News Post

OpenAI releases o1, its first model with ‘reasoning’ abilities

The Verge · 6d

OpenAI has unveiled o1, its latest AI model touted as the first in a series of "reasoning" models. This highly anticipated release, previously rumored as "Strawberry," is designed to handle complex queries more efficiently than its predecessors. O1 employs reinforcement learning and a "chain of thought" process, mimicking human problem-solving steps. The model excels in coding and mathematics, even outperforming GPT-4o on challenging tasks like the International Mathematics Olympiad qualifying exam. However, o1 is slower and significantly more expensive to use than GPT-4o, with API costs up to three times higher. OpenAI is releasing o1 alongside a smaller version, o1-mini. Access is initially limited to ChatGPT Plus and Team users, with plans to expand availability. While o1 shows improved accuracy and reduced hallucinations, it lacks some capabilities of GPT-4o, such as web browsing and image processing. This release marks a step towards OpenAI's goal of developing more human-like artificial intelligence, potentially leading to autonomous systems capable of complex decision-making.