The Sam Altman-run OpenAI has unveiled a new “reasoning” model that has been taught to answer more complicated questions quicker than a person can, as generative AI (GenAI) gains popularity across the board.

The AI startup claims that it educated the “OpenAI o1 model” to think over issues more thoroughly before responding, much as a human would. They learn to hone their thought processes, experiment with various tactics, and own up to their errors via training.

The new AI model may be used by physicists to construct complex mathematical formulae required for quantum optics, by healthcare researchers to annotate cell sequencing data, and by developers across all domains to create and implement multi-step processes.

We’ve created a new line of AI models that are meant to deliberate longer before reacting. In science, computing, and arithmetic, they are able to think through intricate tasks and solve more challenging challenges than earlier versions, the business said.

In experiments, the model achieves comparable results to PhD students on hard benchmark problems in biology, chemistry, and physics.

We also discovered that it does well in coding and math. GPT-4o accurately answered just 13% of the questions in a qualifying test for the International Mathematics Olympiad (IMO), but the reasoning model earned 83%, according to OpenAI.

The 89th percentile was attained in Codeforces competitions, where the coding skills were assessed via tournaments.

Many of the capabilities that make ChatGPT valuable are still missing from this early iteration, such as the ability to post files and photographs and browse the web for information.

However, this is a major breakthrough and a new level of AI power for complicated thinking tasks.

The business said, “Given this, we are resetting the counter back to 1 and naming this series OpenAI o1.”

Additionally, it has created the OpenAI o1-mini, a quicker reasoning model that excels at coding at a lower cost within the “reasoning” series. O1-mini is a strong, economical model for applications that demand reasoning but do not require extensive world knowledge since it is a smaller model and costs 80% less than O1-preview.