New AI models from OpenAI can pause and think before answering, allowing image processing and direct Python code execution in the browser.
OpenAI has just officially launched two new artificial intelligence models o3 and o4-mini on Thursday. These are reasoning models designed to "pause and think" before providing an answer, marking an important step in the increasingly fierce global AI race.
According to OpenAI, o3 is their most advanced reasoning model to date, outperforming previous models in tests of mathematics, programming, reasoning, science, and image understanding. The o4-mini model is introduced as a balanced choice between cost, speed, and performance – three key factors that developers often consider when selecting an AI model.
Technological Breakthrough in a Highly Competitive Context
Notably, these models can generate responses by using tools in ChatGPT such as web browsing, Python code execution, image processing, and creation. Specifically, o3 and o4-mini are the first OpenAI models capable of "thinking with images" – users can upload images like whiteboard sketches or diagrams from PDF documents, and the models will analyze the images during the "chain of thought" before responding.
In terms of performance, o3 achieved an impressive result in the SWE-bench verified test (without custom scaffolding) with a score of 69.1%, significantly outperforming o3-mini (49.3%) and Claude 3.7 Sonnet (62.3%).
OpenAI's release of o3 is a notable turning point, especially since CEO Sam Altman had signaled in February that the company intended to focus resources on a more sophisticated solution integrating o3's technology. However, competitive pressure seems to have prompted OpenAI to change direction.
All three models – o3, o4-mini, and o4-mini-high (a variant spending more time improving reliability) – are now available for OpenAI's Pro, Plus, and Team subscribers, and will also be provided through developer endpoints, including Chat Completions API and Responses API.
In terms of pricing, OpenAI charges relatively low fees for o3 at $10 per million input tokens and $40 per million output tokens. For o4-mini, the price is kept at a level similar to o3-mini, $1.10 per million input tokens and $4.40 per million output tokens.
In the coming weeks, OpenAI plans to release o3-pro, a version of o3 using more computational resources, dedicated to ChatGPT Pro subscribers. Sam Altman has also indicated that o3 and o4-mini may be the final independent AI reasoning models in ChatGPT before GPT-5 launches – a model expected to unify traditional models like GPT-4.1 with reasoning models.
