This is how OpenAI’s new ‘reasoning’ AI model will answer more complex questions

With the rising popularity of generative AI (GenAI), OpenAI, led by Sam Altman, has introduced a new ‘reasoning’ AI model designed to tackle complex questions more swiftly than humans.

The company describes its ‘OpenAI o1 model’ as being trained to take more time to deliberate on problems before providing answers, emulating human thought processes. This training helps the model improve its problem-solving approach, test various strategies, and learn from errors.

The new model is applicable in various fields: healthcare researchers can use it for annotating cell sequencing data, physicists for deriving intricate mathematical formulas for quantum optics, and developers across disciplines for creating and managing multi-step workflows.

“We’ve crafted a new series of AI models that spend more time on reasoning before answering. These models can tackle complex tasks and solve more difficult problems in science, coding, and mathematics compared to previous models,” the company noted.

Tests reveal that the model performs at a level comparable to PhD students on challenging tasks in physics, chemistry, and biology.

In mathematics, the new model significantly outperformed previous versions: while GPT-4o solved only 13% of International Mathematics Olympiad problems, the reasoning model achieved an 83% success rate. It also ranked in the 89th percentile in Codeforces coding contests.

Although still in its early stages, this model lacks some of the features found in ChatGPT, such as web browsing and file uploads. Nonetheless, it represents a major leap forward in AI reasoning capabilities.

“To mark this advancement, we are starting fresh with the series name OpenAI o1,” the company stated.

Additionally, OpenAI has introduced a more affordable variant, the ‘OpenAI o1-mini,’ which is optimized for faster reasoning, particularly in coding tasks. This smaller model is 80% cheaper than o1-preview, making it a cost-effective option for applications requiring reasoning without extensive general knowledge.