OpenAI is set to unveil its next-generation frontier model, tentatively named Orion and speculated to be GPT-5, by December, according to an exclusive report from The Verge. However, OpenAI’s CEO Sam Altman is already raising some doubts about these developments.
Insider sources suggest that, unlike previous releases of GPT-4, Orion will not be made available to the public right away. Instead, OpenAI plans to distribute the new model to a limited number of businesses and partners. These select organizations will then utilize it to develop their own applications and services, mirroring the approach Nvidia is taking with its NVLM 1.0 line of large language models.
Additionally, The Verge indicates that Microsoft intends to host Orion starting in November. While it remains unconfirmed whether the model will officially bear the name GPT-5 upon release, it is being regarded internally as the successor to GPT-4.
Shortly after the article’s release, Altman cryptically commented on social media, branding the rumored information as “fake news.” However, he refrained from explicitly denying the December launch or identifying inaccuracies in The Verge’s report.
This confusion stems from his brief response, which failed to clarify his position or provide details about OpenAI’s plans for Orion.
This news comes on the heels of OpenAI’s recent surprise launch of Project Strawberry, branded as 01 preview and 01-mini. This new model is designed to reason through complex queries across various domains—ranging from science and coding to mathematics—at a speed that surpasses human capabilities. Nevertheless, the reception of the 01 models hasn’t quite matched the excitement that hailed GPT-4, partly due to their restricted functionality, such as the inability to upload files or analyze images, and the high operational costs for OpenAI.
In the lead-up to the 01 preview’s release, Altman shared a sequence of ambiguous tweets featuring the fruit theme. It appears he is once again adopting a similar strategy with Orion. In September, while the company was reportedly finalizing the training of Orion using synthetic data generated by the 01 models, Altman tweeted about enjoying the Midwest skies. Interestingly, Orion is also the name of a prominent winter constellation.
Speculation around the advent of GPT-5 has been rampant for months, gaining traction since the unveiling of GPT-4 in March 2023. Initially, predictions placed its release in the summer of 2024, but expectations shifted to this fall as that timeframe passed. Former Chief Technology Officer Mira Murati had mentioned in June that the next-generation model would not arrive for another 18 months.
Thus, while a December launch aligns with the two-year milestone of ChatGPT’s debut, it’s equally plausible that the rollout could be pushed to 2025, especially given the history of erroneous forecasts.