OpenAI released the benchmark results for the o3 series of artificial intelligence (AI) models on Friday. As the successor to the recently launched o1 series of AI models, the company asserted that o3 greatly exceeds the abilities of the previous version. This series consists of two models, the o3 and the o3-mini, and is designed for advanced reasoning-oriented tasks. At present, the AI models are being rolled out for public safety evaluation, and the o3-mini variant will be introduced to the public early next year. Remarkably, OpenAI’s announcement followed closely on the heels of Google’s unveiling of its Gemini 2. 0 Flash Thinking Mode AI model.
OpenAI Shares Benchmark Scores of the o3 Series AI Models
On the twelfth day of the AI company’s 12-day shipping timeline, OpenAI introduced the o3 series, which follows the o1 AI model. During a live broadcast, CEO Sam Altman pointed out that the o2 designation was omitted due to the UK telecom provider Telefonica utilizing the same brand name.
OpenAI pointed out that, despite the announcement, the models will not presently be available in the public realm. The company has begun offering both o3 and o3-mini in early access to a select group of external researchers for public safety testing. Those wishing to take part in this process can apply here. The application window closes on January 10.
The company emphasized that both the o3 series AI models will outperform their predecessors and will have the capability to execute more intricate tasks in coding, mathematics, and natural language processing. Nevertheless, the complete scope of their features has not been disclosed.
Furthermore, OpenAI provided some benchmark assessments of the o3 AI models from internal testing. The company reported that o3 achieved a score of 71. 7 percent on the SWE-bench benchmark and 96. 7 percent on the AIME 2024 benchmark, surpassing o1. However, a comprehensive benchmark evaluation will only be feasible once the models are released to the public. The launch of the o3-mini is anticipated in January 2025.
OpenAI Offers Unlimited Access of Sora to Paid Subscribers
In a separate announcement, as a “day-13 bonus,” Altman mentioned in a post on X (previously known as Twitter) that ChatGPT Plus and Teams subscribers will have unlimited access to Sora during the final two weeks of December. The CEO noted that this opportunity arose as offices close during the Christmas season, leading to a decrease in server demand.
Rohan Sahai, a product lead at OpenAI, added to the post, stating that the company had enhanced Sora’s blend feature and enabled shared links, allowing users to share AI-generated videos with friends, even if those friends do not possess an account.
