ChatGPT gets o3 and o4-mini: AI that picks its own tools

What's new

On April 16, 2025, OpenAI launched two new reasoning models, meaning AI models that think through multiple steps before answering. They are called o3 and o4-mini. OpenAI calls them the most intelligent and capable models the company has released to date.

o3 is the new flagship model in the o-series, while o4-mini is a smaller and cheaper version with higher usage limits. Both have access to all tools in ChatGPT. It is the first time OpenAI's reasoning models can use and combine them on their own: web search, analysis of uploaded files with Python, visual reasoning and image generation.

What's smart about it

The models are trained to decide for themselves when and how to use each tool. So you don't have to tell them to search the web or open your spreadsheet; they choose on their own. According to OpenAI, they typically respond in under a minute.

They can also think with images, so a photo or a sketch becomes part of the reasoning itself and not just something described at the end.

OpenAI states that o3 makes 20 percent fewer major errors than its predecessor o1 on complex real-world tasks.

Cheaper or better?

On the SWE-bench Verified coding test, o3 scored 69.1 percent and o4-mini scored 68.1 percent. For comparison, o3-mini scored 49.3 percent and Anthropic's Claude 3.7 Sonnet scored 62.3 percent. The figures come from OpenAI's own measurements and have not been independently verified. On the AIME 2025 math exam, o4-mini scored 99.5 percent, according to OpenAI.

For developers, o3 costs $10 per million input tokens and $40 per million output tokens at launch. Tokens are the small chunks of text the AI works with. o4-mini costs $1.10 per million input tokens and $4.40 per million output tokens. The prices are from the time of launch, so current ones should be checked with OpenAI.

For regular users, the models are part of the subscriptions. A monthly price for ChatGPT is not given in the sources.

What it's good for

OpenAI highlights o3 as strong at programming, math, science and image understanding. According to OpenAI, o4-mini is better than its predecessor o3-mini at tasks outside the sciences and at data science, and it is a good choice when you want to run many queries at a lower price.

The models are available from launch day to ChatGPT Plus, Pro and Team users and replace o1, o3-mini and o3-mini-high in the model picker. Free users can try o4-mini by selecting "Think" in the input field. Enterprise and Edu customers get access a week later. Developers can use the models through the Chat Completions API and the Responses API.

OpenAI has also announced o3-pro for Pro subscribers "in a few weeks," with full tool support. The sources do not say whether the models are available in Denmark and the EU on the same terms. The launch is part of the race over reasoning models against Google, Meta, xAI, Anthropic and DeepSeek, among others.