OpenAI has released two new AI reasoning models called o3 and o4-mini today (April 17, 2025), according to information from OpenAI. These new models can solve problems better and can now work with both text and images. They are available to ChatGPT Plus, Pro, and Team users through the ChatGPT platform and API’s.
You might wonder: what exactly is an AI reasoning model? Think of it as a smart computer program that solves problems step-by-step, similar to how you might solve a puzzle. These programs can understand both words and pictures, break big problems into smaller parts, and find solutions more easily.
The new models come with several exciting abilities:
- They can think with and manipulate images – zooming in, rotating pictures, and analyzing what they see
- They can use all ChatGPT tools together to solve complex problems
- They make 20% fewer major errors compared to older models
- o3 achieves 87.7% accuracy on expert-level science benchmarks.
- They perform better on difficult tests in subjects like math, science, and coding
- o4-mini is a faster, more cost-efficient alternative.
Introducing OpenAI o3 and o4-mini—our smartest and most capable models to date.
For the first time, our reasoning models can agentically use and combine every tool within ChatGPT, including web search, Python, image analysis, file interpretation, and image generation. pic.twitter.com/rDaqV0x0wE
— OpenAI (@OpenAI) April 16, 2025
How are these models different from older ones like GPT-4? Here’s what makes them special:
| Feature | o3 and o4-mini | Previous Models |
|---|---|---|
| Image handling | Can zoom, rotate, and manipulate images | Basic image understanding only |
| Text length | Can handle 200,000 words at once | Limited to 128,000 words |
| Accuracy | 20% fewer major errors | More prone to mistakes |
| Tool use | Can combine all tools automatically | More limited tool capabilities |
OpenAI has also made these models safer than previous versions. They’ve improved safety in several important ways:
- Created new safety training data about dangerous topics like creating harmful programs or biological risks
- Added a special safety monitor that caught 99% of potentially harmful conversations during testing
- Ensured both models stay below the “High” risk threshold in important safety categories
Along with these new models, OpenAI introduced Codex CLI, which is a new tool for programmers. Codex CLI is like a helpful coding assistant that works right in your computer’s command line. It can help write code, fix bugs, and solve programming problems. The best part? It’s completely free and open-source, meaning anyone can use and improve it.
If you’re a programmer who wants to use Codex CLI, here’s how you can get started:
- Install it using a tool called npm (a program installer)
- Set up your OpenAI API key
- Type questions or commands in regular language to get coding help
- You can even share screenshots of your code or simple drawings to explain your problems
- Find it at github.com/openai/codex
OpenAI is also launching a $1 million funding program to support projects that use Codex CLI and OpenAI models. They will give $25,000 in API credits (which let you use their AI services) to projects they approve. This could help developers create new and useful tools with these technologies.
While o3 is already available, OpenAI says that an even more powerful version called o3-pro will be released in a few weeks. These improvements show how AI is getting better at thinking and solving problems, bringing us closer to AI that can truly help with complex tasks in our daily lives.
