OpenAI, the developer of ChatGPT, announced on the 31st (local time) that it will launch a small artificial intelligence (AI) model called 'o3 mini' that is capable of advanced reasoning.
'o3 mini' is a small model version of 'o3', an upgraded version of the reasoning model 'o1' (Owon) released by OpenAI in September last year.
According to OpenAI, the 'o3' model focuses on reasoning abilities and has been trained to think before responding, just like 'o1'. OpenAI explained that it can deduce and plan tasks and help find solutions for tasks that span long periods.
Also, similar to 'o1', responses may take a few seconds to a few minutes longer, but OpenAI emphasized that it is more reliable in fields such as physics, science, and mathematics.
'o3' and 'o3 mini' were first unveiled last December. Sam Altman, the CEO, noted at the time, 'We plan to release 'o3 mini' first by the end of January next year, followed closely by 'o3'.
The industry is paying attention as 'o3 mini' is launched at a time when the reasoning model 'R1′ from the Chinese AI startup DeepSeek has gained significant popularity. The DeepSeek model pushed ChatGPT to second place and took first place in the free app download rankings on the U.S. Apple App Store on the 27th.
DeepSeek has stated that the 'R1′ model demonstrated comparable or better performance than leading U.S. models in various tests.
The Institute of Physics at the Chinese Academy of Sciences previously claimed through its official social media account on the 30th that the 'R1' model scored 100 points out of a possible 140 points on a problem posed during the 'Tianmu Cup Theoretical Physics Contest' held in Jiangsu Province on the 17th.
'o3 mini' can be used for free through ChatGPT. Paid subscribers such as ChatGPT Plus, Pro, and Teams can also access it, and it will be distributed to corporate subscribers within a week, according to OpenAI.