Here’s a rephrased version of the text while keeping the HTML tags and structure intact:

—

Update: April 16, 2025, 2:18 p.m. EDT — This article has been refreshed with the most recent information following OpenAI’s livestream presentation of the o3 and o4-mini models.

OpenAI has officially announced the release of its newest reasoning models for ChatGPT: o3 and o4-mini.

In a livestream event earlier today, the AI firm unveiled the new models, highlighting their functions and how users can engage with them. This signifies the public introduction of o3, OpenAI’s most sophisticated reasoning model to date, along with o4-mini, a compact, more efficient variant tailored for tasks such as mathematics and programming.

OpenAI initially presented its series of reasoning models last summer with o1, bypassing o2 and later hinting at o3’s progress in December. While o3 had remained under wraps until this moment, a lighter version—o3-mini—had already been discreetly launched.

During the livestream (notably without CEO Sam Altman), OpenAI illustrated how the new models can autonomously employ tools and interpret visual data like graphs and pictures. According to the company’s formal statement, “For the first time, our reasoning models can independently use and integrate every tool available within ChatGPT—this includes searching the internet, analyzing uploaded documents and various data using Python, deeply reasoning about visual information, and even creating images.”

Both o3 and o4-mini outperform their earlier versions—including o1 and o3-mini—on significant benchmarks like AIME and GPQA. They also exhibit significant advancements in visual reasoning, owing to improved multimodal capabilities.

Initially, OpenAI intended to incorporate these models directly into the forthcoming ChatGPT-5, circumventing standalone launches to streamline its expanding portfolio. However, the firm later changed its approach, choosing to issue o3 and o4-mini separately—once again postponing ChatGPT-5.

This decision has sparked speculation regarding internal difficulties at OpenAI, particularly concerning achieving substantial performance enhancements with GPT-5. Despite the setbacks, CEO Sam Altman insists that the eventual result will justify the delay. Meanwhile, o3 and o4-mini are viewed as robust enough to function independently.

Effective today, both models are accessible via OpenAI’s API and to subscribers of ChatGPT Pro, Plus, and Teams. Availability for Enterprise and Edu users is anticipated to commence next week.