OpenAI has revealed o3, its most advanced thinking model yet. They also announced a cheaper and smaller version called o3-mini, which is expected to come out in early 2025. These models are a big change in how AI is being made. They have amazing abilities in solving problems, writing code, and learning new things. Let’s take a closer look at the o3 models and what they mean for the future of AI.
Table of contents
A New High Standard for Thinking AI
The o3 model is better than ever before on many hard tests. One of its biggest achievements is scoring 75.7% on the ARC Prize’s AGI test. When using more computer power, it even scored 87.5%. This test shows how close we are to making artificial general intelligence (AGI), which is why o3’s performance is a very important step.
In math, o3 did great by solving 96.7% of the problems from the 2024 American Mathematical Olympiad (AIME). It also got 25.2% on the very hard Frontier Math Benchmark—much better than older models, which could barely get past 2%. The system also got 20% better at doing computer tasks compared to the older o1 model. It even got a score of 2727 on Codeforces, a website for coding competitions, beating OpenAI’s top scientist.
Impressive Achievements
- Math and Coding
- 96.7% on AIME: Almost perfect on very hard math problems.
- Codeforces Score of 2727: Shows o3 is very good at competitive coding.
- General Thinking
- 87.7% on PhD-Level Science: Did better than human experts on the GPT Diamond Benchmark.
- Big Step on ARC Benchmark: Scored 75.7%, a big jump in being able to learn and use what it knows in new situations.
What Makes o3 Special?
Unlike older AI models that mostly remember patterns, o3 thinks things through. It creates its own mini-programs in real-time to solve new problems. François Chollet, who created the ARC test, said that o3’s way of working is like Google DeepMind’s AlphaZero. AlphaZero tries out different solutions to find the best one. This skill lets o3 handle tasks that need it to learn and change.
For example, to solve problems on the ARC test, o3 often needs to look at up to 33 million pieces of information for each problem. This heavy work shows how carefully it solves problems, but it also uses a lot of computer power. The low-efficiency version to use o3 costs about $20 for each task. The high-efficiency version uses 172 times more computer power, making it much more expensive and slower.
The Price of Getting Better
Even though o3 can do amazing things, it needs a lot of computer power, which costs money. At about $17 to $20 per task, it currently costs more than a human doing the same task, which is around $5. However, OpenAI believes that the cost will get much better as the technology improves.
o3-Mini: The Cheaper Option
To make their new technology available to more people, OpenAI will release o3-mini in January 2025. This smaller version still works very well, even on medium settings. It’s also faster and cheaper than the older o1 model. In a live demonstration, o3-mini showed how useful it is by creating its own Python code and making ways to check if data is good. The model can also use special tools called API functions to give organized information and make function calls, which makes it more useful.
Challenges on the Way to True AI
Even though o3 is a big step forward, it’s not true AGI yet. Experts says that we will only reach true AGI when humans can no longer think of tasks that are easy for people but hard for AI. The new ARC-AGI-2 test, coming in 2025, is made to challenge today’s AI even more. Early tests suggest that while humans can score 95%, o3 might score below 30%, showing that there’s still a big difference between current AI and real general intelligence.
Pushing Limits with New Ideas
The way o3 is built is a major change in AI. By mixing what it already knows with creating new programs, it goes beyond what older large language models (LLMs) can do. Its ability to put knowledge together in new ways when facing a problem helps it handle unfamiliar challenges well.
It is important to remember that o3 isn’t just making current methods bigger; it’s using completely new ways of doing things. This new thinking shows that new ideas are more important than just using more computer power to make AI better.
What’s Next?
Looking ahead, OpenAI plans to focus on making AI safe and making sure it does what we want it to do. The company has introduced something called Deliberative Alignment, which uses o3’s thinking abilities to set strong safety rules. Also, the ARC Prize Foundation wants to help others create their own versions of o3 to speed up research and new ideas.
In Conclusion
OpenAI’s o3 is a major breakthrough in how AI can think. It pushes the limits of what’s possible in learning new things, solving problems, and writing code. While there are still challenges, its performance sets a new standard for AI development. With the launch of o3-mini and ongoing improvements, OpenAI is helping us move closer to a future where AI systems can handle the hardest and newest problems alongside humans.
| Latest From Us
- Meet Codeflash: The First AI Tool to Verify Python Optimization Correctness
- Affordable Antivenom? AI Designed Proteins Offer Hope Against Snakebites in Developing Regions
- From $100k and 30 Hospitals to AI: How One Person Took on Diagnosing Disease With Open Source AI
- Pika’s “Pikadditions” Lets You Add Anything to Your Videos (and It’s Seriously Fun!)
- AI Chatbot Gives Suicide Instructions To User But This Company Refuses to Censor It
