A few days ago, we learned that OpenAI would be about to unveil Strawberry, an AI that handles user queries in a very different way: it reasons for a few seconds to answer more complex questions. And a few hours ago OpenAI o1 has been announceda new AI model that constitutes a significant advance and represents a new level of generative artificial intelligence capability. OpenAI has given details of the two versions that can start being used now in Europe.
In the news released two days ago, it was specified that this new model is designed for spend more time “thinking” before giving the user an answer. Very different from what is happening with one of the latest models presented by OpenAI with GPT-4o, which in fact, thanks to its great capacity for almost immediate response, its conversational model puts Google itself on the ropes with Gemini Live.
This is exactly what OpenAI has just explained with its new AI model: capable of reasoning more complex tasks and solve more complicated problems than previous models for some science, programming and mathematics tasks.
OpenAI o1 Overview
Free Android
As of today, the first version of these series is deployed in ChatGPT and its API. This is a preview and periodic updates and improvements are expected. Also performance evaluations will be included for the next update currently in development.
OpenAI o1 was trained to spend more time “thinking” or reasoning about problems before responding, just like a person would. Through this training, the AI’s language model learns to optimize its reasoning processes, tries different strategies, and recognizes its mistakes.
AI that “thinks”
In the tests carried out, the next update of the model achieves similar results to students pursuing a doctorate (PhD) to tackle performance testing tasks in physics, chemistry and biology. This model is also capable of delivering very good results in mathematics and programming.
There is a fact that makes the importance of OpenAI o1 very clear, if GPT-4o managed to correctly solve 13% of the IMO (International Mathematics Olympiad) problems, This reasoning model solves 83%
OpenAI o1 Overview
Free Android
Being in an early phase of development, this new model does not have all the multimodal capabilities that GPT-4o offers. That is to say that For now, it’s just a text message and is unable to crawl the web for information or download images and files for analysis.
OpenAI o1-mini
If OpenAI o1 is entirely dedicated to very complex tasks in science, programming, mathematics and other similar fields —how could he help the health research to record cell sequencing data or help physicists generate complex mathematical formulas for quantum optics— OpenAI o1-mini was also announced.
Just in July, the GPT4-o mini was announced as a model with significant capabilities, but not as resource intensive, OpenAI o1-mini goes in the same direction to offer an effective solution for developers. A faster, cheaper model and very efficient in programming.
The selector of the two new models in ChatGPT
Free Android
A smaller model than It’s 80% cheaper than o1-previewwhich makes it a very powerful way to reduce costs for certain types of applications that don’t require as much reasoning.
Both models, o1-preview and o1-mini they are already available today on ChatGPT in Europe and can be manually selected from the template selector, although there are some limitations on weekly requests: 30 messages for o1-preview and 50 for o1-mini.
This index will increase soon with the capacity of enable ChatGPT automatically to select model the most appropriate according to the query. Finally, OpenAI’s goal is to soon launch the 01-mini model to all users who use ChatGPT for free, according to its announcement on the web.
