After twelve consecutive days of important announcements, OpenAI has just unveiled its latest reasoning model, o3, successor to the o1 model launched earlier this year. This announcement comes at a particular time, just after Google presented its own reasoning model, Gemini 2.0 Flash Thinking. For Apple users, who already benefit from AI capabilities through Siri and various integrations in iOS, these advances could foreshadow future improvements in Apple Intelligence.
A new, more efficient and more thoughtful model
OpenAI's o3 model is notable for its ability to “think” before responding, using what OpenAI calls a “private thought chain.” Concretely, the model takes more time – from a few seconds to several minutes – to analyze a question and formulate an answer, but in return offers increased reliability, particularly in the fields of science, mathematics and physics. In particular, it is three times more efficient than its predecessor o1 on the ARC-AGI benchmark, which evaluates the ability of an AI model to reason about new problems.
The model comes in two versions: o3 and o3-mini, the latter being a lightweight version optimized for specific tasks, while being cheaper. Curious fact: the absence of “o2” in the nomenclature would be due to a potential brand conflict with the British telecom operator O2.
A race for innovation between technological giants
This announcement comes during a particularly intense period in terms of AI innovations. Google recently unveiled Gemini 2.0, while OpenAI has made numerous announcements at the end of the year: a new video generation model, a free version of its search engine powered by ChatGPT, and even a telephone service accessible via 1 -800-ChatGPT.
Sam Altman, CEO of OpenAI, is cautious about the general deployment of o3, preferring to wait until a federal testing framework is in place to monitor and mitigate potential risks. This precaution is not trivial: security tests revealed that o1 had a more pronounced tendency to try to deceive users than conventional models.
This development also comes as Alec Radford, one of OpenAI's most accomplished scientists and lead author of the GPT series papers, is leaving the company to pursue independent research. For Apple, which is actively working on its own AI solutions, these advances could influence the development of future features for its devices, particularly with a view to further integration of generative AI in iOS and macOS.
Download the free app ChatGPT
