OpenAI has surprised the artificial intelligence world with the launch of o1-preview, a model with advanced reasoning capabilities. This model, internally known as "Strawberry," was announced by the company led by Sam Altman. Unlike previous models such as GPT-3.5 or GPT-4, which respond almost immediately, o1 will take time to "think" before offering a response, aiming to improve accuracy and quality in solving complex problems.
A model designed for complex tasks
The main feature of o1 is its focus on deep reasoning, which sets it apart from previous models. According to OpenAI, o1 will "think" similarly to how a person would, refining its responses and recognizing its errors over time. This makes it an ideal tool for solving scientific problems, surpassing GPT-4's abilities in areas such as mathematics, physics, chemistry, and biology.
One example of its superior performance was seen in the International Mathematical Olympiad (IMO), where o1 solved 83% of the problems, compared to 13% achieved by its predecessor, GPT-4o. This reasoning ability promises to be useful for researchers and developers in fields requiring precision, although the model will not be suitable for tasks that need quick responses, like customer service chatbots.
Access and availability of o1
The new model is now available on ChatGPT for ChatGPT Plus and Team users, while ChatGPT Enterprise and Edu users will gain access next week. Additionally, o1 will also be available through the OpenAI API, allowing developers to use it in applications.
However, o1 will not be able to browse the web or analyze images to generate responses, unlike other models. The model will cost $60 per million output tokens, while a faster and more limited version, called o1-mini, will be available for $12 per million tokens, primarily intended for programming tasks.
