OpenAI is set to introduce its next-generation AI model, o3, early next year, with safety researchers gaining access to the technology.
The announcement, made during OpenAI’s “12 Days of OpenAI” livestream, also unveiled the existence of a smaller version, o3-mini, which will be available to the public by late January, with the full o3 release following shortly after.
Sam Altman, OpenAI’s CEO, quipped about the model’s name, referencing Telefónica’s O2 cellular network in Europe and OpenAI’s history of unconventional naming. While the public waits, safety researchers can begin exploring o3’s capabilities and offering feedback on potential risks. The model’s preliminary testing results reveal a significant leap in performance, making it a major advancement over its predecessors.
During benchmarking, o3 demonstrated exceptional performance, achieving a 96.7% accuracy on the American Invitational Mathematics Examination, compared to 83.3% by o1. Mark Chen, OpenAI’s senior vice president of research, highlighted that o3 typically missed only one question per test, prompting OpenAI to challenge the model further.
One of the benchmarks included ARC-AGI, a complex test designed to evaluate an AI’s capacity to learn intuitively and adapt in real-time. While no AI system has fully mastered ARC-AGI since its creation in 2019, o3 achieved 75.7% on a low-compute setting and 87.5% with additional resources, surpassing the human baseline of 85%. Greg Kamradt, president of the ARC Prize Foundation, noted the milestone as a significant stride toward artificial general intelligence.
The o3-mini variant offers a more efficient alternative, leveraging OpenAI’s Adaptive Thinking Time API to provide three reasoning modes—Low, Medium, and High. This flexibility allows users to balance response time with processing demands. Despite its smaller scale, o3-mini delivers results comparable to the o1 model while requiring less computational power, signaling accessibility improvements for everyday users.
With o3’s promising advancements, OpenAI continues to push the boundaries of AI technology, emphasizing both performance and safety as the new frontier approaches.
Featured Image courtesy of Anadolu Agency via Getty Images
Follow us for more tech news updates.
Source: https://digitalmarketreports.com/news/31376/openai-announces-o3-model-launching-next-year/