ECNETNews has learned that OpenAI has unveiled its next-generation reasoning models, o3 and o3 mini, in a recent announcement.
During a livestream event, the Senior Vice President of Research highlighted the impressive performance of o3 on various benchmarks, such as achieving 96.7 percent in competition math and 87.7 percent in PhD-level science. Notably, o3 scored 76 percent on the ARC-AGI benchmark, which assesses the capability to acquire new and unique skills dynamically with each task.
This announcement culminates OpenAI’s “12 Days of OpenAI” campaign, where new products and features were revealed each day. Over the past two weeks, OpenAI introduced its AI video generator, vision with Advanced Voice Mode, and a range of enhancements aimed at streamlining ChatGPT for both professional and personal use.
The o3 mini model is crafted to provide a cost-effective solution, offering three distinct effort levels and adjusting its reasoning time based on the challenge’s complexity. CEO Sam Altman remarked on the model’s “incredible cost-to-performance gain.”
While o3 and o3 mini signify significant advancements in AI technology, they are not yet available for general public use. However, OpenAI is initiating early access for safety testing starting today. Applications to participate in the model testing program will be accepted until January 10.
