OpenAI launches o1-preview, a new model for solving complex problems

OpenAI has launched o1-preview, the first in a new series of AI models designed to solve complex problems through deeper reasoning. Unlike its predecessors, these models take more time to think before responding, excelling in scientific, mathematical, and programming tasks. In internal tests, o1-preview outperformed previous models, such as GPT-4, solving 83% of International Mathematics Olympiad problems compared to GPT-4’s 13%.

The model has also excelled in programming competitions, ranking in the 89th percentile in Codeforces contests. In addition, OpenAI has released a more affordable and efficient version, o1-mini, tailored for tasks requiring efficient reasoning without broad general knowledge. o1-mini offers the same coding generation and debugging capabilities at 80% less cost.

The o1-preview model is available for ChatGPT Plus and Enterprise users, as well as developers through OpenAI’s API. While the API currently lacks some advanced features like streaming and system messages, it allows developers to begin integrating this powerful model into their applications, with initial limits of 20 requests per minute (RPM) that will increase over time. This model marks a major step in AI development for tackling more difficult tasks, making it a valuable tool for scientists, programmers, and mathematicians.

Along with its improved reasoning capabilities, OpenAI has implemented new safety measures. These features help the model strictly adhere to safety guidelines, reducing the risk of “jailbreaking” or misuse. In a safety test, o1-preview scored 84 out of 100, compared to GPT-4’s 22.

This initial release is just the beginning, as OpenAI plans to continue enhancing o1-preview with regular updates and add features like web browsing and file uploads to make it even more useful in the future.