OpenAI has just released two new AI models, o3 and o4-mini, designed to think more deeply and solve complex tasks better than ever. These models can now combine all of ChatGPT’s tools: they can browse the internet, analyze files and data, understand images, and even generate new ones.

What makes them stand out? These models are trained to know when and how to use the right tools to answer questions accurately and efficiently, usually in under a minute. This makes them much better at handling big, layered questions and tasks without needing step-by-step instructions.

What’s New?

• o3 is the most advanced model OpenAI has released so far. It excels in areas like coding, math, science, and analyzing visuals (charts, diagrams, photos). It’s great for anyone needing detailed, multi-step reasoning, whether you're writing code, solving a math problem, or breaking down a complex idea. According to experts, o3 makes 20% fewer major errors than previous versions, especially in tough, real-world tasks like programming, consulting, and research.
• o4-mini is a smaller, faster, and more efficient version. It’s designed to handle high-volume tasks without sacrificing much performance. It shines in math, coding, and visual reasoning, and it beats previous small models in tasks across multiple fields, even outside of STEM, like data science or business planning.

Smarter Use of Tools

These new models can do more than just give you answers; they can figure out what tools to use and use them effectively. Let’s say you ask, “How will summer energy use in California compare to last year?” The model can,

• Search online for the latest data,
• Write Python code to forecast usage,
• Create a chart or graph,
• And explain the results all in one go.

And if the first web search doesn’t help, it’ll try a different one. It adjusts and pivots, just like a human would.

Smarter with Images Too

For the first time, ChatGPT can think with images, not just look at them. You can upload a photo of a whiteboard or a sketch, and the model can analyze it, even if the image is blurry or upside down. It can also zoom, rotate, and process the image as part of its reasoning.

This makes it great for visual tasks like solving diagram-based questions or interpreting scientific figures.

Built for the Future

Both models were trained with reinforcement learning, meaning they get better the more they "think." Even when given the same time and resources as earlier models, o3 performs better and continues to improve when given more time to reason. The more challenging the task, the more these models shine.

Whether you're a casual user, a student, a data analyst, or a researcher, these new models bring ChatGPT one step closer to being a reliable digital assistant that can do real work on your behalf.