Sam Altman denies reports of a new OpenAI model in December
The Verge recently reported that OpenAI was preparing to launch a new AI model codenamed "Orion" in December. The report claimed that OpenAI would introduce the model first to its close partners and that Microsoft would host Orion on its cloud computing platform Azure in November.
According to the report, OpenAI views Orion as a successor to GPT-4, but it's unclear if it will be called GPT-5 on release. It's worth noting that both OpenAI and Microsoft declined to comment on the original report. Little is known about the new model at this point. In September, OpenAI's Shaun Ralston posted a graph on X showing the generational improvements the models have made since GPT-3.
Ralston referred to a "GPT-Next" model that will be released this year. Ralston posted that it had been trained on a "compact Strawberry (OpenAI o1) version" and had 100 times more "computational volume" than GPT-4. Interestingly, Orion is also mentioned in this post, albeit as a separate model trained on "10K (Nvidia) H100 GPUs".
Source(s)
