Codename Orion: Will the successor to GPT-4 be released in December?

The next major language model version of OpenAI could be a lot different than usual. As the US tech publication The Verge reports, citing anonymous sources within the company, the successor to GPT-4 could be launched in December. However, OpenAI boss Sam Altman immediately called the report "fake news out of control" on the social network X and an "arbitrary fantasy" – without going into further detail.

According to information available to The Verge, the new Large Language Model (LLM) developed under the code name Orion will not initially be usable via ChatGPT. Instead, OpenAI plans to grant access to close partners first. These partners will be able to use the time advantage to adapt their products or expand them based on the new model. According to internal schedules, Microsoft expects to host Orion on the Azure cloud platform from November.

100 times more powerful than GPT-4?

Orion is said to be 100 times more powerful than GPT-4, but may not be given the name GPT-5. Training was completed in September. This year, OpenAI already released an optimized version of GPT-4 called GPT-4o in May and OpenAI o1 as a specific extension in September. With the release in December, OpenAI allegedly wanted to honor the two-year anniversary of the release of ChatGPT.

The report criticized by Altman further states that OpenAI used its AI model o1 to provide synthetic data for training Orion. In addition, Altman provided a reference to Orion in a cryptic X-post in which he explained that he was looking forward to winter constellations rising soon – after all, the constellation of the same name is best visible from November to February.

(mki)