Since the release of GPT-4 last March, there has been much speculation about the release and potential capabilities of GPT-5, with little comment from OpenAI.

For a company with "open" in its name, OpenAI is as tight-lipped as Apple when it comes to new products.

However, according to Business Insider, a flagship model will be announced as early as this summer, appearing on ChatGPT and will be "very different" from the GPT-4.

The basis for the summer release rumor appears to be third-party companies that have gained early access to OpenAI's new model. These OpenAI enterprise customers are part of OpenAI's bread and butter, bringing in significant revenue to cover the growing costs of operating a larger model.

The rumor of a summer release goes against what OpenAI CEO Sam Altman suggested in an interview with Rex Friedman. He stated that there will be a new model this year, but it will not necessarily be the GPT-5.

"We will release a great model this year, but I don't know what we will call it. I think there are a lot of other important things that need to be released first before we can talk about a model like the GPT-5."

This could include Sora, a video AI model that OpenAI CTO Mira Murati says will be released by the end of the year. [If GPT-5 is not significantly larger than its predecessor, it makes sense that it would take several months.

BusinessInsider's report suggests that they have completed their training and have moved on to the "red team." However, this assumes that we are talking about GPT-5.

Red Team is when a model is placed in extreme conditions and tested for safety issues. The next phase of the red team is to fine-tune the model, correct any problems noted during testing, add guardrails, and make it available to the public.

This phase alone could take several months, as it did for GPT-4. Therefore, GPT-5, which is supposed to be released this summer, may actually be GPT-4.5, not GPT-4.5.

This would relate to the previous generation of releases, which moved from GPT-3 to GPT-3.5 before GPT-4 was introduced. In addition, GPT-3.5-Turbo and GPT-4-Turbo are now available for free on Microsoft Copilot.

Altman has previously stated that GPT-5 will be a significant improvement over any previous generation model. This includes video capabilities (the ability to understand the content of a video), which will greatly improve reasoning.

Only exaggerated comments in interviews have revealed few specific details; we know that it is likely to be larger and have more parameters than the GPT-4, but it is not clear to what extent.

We just got details on the size of its predecessor; Nvidia CEO Jensen Huang revealed during GDC that GPT-4 has 1.8 trillion parameters and requires 30 yottaflops of computing power for training.

However, it is unclear if Nvidia's new exaFLOP GPU platform was used to train GPT-5. A relatively small cluster of Blackwell chips in a data center could train a trillion parameter model in days, not weeks or months.

Altman said it will be much more intelligent than previous models. 'We're excited about being smarter,' he said. 'Overall, it's better.'

There was early speculation that GPT-5 would be the first sign of superhuman intelligence, also known as artificial intelligence (AGI), but it is increasingly believed that it will be a leap forward from GPT-4, but not by much.