- Added OpenAI statement
Update October 26, 2024:
An OpenAI spokesperson confirmed to The Verge that a model code-named "Orion" will not be released this year, though the company plans to release "a lot of other great technology."
According to The Verge, Orion is being discussed internally as a successor to GPT-4. However, the company hasn't yet decided whether the model will be released as GPT-5 when it eventually launches.
Original article from October 25, 2024:
OpenAI CEO Sam Altman says news of Orion AI model launch in December is "random fantasy"
Sam Altman describes a media report about the alleged imminent release of a new AI model called Orion as "pure fantasy".
OpenAI CEO Sam Altman strongly denied a recent media report about a new AI model called Orion that was supposedly set to launch in December. "Fake news out of control" and "just offends me how media is willing to print random fantasy," Altman wrote on X, responding to an article by The Verge.
The Verge's Kylie Robison had claimed, citing unnamed sources, that OpenAI planned to release a new AI model named Orion, with Microsoft engineers already working on integrating it by November. Microsoft, OpenAI's largest investor, recently announced plans to launch agent-based AI in November, which may have contributed to the speculation.
Altman doesn't usually comment so harshly on media reports and rumors. While he dismissed The Verge's report, he added that there's "plenty of great stuff coming your way."
Altman's cryptic hints fuel speculation
Altman recently sparked discussion with a cryptic post on X mentioning he was "excited for the winter constellations to rise soon" - potentially referencing the Orion constellation. He also suggested ChatGPT might receive an update for its second birthday in November.
OpenAI and Microsoft executives have previously signaled plans for a much improved AI model this year. Altman specifically noted expectations for a "significant leap forward" with GPT-5, while CTO Mira Murati indicated PhD-level capabilities within 18 months.
Rumors about an Orion model have been circulating for some time. According to reports, it will significantly outperform GPT-4, in part by training on high-quality synthetic data generated with OpenAI's "reasoning model" o1. OpenAI reportedly hopes that this combination will reduce the error rate compared to previous models.
