OpenAI plans to release new AI models by year-end

OpenAI is reportedly working to release one or two new AI models by the end of the year. While it is unclear if these will include the company’s highly anticipated flagship upgrade, the development process involves rigorous alignment to ensure model goals match researcher and customer expectations.

New models are expected to utilize Looped Transformer technology, similar to the technique used in OpenAI’s Astra. This method, also referred to as ‘Recurrent Depth,’ processes internal token representations through the same transformer layers multiple times. This approach significantly improves performance in areas such as mathematics and coding while reducing costs by allowing smaller models to perform at higher levels.

Before a wide release, OpenAI must complete safety evaluations. Under an executive order, the company is required to provide advanced AI systems to the US government up to 30 days before sharing them with trusted partners for testing.