The last day of OpenAI’s 12-day event ended with the announcement of the o3 reasoning model, which includes a smaller version called the o3-mini. This model is the successor to the o1 reasoning models launched earlier this year.
So, what exactly are the ‘o3’ models? In simple terms, the o3 models are AI systems designed to tackle more complex problems in fields like science, math, and programming. Unlike most AI, which tends to spit out answers based purely on patterns it’s learned, reasoning models like o3 actually “think” things through.
Imagine this, when you’re solving a tough problem, you probably take a moment to plan, consider different approaches, and then come to a conclusion. That’s pretty much what the o3 models are trained to do. They even take a bit longer to respond because they’re carefully reasoning through the task at hand.
What’s also new is that you can adjust how much “thinking time” these models get—low, medium, or high compute. The more compute you give them, the better they perform. However, it must be noted that they are still not perfect.
OpenAI skipped calling this model ‘o2’ because of a trademark issue with the British telecom company O2! CEO Sam Altman even joked about it during a livestream. So, they went with ‘o3.’
The big leap with o3 is how much better it is at reasoning compared to its predecessor, o1. Sam Altman suggests that “on many coding tasks, o3-mini will outperform o1 at a massive cost reduction!”
This makes it more reliable for challenging tasks like solving graduate-level science questions or cracking tricky math problems. In some tests, o3 has outperformed other AI models by huge margins, OpenAI suggested.
That said, reasoning takes time. The o3 models aren’t the fastest, but they make up for it with accuracy. If you’ve ever been frustrated by AI giving you half-baked or wrong answers, this improvement will be perfect for you.
There’s also the smaller sibling, o3-mini, which is designed for specific tasks. OpenAI plans to release o3-mini by the end of January 2024, with the full o3 model coming sometime after.
For now, the o3 models are still in the testing phase. OpenAI is letting safety researchers try them out, and they’ve opened applications for external testers until January 10. Once testing wraps up, the o3-mini version will go public, followed by the full o3 model.
