OpenAI releases 'OpenAI o3-mini', a high-performance AI model that can be used with the free version of ChatGPT



This article, originally posted in Japanese on 11:10 Feb 03, 2025, may contains some machine-translated parts.
If you would like to suggest a corrected translation, please click here.

OpenAI released the inference-specialized model ' OpenAI o3-mini ' on January 31, 2025. OpenAI o3-mini is said to be the most cost-effective model among OpenAI's inference models, and it is appealing that it has higher performance than OpenAI o1 while operating at a lower cost. In addition, it is the first OpenAI inference model to be available to free users of ChatGPT.

OpenAI o3-mini | OpenAI

https://openai.com/index/openai-o3-mini/

Below is a graph comparing the mathematical performance of 'OpenAI o1-preview', 'OpenAI o1', 'OpenAI o1-mini', and 'OpenAI o3-mini'. The performance of OpenAI o3-mini is divided into three types: 'low', 'medium', and 'high' for each effort spent on inference. From the graph, it can be seen that OpenAI o3-mini outperforms OpenAI o1-mini with medium effort, and shows higher performance than OpenAI o1 by increasing the effort spent on inference.



In addition, the OpenAI o3-mini also outperformed the OpenAI o1 in tests solving doctoral-level scientific problems.



Below is a graph comparing the response speed of OpenAI o3-mini (bottom) and OpenAI o1-mini (top), which put a moderate amount of effort into inference. The OpenAI o1-mini took 10.16 seconds to return the first token, while the OpenAI o3-mini was able to return the first token in 7.7 seconds.



OpenAI o3-mini is also available for the free version of ChatGPT. When you actually access the free version of ChatGPT, a button called 'Reason' has been added to the chat input field. If you click this button and then ask a question, an answer will be generated by OpenAI o3-mini.



For example, type in 'Tell me how to make the most of cold ice cream in the winter,' then click 'Why' and click the submit button.



After a few seconds of thinking, the app generated an answer, with useful suggestions such as 'pair it with a warm dessert' and 'enjoy the temperature difference in a warm room.'



The free version of ChatGPT allows you to use the OpenAI o3-mini, which uses a medium amount of effort for inference. ChatGPT Plus users can also use the OpenAI o3-mini, which uses a high amount of effort for inference.

OpenAI o3-mini also provides an API, with input available at $1.1 (about 171 yen) per million tokens and output available at $4.4 (about 686 yen) per million tokens. OpenAI claims that it can be used at a cost 93% lower than OpenAI o1.




Related Posts:

in Review,   Software,   Web Application, Posted by log1o_hf