GPT-5: OpenAI May Launch “Better” Model for ChatGPT This Summer

Most of us are pretty impressed by ChatGPT. Sure, it’s got its shortcomings, but it definitely doesn’t suck, right?

Not according to OpenAI CEO Sam Altman, who has publicly criticism his company’s current large language model, GPT-4, helping fuel new rumors suggesting the AI powerhouse could be preparing to release GPT-5 as soon as this summer.

That’s because, just days after Altman admitted that GPT-4 still “kinda sucks,” an anonymous CEO claiming to have inside knowledge of OpenAI’s roadmap said that GPT-5 would launch in only a few months time.

All of which has sent the internet into a frenzy anticipating what the “materially better” new model will mean for ChatGPT, which is already one of the best AI chatbots and now is poised to get even smarter.

Here’s all the latest GPT-5 news, updates, and a full preview of what to expect from the next big ChatGPT upgrade this year.

GPT-5 Latest News and Updates for March 2024

All eyes are on OpenAI this March after a new report from Business Insider teased the prospect of GPT-5 being unveiled as soon as summer 2024.

The publication says it has been tipped off by an unnamed CEO, one who has apparently seen the new OpenAI model in action. The mystery source says that GPT-5 is “really good, like materially better” and raises the prospect of ChatGPT being turbocharged in the near future.

However, one important caveat is that what becomes available to OpenAI’s enterprise customers and what’s rolled out to ChatGPT may be two different things.

For its part, OpenAI hasn’t confirmed anything beyond saying in the past that GPT-5 (and even GPT-6) were under devleopment, and that the company had something “exciting” in the making this year.

In an earlier March 2024 interview with the Lex Freidman Podcast, Altman said the “honest answer” is he doesn’t know when GPT-5 will be ready, saying that his company has “a lot of other important things to release” before its next major LLM update.

“A lot” could well refer to OpenAI’s wildly impressive AI video generator Sora and even a potential incremental GPT-4.5 release.

This is also the now infamous interview where Altman said that GPT-4 “kinda sucks,” though equally he says it provides the “glimmer of something amazing” while discussing the “exponential curve” of GPT’s development.

In other words, everything to do with GPT-5 and the next major ChatGPT update is now a major talking point in the tech world, so here’s everything else we know about it and what to expect.

What is GPT-5?

GPT is shorthand AI jargon for “Generative pre-trained transformer.” It’s a large language model, or LLM, developed by AI powerhouse OpenAI that serves as the framework for company’s chatbot, ChatGPT – one of the best AI chatbots around.

Another way to think of it is that a GPT model is the brains of ChatGPT, or its engine if you prefer. GPT-5 will be the fifth full release of such a model by OpenAI.

The first was a proof of concept revealed in a research paper back in 2018, and the most recent, GPT-4, came into public view in 2023.

Right now, it looks like GPT-5 could be released in the near future, or still be a ways off. All we know for sure is that the new model has been confirmed and its training is underway.

How long this takes is an unanswerable question, as OpenAI could take as long as they want refining its basic capabilities – or try to rush out the new version as soon as possible to stay in pole position in the AI arms race.

GPT-5: What to Expect and What We Want to See

The first thing to expect from GPT-5 is that it might be preceded by another, more incremental update to the OpenAI model in the form of GPT-4.5.

This might find its way into ChatGPT sooner rather than later, while GPT-5 stays under development and slowly rolls out behind closed doors to OpenAI’s enterprise customers. Let’s take a look at that gossip and everything else to expect from GPT-5.

GPT-4.5 Leak Tips June 2024 Release Window

As demonstrated by the incremental release of GPT-3.5, which paved the way for ChatGPT-4 itself, OpenAI looks like it’s adopting an incremental update strategy that will see GPT-4.5 released before GPT-5.

This means before we get to what we might see in GPT-5, we need to pause to consider GPT-4.5.

In fact, a claimed GPT-4.5 release window may already have been leaked by OpenAI, if internet tech sleuths are to be believed. Apparently, both Microsoft’s Bing and the DuckDuckGo search engine indexed an OpenAI blog post referencing the availability of a GPT-4.5 model with a June 2024 “knowledge cut-off date.”

It follows that GPT-4.5 itself could be released around summer ’24, as OpenAI tries to keep up with newly release rivals like Anthropic’s Claude 3, and ultimately paving the way for GPT-5 to launch in late-2024 or some point in 2025.

Adding even more weight to the rumor that GPT-4.5’s release could be imminent is the fact that you can now use GPT-4 Turbo free in Copilot, whereas previously Copilot was only one of the best ways to get GPT-4 for free.

Throw in the March 2024 Microsoft Surface event and you’ve even got a catwalk for GPT-4.5 to be initially teased, given Microsoft is one of OpenAI’s biggest partners, investors, and even sits on the company’s board.

It’s on the duck duck go one pic.twitter.com/7Jw45v8ybr

— Matthew (@Matthew35342706) March 12, 2024

GPT-5 Confirmed to be Under Development

As well as the likelihood that the release of GPT-4.5 by OpenAI is nearing, we also now have it confirmed that the company is indeed working on GPT-5, so its next-gen LLM is actually under development as opposed to just being an internet pipedream.

OpenAI CEO Sam Altman revealed as much at the start of 2024, speaking to Bill Gates on the tech icon’s Unconfuse Me podcast.

Expect a Major Leap in GPT-5 Parameters vs GPT-4

Now, as we approach more speculative territory and GPT-5 rumors, another thing we know more or less for certain is that GPT-5 will offer significantly enhanced machine learning specs compared to GPT-4.

The headline one is likely to be its parameters, where a massive leap is expected as GPT-5’s abilities vastly exceed anything previous models were capable of. We don’t know exactly what this will be, but by way of an idea, the jump from GPT-3’s 175 billion parameters to GPT-4’s reported 1.5 trillion is an 8-9x increase.

Whether or not OpenAI can match that remains to be seen, but a law of diminishing returns could be emerging based on the fact that the 115x jump from GPT-2 to GPT-3 is almost certainly never going to be repeated.

If OpenAI can approach a 4-5x parameter increase with GPT-5, it would be a phenomenal accomplishment, especially when you consider that the latest industry benchmark, Anthropic’s Claude 3 Opus, is a 2 trillion parameter model, so even doubling the parameters of GPT-4 would be an impressive feat.

Because we’re talking in the trillions here, the impact of any increase will be eye-catching. It’s also safe to expect GPT-5 to have a larger context window and more current knowledge cut-off date, with an outside chance it might even be able to process certain information (such as social media sources) in real-time.

Sora and Multimodality at the Fore of GPT-5

As excited as people are for the seemingly imminent launch of GPT-4.5, there’s even more interest in OpenAI’s recently announced text-to-video generator, dubbed Sora.

Sora is the latest salvo in OpenAI’s quest to build true multimodality into its products right now, ChatGPT Plus (the chatbot’s paid tier, costing $20 a month) offers integration with OpenAI’s DALL-E AI image generator. It lets you make “original” AI images simply by inputting a text prompt into ChatGPT.

With Sora, you’ll be able to do the same, only you’ll get a video output instead. The early displays of Sora’s powers have sent the internet into a frenzy, and even after more than 10 years of seeing tech’s “next big thing” come and go, I have to say it’s wildly impressive.

There’s every chance Sora could make its way into public beta or ChatGPT Plus availability before GPT-5 is even released, but even if that’s the case, it’ll be bigger and better than ever when OpenAI’s next-gen LLM does finally land.

GPT-5 Could Usher in Project Q* Era of AGI

Of course, in any discussion of GPT-5 it’s impossible to ignore the fact that, behind the scenes, OpenAI is working to develop not only its current suite of AI products, but also to bring forth a new breed of Artificial General Intelligence (AGI) technologies currently dubbed Project Q*.

AGI is the term given when AI becomes “superintelligent,” or gains the capacity to learn, reason and make decisions with human levels of cognition. It basically means that AGI systems are able to operate completely independent of learned information, thereby moving a step closer to being sentient beings.

When current AI technologies can’t seem to stop themselves hallucinating information or generating the odd racist image for japes, it’s a hugely controversial prospect to say the least and is being opposed by many within the industry on ethical grounds.

Nevertheless, OpenAI is one of its chief proponents and with the next major GPT release, GPT-5, there’s a chance we’ll learn more about its ultimate aims for Project Q* AGI superintelligence, if not see a finished product, as per Altman’s own statements.

GPT-6 Also “Confirmed” by OpenAI

Why just get ahead of ourselves when we can get completely ahead of ourselves? In another statement, this time dated back to a Y Combinator event last September, OpenAI CEO Sam Altman referenced the development not only of GPT-5 but also its successor, GPT-6.

He stated that both were still a ways off in terms of release; both were targeting greater reliability at a lower cost; and as we just hinted above, both would fall short of being classified as AGI products.

Building in AI? Here are gems from a comprehensive talk by @sama at @OpenAI, from the @ycombinator
alumni reunion weekend:

Most important lessons from early days of building:

🔸Thinking long term will take you far. Open AI experienced [many] pivots. From robotic arms to… pic.twitter.com/ikDJ2CcS4F

— Iba Masood (@IbaMasood) September 24, 2023

GPT-5 Potential Release

Based on what we’ve heard and what we’ve seen in the past, there are a number of possibilities.

Hot of the presses right now, as we’ve said, is the possibility that GPT-5 could launch as soon as summer 2024.

However, with a claimed GPT-4.5 leak also suggest a summer 2024 launch, it might be that GPT-5 proper is revealed at a later days.

Other possibilities that seem reasonable, based on OpenAI’s past reveals, could seeGPT-5 released in November 2024 at the next OpenAI DevDay.

That’s when we first got introduced to GPT-4 Turbo – the newest, most powerful version of GPT-4 – and if GPT-4.5 is indeed unveiled this summer then DevDay 2024 could give us our first look at GPT-5.

That, or GPT-4.5 Turbo could make its way into the public eye at DevDay 2024, both paving the way and somewhat delaying GPT-5;s launch into 2025.

GPT Model Release History and Timeline

To get an idea of when GPT-5 might be launched, it’s helpful to look at when past GPT models have been released. Here’s an overview of OpenAI’s GPT model history.

GPT-1

The original GPT was unveiled back in June 2018, but only as a proof of concept forming a key part of OpenAI’s research paper “Improving Language Understanding by Generative Pre-Training.”

The 117 million parameter model wasn’t released to the public and it would still be a good few years before OpenAI had a model they were happy to include in a consumer-facing product.

GPT-2

The second foundational GPT release was first revealed in February 2019, before being fully released in November of that year. Capable of basic text generation, summarization, translation and reasoning, it was hailed as a breakthrough in its field.

The 1.5 billion parameter model was pre-trained on a dataset of eight million webpages, becoming the first publicly available GPT release, though its use was largely of the experimental variety and confined to the machine learning community.

GPT 3

GPT-3 represented another major step forward for OpenAI and was released in June 2020. The 175 billion parameter model was now capable of producing text that many reviewers found to be indistinguishable for that written by humans.

While still a couple of years away from the release of ChatGPT, OpenAI’s latest effort had everyone in the tech world talking, while The Guardian even published an entire article written by GPT-3 to demonstrate its hitherto inconceivable abilities.

GPT-3.5

Released in March 2022, GPT-3.5 was the bedrock on which ChatGPT was built. It was based on the same 175 billion parameter transformer as GPT-3, but was fine-tuned and trained on data up to June 2021.

The announcement of GPT-3.5 was closely followed in November 2022 by the launch of ChatGPT. The rest, as they say, is history.

GPT-4

As anyone who used ChatGPT in its early incarnations will tell you, the world’s now-favorite AI chatbot was as obviously flawed as it was wildly impressive. Unveiled in March 2023, GPT-4 was OpenAI’s attempt to fix that.

By Altman’s own admission, it’s still a work in progress, but GPT-4 has been a mind-blowing step forward for AI technology that’s now good enough to be widely used not only by lazy college students, but by major businesses as part of their day-to-day operations.

While the actual number of GPT-4 parameters remain unconfirmed by OpenAI, it’s generally understood to be in the region of 1.5 trillion.