OpenAI announces o3 and o4-mini reasoning models for ChatGPT (updated)

UPDATE: Apr. 16, 2025, 2:18 p.m. EDT Now that the OpenAI livestream has ended, this article has been updated with the latest information about the o3 and o4-mini launch.

OpenAI's o3 and o4-mini models for ChatGPT have arrived.

As expected, during a livestream event today, the ChatGPT maker released the newest versions of its reasoning models. OpenAI launched o1, its first reasoning or chain-of-thought model, last summer and skipped over o2 to announce its progress on o3 in late December. While o3 has not been publicly available until today, OpenAI already has a lightweight version, o3-mini, on the market.

This Tweet is currently unavailable. It might be loading or has been removed.

In the livestream, the OpenAI team (sans CEO Sam Altman) described some of the models' capabilities and demoed how users can use o3 and o4-mini. The models have been trained to use tools and can process visuals like charts and images.

"For the first time, our reasoning models can agentically use and combine every tool within ChatGPT—this includes searching the web, analyzing uploaded files and other data with Python, reasoning deeply about visual inputs, and even generating images," read the announcement.

The o3 model is now OpenAI's most advanced reasoning model. The lightweight o4-mini is a smaller model designed to be more cost-efficient, and it's particularly good at math and coding, according to OpenAI. Both perform higher than o1 and o3-mini on benchmarks like AIME and GPQA. They have also improved at visual reasoning, based on multimodal evaluations.

Initially, in an effort to streamline its numerous and somewhat confusing model options, Altman said they wouldn't release o3 and o4-mini as standalone models, and would instead integrate their capabilities into ChatGPT-5. But a few weeks later, the company pivoted and decided to release the two models after all, delaying the ChatGPT-5 launch. Again.

Skeptics say the multiple delays of GPT-5 indicate internal struggles to significantly improve model performance. But Altman says the wait will be worth it. In the meantime, o3 and o4-mini have been deemed strong enough models to exist on their own.

o3 and o4-mini are available today via the API and for ChatGPT Pro, Plus, and Teams subscribers, with Enterprise and Edu access coming next week.