OpenAI just wrapped up its 12-day event called “Shipmas” where it made some amazing announcements. As a proper send-off, OpenAI introduced us to o3, its upcoming reasoning model, and it looks like it will be extremely smart.
During Shipmas, OpenAI announced some other great AI goodies. For starters, it introduced its $200/month ChatGPT Pro plan. This will give users access to the most powerful version of o1 and other great features. Also, the company released Sora, its AI video generator that pretty much broke the internet when the company first showed it off. You can use it if you’re a ChatGPT Plus member.
OpenAI gives us a sneak peek at o3, its latest reasoning model
What happened to o2? Well, it’s in the farm up-state along with Windows 9, the OnePlus 4, and the iPhone 9. OpenAI decided to skip to o3 because there’s a British telecommunication company named O2. So, this was a way to avoid any legal issues down the road.
o3 will be a reasoning model, which is similar to a regular model. However, the key difference is that, instead of giving you the answer all at once, a reasoning model will actually break down the process and show you all of the steps it took to come to the conclusion. Google’s Gemini 2.0 Flash Thinking is a good example of a reasoning model. So, if you want to take a closer look into how a model arrived at its answer, then you’ll want to use reasoning models.
Since this will be OpenAI’s magnum opus, you know that it will come with some insane AI smarts. The company released some statistics on how it performs, and it shows that it’s well past the point of making AI that’s smarter than a human (well, mostly).
For example, the company put the model through the SWE-Bench Verified coding tests, and it beat o1 by 22.8%. Next, OpenAI put o3 through the GPQA (Google-Proof Q&A Benchmark) Diamond science benchmark, and it scored 87.7%. OpenAI also put o3 through the AIME (American Invitiational Mathematics Examination), and it only missed one of the 15 questions. The AIME is an extremely hard math competiton.
It looks like OpenAI really outdid itself this time around. We don’t know when the company will release this model to the public. Just don’t count on it anytime soon, as o1 is still rather new.
