After announcing many features and services in recent days, OpenAI officially previewed its upcoming new artificial intelligence model o3 in its final announcement.
According to OpenAI CEO Sam Altman, the reason for naming it o3 is that OpenAI is not good at naming, and it also wants to avoid a name conflict with the O2 Internet service launched by Spanish telecommunications operator Telefónica in Europe. Therefore, the new artificial intelligence model will not be named o2.
Compared to the previously released o1, the previewed o3 boasts higher performance. OpenAI emphasized that, taking the American Invitational Mathematics Examination test as an example, o3's answer accuracy rate reached 96.7%, while o1's answer accuracy rate was only 83.3%.
However, OpenAI stated that o3 is currently only a preview version and is only open to researchers assisting in security testing. It is expected to be launched in 2025, but the specific launch time has not yet been announced.
In addition to the o3 artificial intelligence model, OpenAI also announced a smaller artificial intelligence model called o3-mini, which incorporates an adaptive thinking time feature that allows users to adjust the o3 artificial intelligence model's reasoning time, allowing the artificial intelligence model to reason for different lengths of time before giving an answer.
OpenAI said that the execution performance of o3-mini is equal to that of o1, but the execution cost can be greatly reduced. It is expected to be launched at the end of January 2025, while o1 is expected to enter the market after the launch of o3-mini.
Today, we shared evals for an early version of the next model in our o-model reasoning series: OpenAI o3 pic.twitter.com/e4dQWdLbAD
- OpenAI (@OpenAI) December 20, 2024
