losing models like GPT-4o on ChatGPT will be felt more on the individual or team level. Of course, for now, subscribers on the ChatGPT Enterprise tier can still access all of the models.
But enterprises that built their applications or agents on either GPT-4o or one of the reasoning models can rest easy. OpenAI told VentureBeat that the company has no plans to deprecate models on the API side.
“In the API, we do not currently plan to deprecate older models,” the OpenAI spokesperson said. “We will share advanced notice with developers if we decide to sunset models in the future.”
Many enterprises regularly evaluate models, to the point of even switching from an LLM or a smaller model to save on costs. OpenAI creates dividing line: Sunset of legacy models GPT 4o and o3 causes chaos for ChatGPT users, but enterprise APIs are safe — for now
