OpenAI surprise launches o1, a new artificial intelligence model that “thinks” before responding

Innovations in the field of artificial intelligence (AI) are constant and increasingly advanced. OpenAI has just presented a new model called “o1”, colloquially known as “Strawberry”. It arrives with the promise of tackling more complex problems than its predecessors, showing a capacity for higher reasoning.

The advantage over the previous model is that it was the first in a new series trained to simulate a thought process. human-like which constitutes a significant milestone in the evolution of major language models.

“We train these models to spend more time thinking about problems before responding, just like any other person would. Through training, they learn to hone their thinking process, try out different strategies, and acknowledge your mistakes“, the company says.

To achieve this condition it was subjected to a new form of training, based on reinforcement learning. This means that GPT o1 has been reward-based training when the objectives have been achieved and penalties when they have not been achieved.

The truth is that this model, although it introduces a fairly important leap in the ambitions of the American company to recreate increasingly sophisticated products, does not have the characteristics of immediate search that make ChatGPT useful.

The performance of this powerful update, as explained by OpenAI, is similar to the reasoning achieved by PhD students in physics, chemistry and biology tasks.

In a qualifying exam for the International Mathematical Olympiad (IMO), GPT-4o only solved 13% of the problems correctly, while the reasoning model obtained a 83%Their coding skills were evaluated in contests and they reached the 89th percentile in Codeforces competitions.

Which means their enhanced reasoning abilities can be especially beneficial when a researcher is faced with complex problems in science, coding, mathematics, and similar fields.

Thus, geneticists can use o1 to annotate cell sequencing data, physicists to generate complicated mathematical formulas needed for quantum optics, and developers across all fields to create and run multi-step workflows.

The company’s new AI model is now available on ChatGPT. But this launch has several peculiarities. First, we are dealing with a preliminary version, so its development is still very active. Second, it will be a limited paid option.

From the drop-down menu in ChatGPT, you can choose the model you prefer. Also listed in that box is the OpenAI o1-mini, a faster model primarily intended for programming tasks. In all cases, it should be noted that o1 will not be able to browse the web and analyze images to provide answers.

Another important fact is that o1 and o1-mini will also be available in the OpenAI API, allowing developers to power applications with these models.

The prices to access these models are quite different. While 01 has a value of 60 dollars For 1 million output tokens, 01-mini is priced at 12 dollars for 1 million output tokens.