Who will be able to stop OpenAI? After launching GPT-4.1, mini and nano a few days ago, the startup is back with two additional models called o3 and o4-mini! Let's put aside the now well-established fact that OpenAI definitely doesn't know how to name its inventions, and focus on the new features of these models.
o3 and o4-mini, well-versed in math, science and programming
o3 and o4-mini are reasoning models, that's say they take longer to respond, but in return their answers are supposed to be more accurate. o3 is the most advanced model with "strong performance" in programming, math, and science. 04-mini is presented as a less expensive alternative that still offers "impressive results" in these areas.
Both models are also capable of searching for information on the internet and analyzing images. They are therefore able to solve complex problems more efficiently—such as finding the exit to a maze, giving the date of an event from a photo of the event in question, or even finding the solution to a puzzle drawn on a blackboard. They can also resize a photo to help their reasoning.
ChatPT Plus, Pro, and Team subscribers can start testing o3 and o4-mini today. An even more powerful o3-pro version is in the pipeline for Pro subscribers. It will be available soon (they can continue to use o1-pro in the meantime).
In the process, OpenAI also unveiled Codex CLI, a command-line interface that allows these new models to write code and modify it locally; this agent can also move files. A tool to install in your computer's Terminal, and which is intended for developers of course.
Source: OpenAI
0 Comments