OpenAI plans to release its next-generation frontier model, code-named Orion and rumored to actually be GPT-5, by December, according to an exclusive report from The Verge. However, OpenAI boss Sam Altman is already pushing back.
According to “sources familiar with the plan,” Orion will not initially be released to the general public, as the previous GPT-4 variants were. Instead, the company intends to hand the new model over to select businesses and partners, who will then use it as a platform to build their own products and services. This is the same strategy that Nvidia is pursuing with its NVLM 1.0 family of large language models (LLMs).
What’s more, The Verge reports that Microsoft is planning to host the new model beginning in November. There is no confirmation yet that Orion will actually be called GPT-5 when it is released, though the model is reportedly considered by its engineers to be GPT-4’s successor.
Within hours of the report’s publication on Friday, Altman coyly denied the claims, stating “fake new is out of control.” However, he did not deny outright that Orion would be released in December, nor did he cite any specific aspects of The Verge’s story as being factually inaccurate.
He has not expounded on his position since publishing that tweet, leaving us confused both about his statement’s meaning and his company’s eventual plans for Orion.
fake news out of control
— Sam Altman (@sama) October 25, 2024
This report comes a little over a month after OpenAI’s surprise release of Project Strawberry, officially known as 01 preview and 01-mini. Its “reasoning” architecture is designed to deduce answers as a human would and accurately solve complex questions on various subjects — including science, coding, and math — faster than a person can.
However, the 01 models have not been met with the same degree of fervor that GPT-4 received, in part because the new models are limited in their functionality, incapable of uploading files or analyzing images, and as VentureBeat notes, are expensive for OpenAI to operate.
In the run-up to 01 preview’s release, Altman published a series of cryptic tweets featuring the fruit. He appears to be doing the same with Orion. In September, just as the company was reportedly finishing up training Orion using synthetic data generated by 01, Altman fired off a conspicuous tweet about visiting the Midwest. As The Verge notes, the dominant constellation in the Northern Hemisphere’s winter sky is, you guessed it, Orion.
i love being home in the midwest.
the night sky is so beautiful.
excited for the winter constellations to rise soon; they are so great.
— Sam Altman (@sama) September 14, 2024
Rumors about GPT-5 have flooded the internet for many months now, basically since the day GPT-4 launched in March 2023. Reports initially put the release date sometime in the summer of 2024, but once that passed, the goalpost had been moved to this fall. On the other hand, former Chief Technology Officer Mira Murati said in an interview this past June (before leaving teh company) that the “next-gen” model was not due out for another year and a half.
So, while a launch later this December seems plausible, timed with the two-year anniversary of ChatGPT, it’s just as likely that it won’t come until 2025 based on how inaccurate all the predictions have been so far.