GPT-5: everything we know about OpenAI’s next frontier model

There’s perhaps no product more hotly anticipated in tech right now than GPT-5. Rumors about it have been circulating ever since the release of GPT-4, OpenAI’s groundbreaking foundational model that’s been the basis of everything the company has launched over the past year, such as GPT-4o, Advanced Voice Mode, and the OpenAI o1-preview.

Those are all interesting in their own right, but a true successor to GPT-4 is still yet to come. Now that it’s been over a year a half since GPT-4’s release, buzz around a next-gen model has never been stronger.

When will GPT-5 be released?

OpenAI has continued a rapid rate of progress on its LLMs. GPT-4 debuted on March 14, 2023, which came just four months after GPT-3.5 launched alongside ChatGPT. OpenAI has yet to set a specific release date for GPT-5, though rumors have circulated online that the new model could arrive as soon as late 2024.

Then again, some were predicting that it would get announced before the end of 2023, and later, this summer. I wouldn’t put a lot of stock in what some AI enthusiasts are saying online.

In June, former CTO Mira Murati affirmed that the next-gen model was still a year and a half out from release, which would certainly rule out a 2024 timeframe.

i have been told that gpt5 is scheduled to complete training this december and that openai expects it to achieve agi.

which means we will all hotly debate as to whether it actually achieves agi.

which means it will.

— Siqi Chen (@blader) March 27, 2023

At least that was the case until October 24, when The Verge put out an exclusive report that claimed the next-gen model, known as Orion, will be launched this December. After publishing the report, however, an OpenAI spokesperson told The Verge that the company won’t be releasing a model code-named Orion this year, but that it had plans to release “a lot of other great technology.”

Where does that leave us? Scratching our heads, for the most part.

We do know, however, that a next-gen model is being worked on. OpenAI announced publicly back in May that training on its next-gen frontier model “had just begun.” As to when it will launch, however, we’re still in the dark.

Availability

Currently all three commercially available versions of GPT — 3.5, 4 and 4o — are available in ChatGPT at the free tier. A ChatGPT Plus subscription garners users significantly increased rate limits when working with the newest GPT-4o model as well as access to additional tools like the Dall-E image generator. Because there’s been very little official talk about GPT-5 so far, you might assume GPT-5 would take the place of GPT-4 in ChatGPT Plus.

According to the report from The Verge, Orion won’t actually release as a part of ChatGPT. Instead, it would reportedly be limited to partnerships with specific companies — at least at first.

It should be noted that spinoff tools like Microsoft Copilot are being based on the latest models, with Copilot secretly launching with GPT-4 before that model was even announced. We could see a similar thing happen with GPT-5 when we eventually get there, but we’ll have to wait and see how things roll out.

Will GPT-5 achieve AGI?

Probably not. We’ve been expecting robots with human-level reasoning capabilities since the mid-1960s. And like flying cars and a cure for cancer, the promise of achieving AGI (Artificial General Intelligence) has perpetually been estimated by industry experts to be a few years to decades away from realization. Of course, that was before the advent of ChatGPT in 2022, which set off the generative AI revolution and has led to exponential growth and advancement of the technology over the past four years.

Achieving AGI is still OpenAI’s overarching goal, and when GPT-5 training was first announced, OpenAI said the following: “OpenAI has recently begun training its next frontier model and we anticipate the resulting systems to bring us to the next level of capabilities on our path to AGI.”

While achieving that AGI goal remains to be seen, GPT-5 is expected to vastly outperform OpenAI’s currently available models in terms of both complexity and efficiency, offering improvements in its natural language processing, content generation abilities, a larger knowledge base, and enhanced reasoning skills. Claude 3.5 Sonnet’s current lead in the benchmark performance race could soon evaporate.

Last year, Shane Legg, Google DeepMind’s co-founder and chief AGI scientist, told Time Magazine that he estimates there to be a 50% chance that AGI will be developed by 2028. Dario Amodei, co-founder and CEO of Anthropic, is even more bullish, claiming last August that “human-level” AI could arrive in the next two to three years. For his part, OpenAI CEO Sam Altman argues that AGI could be achieved within the next half-decade.

Interesting find: One year ago, forecasters estimated AGI to be ready by 2057.

Given the rapid pace of AI these past few weeks, AGI is now expected to be ready by October 2032. 🤯 pic.twitter.com/vHp6izeBAI

— Rowan Cheung (@rowancheung) March 28, 2023

During a recent interview with Dartmouth Engineering, OpenAI Chief Technology Officer Mira Murati, described the capability gap between GPT-4 and GPT-5. “If you look at the trajectory of improvement, systems like GPT-3 were maybe toddler-level intelligence,” Murati said. “And then systems like GPT-4 are more like smart high-schooler intelligence. And then, in the next couple of years, we’re looking at Ph.D. intelligence for specific tasks. Things are changing and improving pretty rapidly.”

The comments from Murati match closely with what CEO Sam Altman has stated publicly about what’s coming next. In June, Altman said that he expects GPT-5 to be a “significant leap forward,” but that they “still have a lot of work to do on it.”

Pushback to GPT-5

Concerns about a model significantly more powerful than GPT-4 have been raised from very early on. Shortly after the release of GPT-4, a petition signed by over a thousand public figures and tech leaders has been published, requesting a pause in development on anything beyond GPT-4. Significant people involved in the petition include Elon Musk, Steve Wozniak, Andrew Yang, and many more.

The eye of the petition is clearly targeted at GPT-5 as concerns over the technology continue to grow among governments and the public at large.

a few quick updates about safety at openai:

as we said last july, we’re committed to allocating at least 20% of the computing resources to safety efforts across the entire company.

our team has been working with the US AI Safety Institute on an agreement where we would provide…

— Sam Altman (@sama) August 1, 2024

In terms of its safety, Altman has posted on X (formerly Twitter) that OpenAI would be “working with the US AI Safety Institute,” and providing early access to the the next foundation model.