On Thursday, OpenAI unveiled an advanced iteration of ChatGPT referred to as “o1.” This model marks the beginning of a series focused on enhancing reasoning capabilities, allowing it to tackle intricate inquiries and perform mathematical, coding, and scientific tasks at a speed surpassing that of human beings.
The “o1” model, also known by its codename “Strawberry,” is being launched alongside a smaller and more cost-effective variant called “o1-mini.” In contrast to existing AI models like GPT-4o, which is currently OpenAI’s most robust option, o1 can manage challenges that were previously insurmountable for these earlier systems.
This new model employs a reasoning process, akin to human thought, before reaching a conclusion. Rather than simply producing an answer in one go—typical of many large language models—o1 engages in a thoughtful exploration of the problem.
The rollout of this technology is part of a broader initiative aimed at developing AI that can logically and methodically address complex issues. As reported by The New York Times, the goal is to create systems that solve problems in a structured manner, progressing through a series of steps that build upon each other, much like human reasoning.
Starting today, users of ChatGPT Plus and Team can access both o1-preview and o1-mini. Users with Enterprise and Edu accounts will gain access next week. OpenAI has also expressed intentions to make o1-mini available to free ChatGPT users, although a specific release date has yet to be confirmed.
Bob McGrew, OpenAI’s chief research officer, shared his thoughts on o1 with The Verge, suggesting, “There are many aspects where it might feel somewhat alien, yet there are equally surprising elements that resonate with human characteristics.”