OpenAI has unveiled its next-generation reasoning models, o3 and o4-mini, marking a major leap in AI capabilities with the introduction of built-in agentic abilities. These new models are designed to independently use tools such as web browsing, file analysis, image generation, and visual reasoning—bringing the vision of an autonomous AI assistant closer to reality.
In a blog post, OpenAI emphasized that o3 and o4-mini are the company’s first agentic reasoning models, capable of understanding complex, multi-step instructions and deciding when and how to use integrated tools in ChatGPT. This evolution represents a critical milestone toward creating a fully agentic version of ChatGPT—one that can perform tasks on a user’s behalf with minimal input.
Meet the New Models
o3: OpenAI’s Most Powerful Reasoning Model Yet
OpenAI describes the o3 model as its most advanced reasoning engine to date, demonstrating exceptional performance across a wide range of domains, including:
-
Coding
-
Mathematics
-
Science
-
Visual understanding
The model is particularly adept at solving multi-layered problems where answers aren’t immediately obvious. Its visual reasoning skills stand out, enabling it to interpret and analyze images, charts, and diagrams with impressive accuracy.
o4-mini: A Smaller Model with Big Potential
The o4-mini is a compact yet capable model designed for speed and cost-efficiency. Despite its smaller size, it delivers high-quality results in coding, math, and visual reasoning tasks. It outperforms the previous o3-mini model in several areas such as data science and offers higher usage limits, making it ideal for frequent users.
OpenAI also released a high-capacity version called o4-mini-high, available to select users for more intensive tasks.
Visual Reasoning: “Thinking with Images”
One of the standout features of these new models is their visual chain-of-thought capability. As OpenAI puts it, the models “don’t just see an image—they think with it.” Users can now upload visuals such as:
-
Textbook diagrams
-
Whiteboard photos
-
Hand-drawn sketches
Even if the images are blurred, reversed, or low-resolution, the models can interpret them and perform reasoning tasks. With tool use enabled, users can also transform images in real-time, including rotating, zooming, or modifying them as part of a larger analysis.
Availability and Access
The new models—o3, o4-mini, and o4-mini-high—are currently available through the ChatGPT model selector for:
-
Plus
-
Pro
-
Team subscribers
Free-tier users can also try the o4-mini model by selecting the “Think” option in the ChatGPT composer. Additionally, OpenAI announced plans to introduce an even more capable model, o3-pro, in the near future.
