OpenAI has unveiled OpenAI o1, the first artificial intelligence (AI) model in its new series equipped with ‘reasoning’ capabilities. Nicknamed the ‘Strawberry’ model, its creators said that the new platform is not only designed to handle more complex queries but has already demonstrated significant improvements in reasoning over its predecessors. This includes solving 83% of the problems contained in the International Mathematics Olympiad qualifier, compared to the 13% of queries solved by o1’s antecedents.
“We trained these models to spend more time thinking through problems before they respond, much like a person would,” said OpenAI, claiming that o1 also performed similarly to PhD students in solving complex tasks in physics, biology and chemistry exams. “As an early model, it doesn’t yet have many of the features that make ChatGPT useful, like browsing the web for information and uploading files and images… but for complex reasoning tasks [o1] is a significant advancement and represents a new level of AI capability.”
Safety first for o1
OpenAI also noted the advancements in safety and security with this new series, stating that o1 has outperformed its predecessors in adhering to safety guidelines. The firm also claimed that the series has been specifically trained to better handle attempts to bypass safeguards, known as jailbreaking, a key challenge in AI safety.
The company claimed that in tests, the o1-preview model scored 84 on jailbreaking assessments, compared to GPT-4o’s score of 22.
OpenAI’s focus on safety and security comes amidst the global broader push to align AI models with safety protocols, supported by the memorandum of understanding signed between the UK and the US to establish greater AI safety.
Last month, OpenAI and Anthropic signed agreements to collaborate on AI safety research, testing, and evaluation with the US Artificial Intelligence Safety Institute, part of the US Department of Commerce’s National Institute of Standards and Technology (NIST).
Currently, an early preview of the new reasoning models is made available in ChatGPT and the API, said the AI company. In addition to model updates, OpenAI expects to add browsing, file and image uploading, and other features.
In addition to its reasoning capabilities, OpenAI will release a smaller, more cost-effective model in the series, o1-mini, specifically designed to excel at coding while being 80% cheaper than the larger model.
Recently, OpenAI announced it surpassed one million paid business users for its enterprise services, including ChatGPT Enterprise, ChatGPT Team, and ChatGPT Edu.
Following the launch of the ChatGPT Team in January 2024, OpenAI’s business customer base is said to have grown from 600,000 in April to one million this month, representing a 67% increase in just under five months.
Read more: Microsoft abandons observer seat on OpenAI board
