4 things to know about o3, the new AI model presented by OpenAI

© Unsplash/Levart_Photographer

OpenAi had a lot to talk about as 2024 drew to a close. For several days, the world's most famous AI company unveiled its work to improve its ChatGPT artificial intelligence. Among them, the new o3 intelligence model promises to push the boundaries of everything we knew before. Here are 4 things to know about o3.

Gone are the days of instant and sometimes approximate answers. o3 is distinguished by its ability to “think” before responding. Unlike traditional models that work like parrots, o3 uses a “private thought chain” to analyze problems from all angles.

More interestingly, users can now choose between three levels of thinking: low, medium, or high. The higher the level of thinking, the longer the model can take to formulate an answer. In return, o3 takes the time to fact-check and avoid the pitfalls that usually trip up other AI models.

The numbers are staggering: o3 smashes records on virtually every standardized test.He scored a mind-blowing 96.7% on the American AIME mathematics exam and achieved an 87.7% score on PhD-level questions in science. In programming, he ranked among the top 1% of developers worldwide with a Codeforces rating of 2727, a level that only the best programmers in the world achieve.

200% Deposit Bonus up to €3,000 180% First Deposit Bonus up to $20,000

But most impressive is his performance on the ARC-AGI benchmark, where he achieved 87.5% in high-performance mode. Designed to assess an AI system's ability to learn new skills outside of its training data, this test puts o3 at a level never before achieved. However, this exceptional performance comes at a significant cost: each query in high-performance mode can cost thousands of dollars in computing power.

OpenAI also unveiled o3-mini, a lightweight version of the main model, which offers remarkable flexibility for everyday tasks. Although more modest than its big brother, it already surpasses the o1 model in many aspects. Its particularity ? It allows to adjust the thinking effort according to the complexity of the task, thus offering an optimal balance between speed and precision.

The release of o3-mini is planned for the end of January 2025, followed shortly thereafter by that of o3. This phased rollout strategy reflects OpenAI's caution, as it seeks to ensure that these new models adhere to its security principles through a new technique called “deliberative alignment.”

Because while these models are more powerful, they also raise important questions about their potential to mislead users, something the company takes very seriously.

• o3 introduces a three-speed thinking system, allowing the “thinking” time to be adapted as needed
• The model's performance exceeds expectations, even surpassing human capabilities in some areas
• A more accessible mini version offers an excellent compromise between power and efficiency, available from January 2025

📍 To not miss any Presse-citron news, follow us on Google News and WhatsApp.