OpenAI presents o1, its new AI with better reasoning in mathematics and programming

“We have developed a new series of AI models designed to spend more time thinking before responding. They can reason about complex tasks and solve more difficult problems than previous models in science, programming and mathematics,” he announced. OpenAIthe company behind GPT-4o and ChatGPT, in a statement.

Unlike conventional models, which use a statistical model to calculate what “sounds” right to a request, GPT o1 was trained with another system, trial, error and positive feedback, to generate its own reasoning path; this makes it more efficient for some things, and worse than GPT for others.

According to the company, the OpenAI o1 model “performs similarly to PhD students on challenging benchmark tasks in physics, chemistry, and biology. We also found that excels in math and coding. In a qualifying exam for the International Mathematical Olympiad (IMO), GPT-4o solved only 13% of problems correctly, while the reasoning model scored 83%. Their coding skills were tested in contests and they reached the 89th percentile in Codeforces competitions.”.

At the same time, they warn, does not have the same functions as GPT-4o: It cannot search for information on the web, nor generate images. But it is less mind-blowing (although it still invents content from time to time), and unlike GPT-4o it does not try to respond to the user’s request immediately, but rather takes more time to analyze variables and understand whether it is responding adequately to the request.

“These enhanced reasoning capabilities can be particularly useful when tackling complex problems in science, coding, mathematics, and similar fields, the company says. For example, o1 can be used by healthcare researchers to annotate cell sequencing data, by physicists to generate complicated mathematical formulas needed for quantum optics, and by developers across all fields to create and run multi-step workflows..”

In the words of Sam Altman, the company’s CEO, “o1 is still flawed, still limited, and still seems more impressive on first use than after spending more time with it. But it is also the beginning of a new paradigm: an AI that can perform complex, purpose-driven reasoning.”l.”

The progress of this language model is significant enough – and experimental – that the company has decided to “reset” the counter: this new model is called OpenAI o1, and it will have two versions: the full version, and OpenAI o1-mini, which is cheaper and more limited than the full version.

How to use it

• ChatGPT Plus and Team users will be able to access o1 templates in ChatGPT. Both o1-preview and o1-mini can be hand-picked in the template picker.
• ChatGPT Enterprise and Edu users will have access to both models starting next week.
• According to the company, it will eventually be rolled out to users of the free version of ChatGPT.

At the developer level, the cost of using the OpenAI o1 API is higher than that of GPT-4o. As reported by The Verge, OpenAI o1 costs $15 per million tokens (bits of meaning in the order that the user writes down) analyzed, and $60 per million tokens generated. In comparison, GPT-4o costs $5 per million tokens analyzed and $15 per million tokens generated.

The announcement comes on the heels of the news that The company is seeking a new round of financing for $6.5 billion which could give it a stock market value of $150 billion.