OpenAI’s rumoured Orion model sparks excitement and speculation

OpenAI’s latest AI model, code-named Orion, is reportedly set to debut by December, with limited access initially granted to a few corporate partners, according to sources. Unlike previous releases available broadly on ChatGPT, Orion will first be shared with select companies, including key partner Microsoft. Engineers at Microsoft are preparing to deploy Orion on Azure by November, suggesting early access could be imminent.

Although Orion is seen as the successor to GPT-4, OpenAI has yet to confirm if the model will officially carry the GPT-5 designation. Publicly, OpenAI has downplayed the reports, with CEO Sam Altman dismissing them as “fake news.” An OpenAI spokesperson later clarified that the company has “no plans to release a model code-named Orion this year,” but they confirmed a commitment to releasing new technology.

Sources indicate that Orion could be up to 100 times more powerful than GPT-4 and separate from OpenAI’s o1 reasoning model, launched in September. Orion’s development has likely involved synthetic data generated by o1, referred to internally as “Strawberry.” OpenAI celebrated completing Orion’s training last month, which coincides with a cryptic post by Altman hinting at the model’s arrival, mentioning his excitement for “winter constellations.”

Orion is expected to advance OpenAI’s goal of creating a model capable of artificial general intelligence (AGI), a significant leap from current large language models. The prospect of Orion has drawn speculation, both for its potential capabilities and its selective release strategy, signalling OpenAI’s commitment to carefully refining its technology for high-level applications.