Speculation has surrounded the release and potential capabilities of GPT-5 since the day GPT-4 was released in March last year. With very little comment from OpenAI on the topic.
For a company with “open” in its name, OpenAI is almost as tight lipped as Apple when it comes to new products — dropping them on X out of nowhere when they feel the time is right.
However, Business Insider reports that we could see the flagship model launch as soon as this summer, coming to ChatGPT and that it will be “materially different” to GPT-4.
The basis for the summer release rumors seems to come from third-party companies given early access to the new OpenAI model. These enterprise customers of OpenAI are part of the company's bread and butter, bringing in significant revenue to cover growing costs of running ever larger models.
Will GPT-5 come out this summer?
The summer release rumors run counter to something OpenAI CEO Sam Altman suggested during his interview with Lex Fridman. He said that while there would be new models this year they would not necessarily be GPT-5.
"We will release an amazing model this year, I don't know what we will call it," he said. "I think before we talk about a GPT-5-like model we have a lot of other important things to release first."
This could include the video AI model Sora, which OpenAI CTO Mira Murati has said would come out before the end of this year.
I think before we talk about a GPT-5-like model we have a lot of other important things to release first.
Sam Altman, CEO OpenAI
Altman first revealed GPT-5 had started training in November of last year and if it isn't significantly larger than its predecessor it would make sense this could take months of work.
The report from Business Insider suggests they've moved beyond training and on to "red teaming", especially if they are offering demos to third-party companies. Although that does assume it is GPT-5 we're talking about.
Red teaming is where the model is put to extremes and tested for safety issues. The next stage after red teaming is fine-tuning the model, correcting issues flagged during testing and adding guardrails to make it ready for public release.
That stage alone could take months, it did with GPT-4 and so what is being suggested as a GPT-5 release this summer might actually be GPT-4.5 instead. After all there was a deleted blog post from OpenAI referring to GPT-4.5-Turbo leaked to Bing earlier this year.
That would tie in to the previous generation releases, going from GPT-3 to GPT-3.5 before GPT-4 came out. We also saw GPT-3.5-Turbo and then GPT-4-Turbo, which is now available for free in Microsoft Copilot.
What can we expect from GPT-5?
Here's my conversation with Sam Altman (@sama), his 2nd time on the podcast. We talk about the board saga, Elon lawsuit, Ilya, Sora, GPT-5, $7 trillion in compute, open source, and AGI. This was a truly fascinating conversation.It's here on X in full, and is up on YouTube,… pic.twitter.com/h2ANXZpcrzMarch 18, 2024
Altman has previously said that GPT-5 will be a big improvement over any previous generation model. This will include video functionality — as in the ability to understand the content of videos — and significantly improved reasoning.
Beyond hyped up comments in interviews few specific details have been revealed. We know it will likely be larger and have more parameters than GPT-4 but by how much it isn’t clear.
We’re only just getting details of the size of its predecessor. Nvidia CEO Jensen Huang revealed during GDC that GPT-4 had 1.8 trillion parameters and required 30 yottaflops of compute power to train — that is like having a billion PS5s running constantly for 93,000 years.
I am excited about it being smarter. It is getting better across the board.
Sam Altman, OpenAI CEO
However, what we don’t know is whether they utilized the new exaFLOP GPU platforms from Nvidia in training GPT-5. A relatively small cluster of the Blackwell chips in a data centre could train a trillion parameter model in days rather than weeks or months.
Altman has said it will be much more intelligent than previous models. "I am excited about it being smarter," said Altman in his interview with Fridman. "It is getting better across the board."
There was some speculation early on that GPT-5 would be the first indication of a superhuman intelligence, also known as artificial general intelligence (AGI) but it looks increasingly like while it will be a leap over GPT-4, it won't be that much of a leap.
