OpenAI has unveiled a new model of artificial intelligence called o1, focusing on answering questions in the fields of science, mathematics, and programming. The unique aspect of o1 compared to previous models is its ability to “think” before providing an answer, allowing it to respond to complex commands such as creating games based on specifications.
The model comes in three versions: o1, o1-preview, and o1-mini. In testing against the mathematics competition AIME, o1 achieved a score of 74.4%, while o1-mini scored 70%. In comparison to real competitors, these scores place them in the top 500 high school students in the United States. In programming competition testing on Codeforce, o1 achieved an Elo rating of 1673, while o1-mini scored 1650, placing them in the 86th percentile among all competitors.
Due to specific training, when tested on general knowledge questions like MMLU, the score of o1-mini may be lower than GPT-4o, but other models in the o1 series perform better.
In terms of speed, if GPT-4o takes 3 seconds to answer a question, o1-mini takes 9 seconds, and o1-preview takes up to 32 seconds.
Users of the paid version of ChatGPT can now choose to utilize o1-mini and o1-preview, but with a limit on the number of messages per week. API access is restricted to Tier 5 customers (those who have spent over $1,000) and limited to 20RPM. However, there are plans to make o1-mini available to all users for free in the future.
TLDR: OpenAI introduces a new AI model, o1, with a focus on answering questions in science, mathematics, and programming. It offers different versions, with varying capabilities and speed, and is currently available to Tier 5 paying customers with plans for free access to all users in the future.
Leave a Comment