OpenAI revealed the benchmark results for its new o3 series of artificial intelligence (AI) models on Friday. Positioned as the successor to the recently introduced o1 series, the company asserts that o3 offers significantly enhanced capabilities compared to its predecessor. This new series comprises two models: the o3 and o3-mini, with a focus on advanced reasoning tasks. Currently, these AI models are available for public safety testing, and the o3-mini is scheduled for public release early next year. OpenAI’s announcement followed closely behind Google’s launch of its Gemini 2.0 Flash Thinking Mode AI model.
OpenAI Launches Public Safety Testing for o3 Series AI Models
On the twelfth day of a planned 12-day shipping schedule, OpenAI announced the introduction of the o3 series, succeeding the o1 AI model. During a live stream, CEO Sam Altman noted that the interim name o2 was omitted due to a trademark conflict with the UK telecommunications service provider Telefonica, which holds the same brand name.
While OpenAI made the announcement, it clarified that the models are not yet available to the general public. The company has initiated early access for select external researchers to conduct public safety testing on both the o3 and o3-mini models. Interested participants can apply here, with the application window closing on January 10.
According to OpenAI, both models in the o3 series are designed to outperform their predecessors, offering capabilities to tackle more complex tasks in coding, mathematics, and natural language processing. However, a complete overview of their functionalities has yet to be disclosed.
In addition, OpenAI presented some benchmark evaluations from internal tests of the o3 models. The company reported that the o3 achieved a score of 71.7 percent in the SWE-bench benchmark and 96.7 percent in the AIME 2024 benchmark, outpacing the o1 model. A comprehensive evaluation of the benchmarks will be attainable only once the models are made publicly accessible. The o3-mini is anticipated to launch in January 2025.
OpenAI Grants Unlimited Access to Sora for Paid Subscribers
In a separate development, Altman took to X (formerly Twitter) to announce a “day-13 bonus,” where ChatGPT Plus and Teams subscribers will enjoy unlimited access to Sora during the last two weeks of December. Altman indicated that this initiative is feasible due to reduced server loads as many offices close for the Christmas holiday.
Rohan Sahai, a product lead at OpenAI, further elaborated on the post, stating that the company has enhanced Sora’s blend feature and introduced shared links, enabling users to share AI-generated videos with friends, regardless of whether they have an account.
