What's New in GPT-5? Explore OpenAI's Latest AI Model and Its Four Innovative Variants

GPT-5 is officially here. We already knew about the model from OpenAI because of leaks and teases from company CEO Sam Altman, but we now have confirmed details about it.

There are four variations of the GPT-5 model: gpt-5, gpt-5-mini, gpt-5-nano, and gpt-5-chat.

At the time of publication, OpenAI's stream unveiling the model is still going, so more details will emerge soon. The company's blog post breaks down performance, features, and more details about GPT-5.

The fact that there are four variants will disappoint some people. Rumors suggested that the release of GPT-5 would usher in a simplified naming structure. The growing number of OpenAI models can create confusion since average users may be unaware of the differences between each model.

But OpenAI has taken a step toward simplifying how its models are presented to users. GPT-5 will be presented as if it were just one model, but in the background it will automatically switch to the most appropriate model for your query.

GPT-5 is available as of today. All ChatGPT users can access the new model, though there is a cap on how much free users can use GPT-5. Once a free user hits that cap, ChatGPT will default to the less powerful gpt-5-mini.

Introducing GPT-5 - YouTube

Watch On

OpenAI will deprecate all previous GPT models with the release of GPT-5. The company even had GPT-5 write a eulogy for its predecessors.

A GitHub blog post that briefly appeared ahead of OpenAI's event outlined the highlights of the four new models:

• gpt-5: Designed for logic and multi-step tasks.
• gpt-5-mini: A lightweight version for cost-sensitive applications.
• gpt-5-nano: Optimized for speed and ideal for applications requiring low latency.
• gpt-5-chat: Designed for advanced, natural, multimodal, and context-aware conversations for enterprise applications.

"OpenAI's most advanced model, offering major improvements in reasoning, code quality, and user experience," said the GitHub post.

Ahead of the GPT-5 announcement, Altman hyped up the model using an unusual strategy. The CEO suggested he is scared of GPT-5 and compared the model to the Manhattan Project:

"There are moments in the history of science where you have a group of scientists look at their creation and just say, you know, ‘What have we done?’"

Altman also shared a picture of the Death Star from Star Wars before the company's event for GPT-5.

Altman's tactics are similar to when a car salesman says a dealership's "prices are so low, they're scary." Presumably, he does not believe GPT-5 is going to destroy Earth, Alderaan, or any other planet.

He has stated that "someday soon something smarter than the smartest person you know will be running on a device in your pocket, helping you with whatever you want."

It doesn't seem like the "someday" Altman referred to is today, or he likely would have been more direct. Though his comparison to a team of PhD-level students is high praise for GPT-5.

Coding with GPT-5

During the presentation unveiling the new model, OpenAI's team called GPT-5 its best coding model to date. The presenters had the tool generate hundreds of lines of code for a website.

OpenAI shared that it tested GPT-5 for over 5,000 hours.

ChatGPT, Gmail, and Google Calendar

Starting with the release of GPT-5, ChatGPT can integrate with Google Calendar and Gmail to help plan your day. The feature will be available first for Pro users. It will then roll out to Plus, Team, and Enterprise users.

This is the latest step in making ChatGPT a more personal assistant. OpenAI added support for memory early in 2024.

During the demo, ChatGPT found an unread email and suggested a schedule to the presenter.

GPT-5 vs. GPT-4

It should come as no surprise that GPT-5 is better than GPT-4. Tech tends to trend that way, after all.

But considering Altman called GPT-4 "mildly embarrassing at best" and said GPT-4 "kind of sucks," a jump in quality seemed necessary.

Altman promised that GPT-5 would be "materially better" than GPT-4, and that seems to be the case based on the figures shared by OpenAI.