OpenAI introduces O3 and O4-mini: the smartest ChatGPT model debuts, and "image inference" is a new breakthrough

OpenAI announced two new ChatGPT models: o3 and o4-mini, which OpenAI says are not only the smartest and most capable models available, but also include unprecedented capabilities: the ability to really "think" images, not just look at pictures, but also incorporate images into the inference process.

Both models are available to ChatGPT Plus, Pro and Team plan users, and will be available to ChatGPT Enterprise and Edu users within a week. At the same time, developers can now use the two new models through the Chat Completions API and the Responses API.

O3: All-round player, reasoning and visual understanding are fully upgraded

According to OpenAI, the o3 model excels in several areas, especially setting new standards in tasks such as programming, mathematical operations, scientific reasoning, and visual understanding. This model also supports the integration of several built-in tools, including:

Web search

Python arithmetic tools

Image analysis and generation

File parsing function

This means that users can now not only ask questions, but also allow AI to actually help analyze data, look at pictures and interpret, and even automatically generate images or code.

O4-MINI: High-efficiency inference machine, suitable for a large number of query needs

Compared to O3, O4-mini focuses on cost and speed optimization. Although the model is small, its inference capabilities are still excellent, especially suitable for applications that require high frequency and high throughput.

OpenAI says that if you need to ask hundreds of questions a day, o4-mini is definitely the best choice for you. And because of the excellent performance and low computing cost, users can enjoy higher usage quotas.

New ability to think about images: from looking at pictures to "thinking with pictures"

One of the biggest highlights of both models is the ability to "think with images" for the first time.

While traditional AI models can only see pictures, o3 and o4-mini can now incorporate images into the entire inference process. This means that if you pass an image to the model, it will not just tell you what is in the picture, but will treat that image as a reasoning clue, analyze and integrate with the text data, and answer in more depth.

This breakthrough is a boon for fields such as education, healthcare, design and engineering.

Ask GPT to reason about what to write in an inverted note in a picture You can ask ChatGPT to identify the port vessel in the photo and deduce relevant information The new version of the model is online, and ChatGPT users can experience it immediately

From now on, users of ChatGPT Plus, Pro, and Team will see:

O3

o4-mini

o4-mini-high

They will replace the old O1, O3-Mini and O3-Mini-High. Although Pro users can currently choose the o1-pro model, OpenAI said that the o3-pro will be launched in the next few weeks, when full tool support will be available.

Developer tools are upgraded synchronously, and the Responses API enhances inference capabilities

In addition to ChatGPT users, developers can now use o3 and o4-mini through the Chat Completions API and the newly upgraded Responses API.

The Responses API adds several new features, including:

Reasoning summaries

Support for persisting tokens for inference processes (improved performance)

Built-in tools such as web search, file search and code interpreter will be supported soon

This allows developers to build smarter, more reasoning applications.

With the introduction of o3 and o4-mini, OpenAI has once again raised the ceiling of AI models. Not only the reasoning ability has been comprehensively upgraded, but also the ability to truly understand "looking at pictures and thinking" has been added. Whether you're a developer, a business user, or a curious everyday user, both models offer more options and possibilities.

This article OpenAI launches o3 and o4-mini: the smartest ChatGPT model debuts, and "image inference" becomes a new breakthrough first appeared in Chain News ABMedia.